TO

BURT ORE BEN

firm friend and

constructive critic
down the decades

Copyright C 1992, 1990 by the President and Fellows of Harvard College

All rights reserv ed
Printed in the United States of America
10 9 8 7 6 5 4 3 2 1

First Harvard University Press

paperback edition, 1992

Library of Congrtss Cataloging-in-PwbliUltion Data

Quine, w. V. (Willard Van Orman)

Pursuit of truth I W. V. Quine. - Rev. ed.
p. cm.
Includes bibliographical references and index.
ISBN 0-674-73951-5 (pbk. : alk. paper)
I. Meaning (Philosophy) 2. Reference (Philosophy) 3. Knowledge,
Theory of. 4. Semantics (Philosophy) I. Title.
B94 5·Q5 3P87 1992
UI-dc20

92-5606
CIP
 Preface to the Revised Edition

In May 1990, a mere four months after first

this book
appeared, I was in th e g allan t little Republic of San Marino
for a week-long international colloquium on my philoso­
phy. Six months later I was in medieval Girona, in Cata­
lonia, giving the Josep Ferrater Mora Lectures-fifteen
hours of them and five of discussion. Donald Davidson,
Bur to n Dreben, Dagfinn F0Uesdal, and Roger Gibson
were all im por ted wit h me, to add depth and zest to the
discussion. The busy months of p re pa ration and the stimu­
lating exch an ges on these occasions sparked thoughts that
would have made for a better book if the chronology had
b een inverted. I am approximating such an invers io n as
best I can by this ear ly re vis ed edit io n .

Old §13 , "Ontological r ela ti vit y, has become more

"

emphatically "Ontology defused," and incorporates bits

from my responses in the projected San Marino volume.
My treatment of domestic meaning in §22 is utterly
changed, and so also, thanks to Davidson's and F0Uesdal's
a b et ting , are §§28-29 on propositional attitudes.

March 1992 W.V.Q.

 Preface to the First Edition

In these pages I have undertaken to upda te, sum up, and

c l a ri fy my variously intersecting views on cognitive mean­
ing, objective reference, and the grounds of kn owledge.
S ome of the progress is expository and some substantive.
The substance has been precipitating sporadically over the
past ten years, and some of it has surfaced in lectures, infor­
mal d iscuss ions, and scattered paragraphs. In interrelating
these thoughts I have occasionally found a faulty joint and
ha ve firmed it up to my satisfaction.
I intend this little book no less for my past readers than for
my new ones, so I have curbed my exposition of things
already belabored in my other books. I do retrace fa miliar
ground where 1 s ee an impro ve ment in the idea or its pre­
sentation, and also where the new reader needs a little
briefing to be kept abreast.
The bits of the book that have pre viousl y appeared in
pr in t add up to a scant nine pages, and are ident ified on a
back page. Unpublished l ect u res were a richer source. My
lecture "The Mentalistic Heritage" in Calcutta, 1983, is a
source of §3 I, and "The Forked A nim al" y ielded earlier
parts of Chapter IV. That lecture was the third of four
Immanuel Kant Lectures that I gave at Stanford in 1980.
The tide of the series offour was "Science and Sens ibilia , a "

take off of John Austin 's takeoff of J ane Austen. The four
lectures appeared as a little b ook in Italian, La scienza e i dati
viii P REF A C E T O F IRS T E D I T I ON

di senso, translated by Michele Leonelli ( Rome: Armando,

1987) . Instead of publishing them intact in English, I have
used portions of them in subsequent publications, as here.
Much of my lecture "Three Indeterminacies, " presented
at the Quine symposium at Washington University in April
1988, is woven into Chapter I, and bits into Chapter V.
That lecture is to appear in the symposium volume, Barrett
and Gibson, editors, Perspectives on Quine ( Oxford: Black­
well) . Another overlapping publication in the offing is
"Truth," written at the request of the Institut International
de Philosophie and slated for Philosophical Problems Today
( The Hague: Nijhofl) . I drew heavily on it for Chapter V ,
b y prior arrangement.
I am blessed with bright and earnest readers . Leonelli
wrote me from Pis a that my new blend of reification with
observation gave him una sona di crampo mentale. After two
letters I began to feel the cramp myself. Result: a substantial
revision ofChapters I and II. A letter from Felix Miihlholzer
in Munich prompted me to insert a couple of paragraphs
recognizing the untidy side of scientific method . A
difficulty spotted by Lars Bergstrom of Stockholm is now
noted and dealt with in the text, and my indebtedness to
Donald D avidson, Dagfinn F011esdal, and Roger Gibson is
noted at appropriate points. I am much indebted to Burton
Dreben, who has read earlier drafts with care and insight
and has made many helpful suggestions.

W.V .Q.
 CONTENTS

I . EV I D EN C E I

I. Stimulation and prediction 1

2. Observation sentences 2
) . Theory-laden? 6
4· Observation categoricals 9
5. Test and refutation 12
6. Holism I)
7. Empirit:4l content 1 6
8. Norms and aims 19

II . REF ER EN C E 23

9. Bodies 2)
10. V.lues of variables 25
11. Utility of reification 29
12. Indifference ofontology )1
I). Ontology defosed ))

I I I . ME A N I N G 37
'4. Thefold linguist's entering wedge )7
15. Stimulation again 40
16. To each his own 42
17. Translation resumed 44
18. Indeterminacy of translation 47
19· Syntax 49
x C O NTE NTS

20. Indeterminacy of reference 50

21. Whither meanings? 52
22. Domestic meaning 53
23. Lexicography 56

I V. I N TE N S I O N 61

24. Perception and observation sentences 61

25. Perception extended 63
26. Perception of things 64
27. Belief and perception 65
28. Propositional attitudes 67
29. Anomalous monism 71
30. Modalities 7 3
31. A mentalistic heritage 74

V. TRUTH 77
32. Vehicles of tru th 77
33. Truth as disquotation 79
34. Paradox 82
35. Tarski's construction 84
36. Paradox skirted 86
37. Interlocked hierarchies 88
38. Excluded middle 90
39. Truth versus warranted belief 93
40. Truth in mathematics 94
41. Equivalent theories 95
42. Irresoluble rivalry 98
43. Two indeterminacies 101

References 105
Credits 109
Index 111
PURSUIT OF TRUTH
 O'W'�l.lI Tel cfHxI.1I6�1IQ.
PLATO

Save the surface and you save all.

SHERWIN-WILLIAMS
 I

EVIDE NCE

1. Stimulation and p rediction

From impacts on our sensory surfaces, we in our collective
and cumulative creativity down the generations have pro­
jected our systematic theory of the external world. Our
system is proving successful in predicting subsequent sen­
sory input. How have we done it?
Neurology is opening strange new vistas into what goes
on between stimulation and perception. Psychology and
more particularly psycholinguistics may be looked to for
something to say about the passage from perception to ex­
pectation, generalization, and systematization. Evolution­
ary genetics throws further light on the latter matters, ac­
counting for the standards of similarity that underlie our
generalizations and hence our expectations. The heuristic of
scientific creativity is illuminated also, anecdotally, by the
history of science.
Within this baffling tangle of relations between our sen­
sory stimulation and our scientific theory of the world,
there is a segment that we can gratefully separate out and
clarify without pursuing neurology, psychology, psycho­
linguistics, genetics, or history. It is the part where theory is
tested by prediction. It is the relation of evidential support,
2 P U R S U I T OF T R U T H

and its essentials can be schematized by means oflittle more

than logical analysis .
Not that prediction is the main purpose of science . One
major purpose is understanding . Another is control and
modification of the environment. Prediction can be a pur­
pose too , but my present point is that it is the test of a theory,
whatever the purpose.
It is common usage to say that the evidence for science is
observation, and that what we predict are observations . But
the notion of observation is awkward to analyze. Clarifica­
tion has been sought by a shift to observable objects and
events. But a gulf yawns between the m and our immediate
input from the e xternal world, which is rather the trigger­
ing of our sensory receptors. I have cut through all this by
settling for the triggering or stimulation itself and hence
speaking, oddly perhaps, of the prediction of stimulation .
By the stimulation undergone by a subj ect on a given occa­
sion I j ust mean the temporally ordered set ofa11 those of his
exteroceptors that are triggered on that occasion.
Observation then drops out as a technical notion. So does
evidence, if that was observation . We can deal with the
question of evidence for science without help of 'evidence '
as a technical term . We can make do instead with the notion
of observation sentences .

2. Observation sentences

We were undertaking to examine the evidential support of

science. That support, by whatever name, comes now to be
seen as a relation of stimulation to s cientific theory. Theory
consists of sentences, or is couched in them; and logic con­
nects sentences to sentences. What we need, then, as initial
 EV I D EN C E 3
links in those connecting chains, are some sentences that a re
direcdy and firmly associated with our stimulations. Each
should be associated affirmatively with some range of one's
stimulations and negatively with some range. The sentence
should command the subject'S assent or dissent outright, on
the occasion of a stimulation in the appropriate range, with­
out further investigation and independendy of what he may
have been engaged in at the time. A further requirement is
intersubj ectivity: unlike a report of a feeling, the sentence
must command the same verdict from all linguistically
competent witnesses of the occasion.
I call them observation sentences. Examples are 'It's rain­
ing' , ' It's getting cold', ' That's a rabbit' . Unlike ' Men are
mortal' , they are occasion sentences: true on some occasions,
false on others . Sometimes it is raining, sometimes not.
Briefly stated, then, an observation sentence is an occasion
sentence on which speakers of the language can agree out­
right on witnessing the occasion. See further § I S.
Observationality is vague at the edges . There are grada­
tions in an individual 's readiness to assent. What had passed
for an obser vation sentence, say 'That's a swan', may to the
subject's own surprise leave him undecided when he en­
counters a black specimen. He may have to resort to con­
vention to settle his usage. We shall need now and again to
remind ourselves thus of the untidiness of human behavior,
but meanwhile we foster perspicuity by fancying bound­
aries.
The range of stimulations associated with an observation
sentence, affirmatively or negatively, I call its affirmative or
negative stimulus meaning for the given speaker. Each of the
stimulations, by my definition, is global: it is the set of all
the triggered exteroceptors, not just the ones that happened
 PURSUIT OF T R U T H
to elicit behavior. Hence the stimulations encompassed in a
stimulus meaning will differ wildly from one another in
their ineffective firings, but in their effective core they are
bound to be similar to one another in some respect, by the
subj ect ' s lights; l similar, that is, in eliciting similar behav­
ior. His according them all the same obser vation sentence is
itself a case of similar elicited behavior.
An obse rvation sentence may consist of a single noun or
adjective, thought of as a sentence; thus ' Rain' , 'Cold', or
'Rabbit' , for ' It's raining ' , 'It's cold ' , ' It's a rabbit'. Obser­
vation sentences also may be compounded to form further
observation sentences, for example by simple conjunction:
'The sun is rising and birds are singing'. Another way of
compounding them is predication: 'This pebble is blue', as a
compound of 'lo, a pebble' and 'lo, blue'. An equivalent
rendering is simply 'Blue pebble' ; they have the same
stimulus meanin g . But they are not equivalent to the mere
conjunction 'Lo , a pebble, and 10, blue'. Their c onnection is
tighter. The conj unction is fulfilled so long as the stimula­
tion shows each of the component observation sentences to
be fulfilled somewhere in the scene-thus a white pebble
here, a blue flower over there. On the other hand the predi­
cation focuses the two fulfill m ents, requiring them to coin­
cide or a mply overlap . The blue must encompass the peb­
ble. It may also extend beyond; the construction is not
symmetric.
What brought us to an examination of observation sen­
tences was our q uest of the link between observation and
theory. The observation sentence is the means of verbaliz­
ing the prediction that checks a theory. The requirement

1 Hence perceptually similar, not receptually. Roots of Reformce,

pp. 16-18.
 EV ID EN C E s

that it command a verdict outright is what makes it a fmal

checkpoint. The requirement of intersubjectivity is what
makes science objective.
Observation sentences are thus the vehicle of scientific
evidence, we might say-though without venturing a defi­
nition of 'ev idence' itself. But also they are the entering
wedge in the learning of language. The infant's first acqui­
sitions in cognitive language are rudimentary observation
sentences, including 'Mama', 'Milk', and the like as one­
word observation sentences. They become associated with
stimulations by the conditioning of responses. Their direct
association with concurrent stimulation is essential if the
child is to acquire them without prior language, and the
requirement of intersubj ectivity is essential in order that he
learn the expressions from other speakers on appropriately
shared occasions.
That observation sentences serve in both ways-as vehi­
cles of scientific evidence and as entering wedge into lan­
guage-is no cause for wonder. Observation sentences are
the link between language, scientific or not, and the real
world that language is all about.
Observation sentences as I have defined them far exceed
the primitive ones that are the child's entering wedge.
Many of them are learned not by simple conditioning or
imitation, but by subsequent construction from sophis­
ticated vocabulary. The requirement of direct correspon­
dence to ranges of stimulation can be met either way.
Which ones are learned directly by conditioning, and which
ones indirectly through higher language, will vary from
person to person. But the two requirements, intersubjectiv­
ity and correspondence to stimulation. assure us that any
observation sentence could be learned in the direct way. We
6 P U R S U I T OF T R U T H
hear our fellow speakers affirming and denying the sentence
on just the occasions when we are stimulated in the charac­
teristic ways, and we join in.

J. Theory-laden?
My definition of observation sentence is of my devising,
but the term is not. Philosophers have long treated in their
several ways of what they called observation terms or ob­
servation sentences. But it has now become fashionable to
question the notions, and to claim that the purportedly ob­
servable is theory laden in varying degrees. It is pointed out
-

that when scientists marshal and check their own data or

one another's, they press no fa rther than is needed to assure
agreement among witnesses conversant with the subject ;
for they are reasonable men. 'The mixture is at 180°C' and
'Hydrogen sulfide is escaping are observational enough for
'

any of them, and more recondite reports are obse rvatio nal
enough for some. I agree that the practical notion of obser­
vation is thus relative to one or another limited community,
rather than to the whole speech community. An observa­
tion sentence for a community is an occasion sentence on
which members of the community can agree outright on
witnessing the occasion.
For philosophical purposes we can probe deeper how­,

ever, and reach a single standard for the whole speech com­
munity. Observable in this sense is whatever would be at­
tested to on the spot by any witness in command of the
language and his five senses. If scientists were perversely to
persist in demanding further evidence beyond what sufficed
for agreement, their observables would reduce for the most
part to those of the whole speech community. Just a few,
 EV I D EN C E 7
such as the indescribable smell of some uncommon gas,
would resist reduction.
But w hat has all this to do with a sentence's being theory­
la den or theory-free? My definition distinguishes obser­
vation sentences from others, whether relative to special
communities or to the general one, without reference to
theory-freedom. There is a sense, as we shall now see, in
which they are all theory-laden, even the most primitive
ones, and there is a sense in which none are, even the most
p rofessional ones.
Think first of primitive ones, the entering wedge in lan­
guage learning. They are associated as wholes to appropri­
ate ranges of stimulation, by conditioning. Component
words are there merely as component syllables, theory­
free. But these words recur in theoretical contexts in the
fullness of time. It is precisely this s haring of words, by
observation sentences and theoretical sentences, that pro­
vides logical connections between the two kinds of sen­
tences and makes observation relevant to scientific theory.
Retrospectively those once innocent observation sentences
are theory-laden indeed. An observation sentence contain­
ing no word more technical than 'water' will join forces
with theoretical sentences containing terms as technical as
'H20' . Seen holop hrastically, as conditioned to stimulatory
situations, the sentence is theory-free; seen analytically,
word by word, it is theory-laden. Insofar as observation
sentences bear on science at all, affording evidence and tests,
there has to be this retrospective theory-lading along with
the pristine holophrastic freedom from theory. To impugn
their observationality thus retrospectively is to commit
what Firth (p. 100) called the fallacy of conceptual retrojec­
tion.
8 P U R SU IT OF TRU TH
More sophisticated observation sentences, inc lud ing
those of specia li zed scientific communities , are sim ilarly
tw o-faced, even tho ugh learned by composition rather than
direct conditio ning. What qu alifies them as obse rvation
sentences is still their holophras tic associ ation with fIXed
ranges of sensory stimul ation, however that association be
acq uired. Holophrastical ly they function sti ll as theory­
free, like C. I. Lewis's "expressive" sentences (p. 1 79),
tho ugh when taken retrospectively word by word the self­
same senten ces are theory- laden , like his "ob jective" ones.
When epistemology rounded the linguistic tum, talk of
observable objects ga ve way to tal k of observation terms. It
was a g ood mo ve, but not good enough. Observation sen­
ten ces w ere distinguished from theoret ical ones only
derivati vely, as containing observation terms to the exclu­
sion of theory-laden or theoretical terms. Consequently
Rei chenbach and others felt a need for "bridg e p rincip les "
to re late the two kin ds of sentences. No bridge is wanted,
we now see, and bridging is the w rong figure. Start in g with
sentences as we have done rather than with terms, we see no
bar to a sharing of voc ab ulary by the two kinds of sentences ;
and it is the shared vocabulary that links them.
Starting with sentences has conferred the further boon of
freeing the definition of observation sentence from any de­
pendence on the dist inction between the theory -free and the
theo ry-laden. Yet a third adv antage of this move is th at we
can then study the acquisition and use of observ ation sen­
tences without p rejudging what objects, if any , the compo­
nent word s are meant to refer to. We thus a re freed to
spec ulate on the nat ure of reification and its utility for
scientific theory -a topic for Chap ter II. Taking terms as
starting point wou ld h ave me ant finessing reification and
 EVI D ENC E 9
con ceding objective reference out of hand, without consid­
e ring what it is for or what goes into it.

4. Observation categoricals
The support of a theory by observation stands forth most
expl icitly in experiment, so let us look i nto that. The sci en­
t ist has a backlog of accepted theory, and is considering a
hypothesis for possible in corporation into it. The theo ry
tell s him that if the hypothesis under consideration is t rue,
then, whenev er a certain obser vable situation is set up, a
cert ain effect should be observed. So he set s up the situation
in qu estion. If the p redi ct ed effect fail s to appear, he aban­
don s hi s hypothesis. If the effect doe s appear, hi s hypothesis
may b e t rue and so can be tentatively added to hi s ba cklog of
theory.
Th us suppose a team of fi eld mineralogist s have tu rned
up an unfam iliar crystalline mineral of a distinc tively p in k­
ish cast. They speak of it provisionally as litholite, for want
of a better name. One of them conjectures its ch emical
co mposition. This is the hypothesis, of whi ch I shall spare
m yself the detail s. From his ba cklog of ch emi cal lore he
reason s t hat if this chemical hypoth esis is true , then any
pie ce oflitholite should em it h ydrogen sulfide when he ated
above 180°C. Th es e last p rovi sion s a re the observables ; for
o ur min eralogist and hi s colleagues know litholite when
they see it and hydrogen sul fide when they smell it, and
t hey can read a the rmometer.
The t est of a hypothesis thu s hin ges on a lo gical rel ation
of i mpli cation. On one side, the theoretical, we h ave the
backl og of accepted theory plus the h ypothesis. This com­
bination does the implying. On the other side, the observa-
10 P U R SU I T OF T R U TH
tional, we have an implied generality that the experimenter
can directly test, directly challenge-in this case by heating
some of the pink stuff and sniffing.
A generality that is compounded of observables in this
way-'Whenever this, that'-is what I call an observation
categorical. It is compounded of observation sentences. The
'Whenever' is not intended to reify times and quantify over
them. What is intended is an irreducible generality prior
to any objective reference. It is a generality to the effect
that the circumstances described in the one observation sen­
tence are invariably accompanied by those described in the
other.2
Though compounded of two occasion sentences, the ob­
servation categorical is itself a standing sentence, and hence
fair game for implication by scientific theory. It thus solves
the problem of linking theory logically to observation, as
well as epitomizing the experimental situation.
That situation is where a hypothesis is being tested by
an experiment. An opposite situation is equally familiar: a
chance observation may prompt us to conjecture a new
observation categorical, and we may invent a theoretical
hypothesis to explain it. For example, we might notice wil­
lows leaning over a stream. This suggests the observation
categorical:

(I) When a willow grows at the water's edge, it leans

over the water.
2
The observation categorical is not to be confused with the observa­
tion corulitioltl,z, a less fruitful notion that I ventured in 1975. The observa­
tion conditional is formed from two standing sentences each of which has
been built upon an observation sentence with help of theory. See Theories
arul Things, pp. z6-l,7.
 EVI D E N C E II

This suggests, in tum, a theoretical hypothesis : 'A willow

root nourishes mainly its own side of the tree' . Taken t o­
geth er with prior bits o f theory , such a s that roots get more
nourishment from wetter ground, and that nourishment
p ro m ote s the growth of branches , the hypothesis is found
to imply the obser vation categorical. Other ob se rvation
cate go ricals will be imp lied too, and the continued testin g
of th e hypothesis would proceed b y t esting various of
them , a long with further testing of the one that happened to
sugge st the hypothesis.
The o bservation categorical (1) exceeds my definition in a
sub tle way : it is not compounded of two self-su fficient o b­
serv ation sentences. It cannot be read 'When a w illow
grow s at the water 's edge, a willow leans o ver the water'.
The component o bse rvation sentences ha ve to bear not ju st
on the same scene, thi s time, but on the same part of the
scene, the same w il low. Such was the force of'it' in (1). We
h ave what may be called afocal observation categorical , as
distinct from a free one.
In §2 we saw a contra st between conjunction and predica­
tion. Now the free o bservation c atego ri cal genera lizes
merely on a conjunction, and claims that every occasion
p re sent ing the one feature w ill p resen t the other somewhere
about. The focal observation categorical generalizes ra ther
on a pre dica tiona l observation sentence. (1) gener alizes o n
the predication 'This riverine willow leans over the w ater'
t o say that they all do.
A more succinct p red icati onal observation sentence is
'This ra v en i s blac k', o r 'Black raven'. It g enerali z es to the
focal observation categ ori c a l 'When ever there is a raven , it
is bla c k or, succinctly, 'All ravens are b la ck .
', '
 P URS U IT OF TRU TH
.5. Test and refotation
An observation categorical is tested by pairs of observa­
tions. It is not conclusively verified by observations that are
conformable to it, but it is refuted by a pair of observations,
one affirmative and one negative-thus observation of
litholite at I80Ge but absence of hydrogen sulfide, or obser­
vation of riverine willows leaning away from the water.
The free observation categorical 'When the sun comes up
the birds sing' is refuted by observing sunrise among silent
birds.
The observational test of scientific hypotheses, in turn,
and indeed of sentences generally, consists in testing obser­
vation categoricals that they imply. Here again, as in the
case of the observation categorical itself, there is no conclu­
sive verification, but only refutation. Refute an observation
categorical, by an affirmative and a negative observation,
and you have refuted whatever implied it.
Traditional epistemology sought grounds in sensory ex­
perience capable of implying our theories about the world,
or at least of endowing those theories with some increment
of probability. Sir Karl Popper has long stressed, to the
contrary, that observation serves only to refute theory and
not to support it. We have now been seeing in a schematic
way why this is so.
But again we must bear in mind, as in §2, that we are
schematizing: positing sharp boundaries where none can be
drawn. The pair of observations in purported refutation of
an observation categorical may be indecisive because of un­
foreseen indecision over the stimulus meaning of one of the
pair of observation sentences, as in the case of the black
swan or an albino raven. A theory that implied the observa­
tion categorical 'All swans are white', or 'All ravens are
 EV ID E N C E 13
black', might or might not be refuted by the discovery of
the odd specimen, depending on our own decision regard­
ing the vague stimulus meaning of the word. In both exam ­
ples the verbal usages actually adopted, which do admit
black swans and blond ravens, are the ones that make for the
s moother terminology in the overall theory.
It is clearly true, moreover, that one continually reasons
not only in refutation of hy potheses but in support of them.
This, however, is a matter of arguing logically or probabil­
istically from other beliefs already held. It is where the tech­
nology of probability and mathematical statistics is brought
to bear. Some of those supporting beliefs may be observa­
tional, but they contribute support only in company with
others that are theoretical . Pure observation lends only
negative evidence, by refuting an observation categorical
that a proposed theory implies.

6. Holism
Let us recall that the hypothesis regarding the chemical
composition of litholite did not im p ly its observation
categorical single-handed. It implied it with the help of a
backlog of accepted scientific theory. In order to deduce an
observation categorical from a given hy pothesis, we may
have to enlist the aid of other theoretical sentences and of
many common-sense platitudes that go without saying,
and perhaps the aid even of arithmetic and other parts of
mathem atics.
In that situation, the falsity of the observation categorical
does not conclusively refute the hy pothesis. What it refutes
is the conjunction of sentences that was needed to imply the
observation categorical. In order to retract that conjunction
14 P URS U IT OF TRU T H
we do not ha ve to retract the hypothesis in question; we
could retract some other sentence of the conj unction in­
stead. This is the important insight called holism. Pierre
Duhem made muc h of it early in this century , but not too
m uch.
The scientist thinks of his experiment as a test spe cifically
of his new hypothesis , but o nly because t his was the sen­
tence he was wonde ring about and is prepared to re ject.
Moreo ver, there are also the situations where he has no
preconcei ved hypothesis, but just happens upon an anoma­
lo us phenomenon . It is a case of his happening upon a
co unter-instance of an observation categorical which, ac­
cording to his c urrent theory as a whole, ought to have been
true. So he loo ks to his theory with a c ritical eye.
Over-Iogici zing, we may pict ure the accommodation of
a failed observation catego rical as follows. We ha ve before
us some set S of pu rported truths that was fo und jointly to
imply the false categori cal. Implication may be taken here
simply as deducibility by the logic of t ruth functio ns,
quantification, and identity. (We can always pro vide for
more substan tial conseq uences by incorporating appropri­
ate premisses explicitly into S.) Now some one or more of
the sentences in S are going to have to be rescinded. We ex­
empt some membe rs of S from t his thre at on determining
that the fateful implication still holds witho ut their help .
Any p urely logical truth is thus exempted, since it adds
nothing to what S wo uld logically imply anyway ; and sun­
d ry irrele vant sentences in S will be exempted as well . Of
the remaining mem bers of S, we rescind one that seems
most suspect, or least c rucial to our o verall theory. We heed
a maxim of mi nimum m utilation. If the remaining mem­
bers of S s till conspire to imply the fa lse categorical, we try
 EVI DENCE 15

res cinding another and restoring the first. If the false

categorical is still implied, we try rescindi ng both. We con­
tinue thus until the implication is defused.
But this is only the beginning. We must also track down
sets of s entences elsewhere, in our overall th eo ry, that im­
pl y th ese newly rescinded beliefs ; for thos e must be defused
too. We continue thus until co nsistency seems to be re­
s tored. Such is th e mutilation that the ma xim of min imum
muti lation is meant to minim ize.
In p articular the maxim constrains us, in our choice of
what sentences of S to rescind, to safeguard any purely
mathematical t ruth ; for mathemat ics in filtrates all branches
of our system of the world, and its disruption would r e­
verb erate intolerably. If asked why he spare s mathematics,
the scientist will perhaps say that its laws are necessaril y
true; but I think w e have here an explanation, rather, of
mathem atical necessity itself. It resides in our unstated pol­
i ey of shield ing mathematics by exercising our freedom to
re ject other beliefs instead.
So the choice of which of th e b eliefs to r eject is i ndiffer ent
only so far as the failed observation categorical is con­
cerned, and not on other counts. It is well, we saw, not to
rock the boat more than need be. Simplic ity of the resulting
theory is another guiding consideration, however, and if
the scien tist sees his way to a big g ain in simplicity he is even
prepared to rock the boat very considerably for the sake of
it. But th e ultimat e objective is so to choos e the r evision a s
to m aximi ze future success i n pr ediction: future coverage of
true observation categoricals. There is no recipe for this,
but m aximization of simplicity and minimi zation of muti­
lation are maxims by which science strives for v indication
in future predi ctions.
16 PURSUIT OF T R UTH
I t is difficult to see how anyone can question holis m, in
the sense now b efo re us. G riinbaum has indeed argu ed
against ho lism , but in a stronger sense than is h ere enter­
tained. He co nstrues holism as claiming that when a predic ­
tion fai ls , we can alw ays save the threatened hy pothesis by
so revising the bac klog of accepted theory that it , plus the
threatened hypothesis, w ill imp ly the failure of the predic­
tion. I am making no such p resumption. Inacti vatin g the
false implication is all that is at stake. Explaining the un ex­
pected counter-observation is quite anothe r step of scien­
tific progress, which may o r may not be made in th e fu llness
of t ime.
Holism in t his moderate sense is an obvious but vita l
correction of the nai ve conception of scient ific sentences as
endowed each with its own s epa rab le empirical content.
Content is shared, e ven by mathematics insofar as it g ets
applied.

7. Empirical content
Stimu lus meanings ha ve fu zzy boundaries, as witness ag ain
the black swan and the a lbino raven. If we i magine sha rp
dema rcation, howev er, we can bu ild up to a decepti ve ly
p recise but withal inst ructi ve definition of empirical con­
tent.
C all an observation categorical a na lytic for a given
speake r if, as in 'Robins a re birds ', the affirmati ve stimulus
meaning fo r him of the one compon ent is inc luded in t hat
of the other. Otherwise synthetic. Call a sentence o r set of
sentences testable if it imp lies some synthetic observation
categorica ls . Call two observa tion categoricals synonymous
 EVIDE N C E 17

if th eir respective components have the same stimulus

mea nings. Then the empirical content of a testable sentence or
s et of sentences for that speaker is the set of all the synthetic
ob s ervation categoricals that it implies, plus all synony­
mo us ones. I add the synonymous ones so that merely ver­
b al variation will not obstruct sameness of content.
Having thus defined empirical content and hence empir­
ica l equivalence for the individual speaker, we can call two
sentences or sets of sentences equivalent for a whol e com­
munity when equivalent for each member.
Some unconjoined single sentences qualify as testable,
notably the synthetic observation categoricals themselves.
For the most part, however, a testable set or conjunction of
sentences has to be pretty big, and such is the burden of
holism. It is a question of critical semantic mass.
We must recognize, along with the idealization of
stimulus meanings, a significant degree of idealization in
the foregoing account of hypothesis-testing. The scientist
does not tabulate in advance the whole fund of theoretical
tenets and technical assumptions, much less the common­
sense platitudes and mathematical laws, that are needed in
addition to his currently targeted hypothesis in order to
imply the observation categorical of his experiment. It
would be a Herculean labor, not to say Augean, to sort out
all the premisses and logical strands of implication that ulti­
m a tel y link theory with observation, if or insofar as linked
they be.
Worse, it seems that in many cases no such marshaling of
tacit p remisses could quite clinch the observation categor­
ical, becau se of vagueness. The situation is illustrated by the
near-platitude
18 P U RSU I T O F T R U TH

(I) Sodium chloride dissolves in water.

Notoriously , this is tenab le on ly ceteris paribus, and the cetera
are left vag ue. Normally one j ust treats (1) as tr ue and ad­
mits it to the ba cklog of auxiliary tenets , impli citly or ex­
p licitly . If an experimenter faced with a negative res ult
e le cts to save his hypothesis by tampering with the auxiliary
tenets, and with (1) in parti cular, he will do so by devel­
oping a subsidiary th eory to a ccount for an ex ception to (1).
In general (1) is a ccepted as a vag ue statement of strong
probability , open to ques tion only where the improbable
counter-instan ce can be p la usibly accounted for.
Similar cushioning s hie lds much of s cien ce , it would
seem, from the simp le probe of observ ation categori cals. It
has e ven been arg ued th at our broadest s cientifi c laws es­
cape eviden ce a ltogether. Yosida writes (pp. 207-208) that
they "may be come o ut offashion , . . . they are never refuted
by direct observation, they are the old so ldiers who never
die but only fade away."
The point of the do ctrine of observation categori cals ,
meanwhile , i s t o exp lain the bearing o f sensory stimu lation
upon s cientific theory so long and insofar as scien ce has not
parted its empiri cal moorings. My con cern has been with
the central logi ca l structure of empirica l evidence. In fused
phrase s of Kant and Russe ll, it is a question of how our
know ledge of the external world is possible. S cience does
stay responsive somehow to senso ry stimu lation both early
and late, b ut its mode of response after the parting of the
moorings eludes my s chematism. My defmition of empir­
i ca l content, a ccordingly , app lies only to sentences and sets
of sentences that are testable in the defined sense of flatly
implying syntheti c observa tion categori ca ls.
 EVI D EN CE 19

8. Norms and aims

I a m of that large minority or sma ll majority who repudiate
th e Cartesian dream of a foundation for scienti fic certainty
fir mer than scientific method itself. But I remain occupied,
we see, with what has been central to traditional epistemol­
og y . namely the re lation of science to its sensory data. I
app ro ach it as an input-output relation wit hin flesh -and­
blo od deni zens of an antecedently acknowledged external
worl d, a relation open to inq uiry as a chapter of the s cience
of that world. To emphasize my dissociation from the
Cartesian dream, I have written of neural receptors and
their stimulation rathe r than of sense or sensibilia. I call the
pursuit natura lized epistemology, but I have no quarrel
with traditionalists who p rotest my retention of the latter
word. I agree with them that repudiation of the Cartesian
dream is no minor deviation.
But they are wrong in protesting that the normative ele­
m en t, so characteristic of epistemo logy, goes by the board.
Insofar as theoretical epistemology gets natural ized into a
chap ter of t heoretical science, so norma tive epistemo logy

gets na turali zed into a chapter of engineerin g: the technol­

og y of anticipating senso ry stimulation.
The most notable norm of naturalized epistemolo gy ac­
tuall y coincides with that of traditional epistemology. It is
simply the watchword of empiricism: nihil in mente quod non
prius in sensu. This is a p rime specimen o f natura li zed epis­
temology, for it is a finding of nat ura l science i tsel f, ho w­
eVer fallible, that our info rma tion about the wo rld co mes
o nly through impacts on our sensory receptors. And still
the point is normative, wa rning us against telepaths and
s o oth s a y e rs
.
20 P U R SU IT O F T R U TH
Moreover, naturalized epistemology on its normative
side is occupied with heuristics generally-with the whole
strategy of rational conjecture in the framing of scientific
hypotheses. In the present pages I have been treating rather
of the testing of a theory after it has been thought up, this
being where the truth conditions and empirical content lie;
so I have passed over the thinking up, which is where the
normative considerations come in. Ullian and I did go into
it somewhat in The Web ofBelieJ, listing five virtues to seek
in a hypothesis: conservatism, generality, simplicity, re­
futability, and modesty . Further counsel is available anec­
dotall y in the history of hard science. In a more technical
vein, normative naturalized epistemology tangles with
margin of error, random deviation, and whatever else goes
into the applied mathematics of statistics. (See §s.)
But when I cite predictions as the checkpoints of science,
I do not see that as normative. I see it as defining a particular
language game, in Wittgenstein's phrase: the game of sci­
ence, in contrast to other good language games such as
fiction and poetry. A sentence's claim to scientific status
rests on what it contributes to a theory whose checkpoints
are in prediction.
I stressed in § I that prediction is not the main purpose of
the science game. It is what decides the game, like runs and
outs in baseball. It is occasionally the purpose, and in primi­
tive times it gave primitive science its survival value. But
nowadays the overwhelming purposes of the science game
are technology and understanding.
The science game is not committed to the physical, what­
ever that means. Bodies have long since diffused into
swarms of particles, and the Bose-Einstein statistic (§I3)
has challenged the particularity of the particle. Even telep-
 EVI D E N C E 21

a th y and clairvoy ance are scientific options, however mor­

ibu nd. It wou ld take some extraordi nary evidence to en­
li ven them , but, if th at were to h appen . then empir icism
i tsel f-the crowning norm , we s aw , of n atural ized epis­
temology -wo uld go by the bo ard. For remem ber tha t tha t
norm . and natura li zed episte molog y itse lf, are integral to
scien ce, and science is falli ble and corri gi ble.
Science a fter such a conv ulsion would still be science , the
s ame old language g ame, hinging still on checkpoin ts in
sensor y prediction. The co ll apse of empiricism would ad­
m it extra input by telepathy or revelation , but the test o f the
resulting science wou ld st ill be predicted sensation.
In that extremity it might indeed be wel l to mo dify the
game itself, and take on as further checkpo ints the predict­
ing o f telepathic and divine inp ut as well as of sensory input.
It is idle to bulw ark definitions ag ainst implausible cont in­
gen Cles.
 II

REFERENCE

9. Bodies
There were advanta ges. we saw (§3) . in sta rtin g with obser­
vation sentences rather than terms. One advan ta ge was that
the nature and utili t y of reification could be deferred for
consideration un ti l an e piste molog i cal se tti ng had b een
sketched in. We are now at that stag e .

Incipi ent reification can alread y be sensed in the prediC"a­

tional ob se rvation sentences (§2) . That mode of combina­
tion favors. as components. observation sentences that fo­
cus on conspicuously limited portions of the scene; for the
co mp ound expresses coincidence of such foci.
A second step of reification, and a step beyond ordinary
ob ser vati on senten ces w as reco gn i zable in th e move to fo­
,

cal ob serva t i on categoricals (§4). I think of the child as first

mastering this construction. like the free observation cate­
gorical, simply as a generali ze d expression of exp ectat ion :
whene ver this. that. For her the difference between the two
kinds of ca tegorical would not at first obtrude. T h e differ­
ence is, we recall . that the focal categ o rical requires the two
features-'Raven' and 'Black', say-to fuse in the scene ,

while the free categ orical does not. However, the scenes
first associated with 'Raven' will show a raven at the salient
focus, and those first associated with 'Black' will show
 P U R SU I T O F TR U TH
black at the salient focus. Insofar, the free categorical al­
ready meets the focal demand. The difference between the
free and the focal in other cases, and between conjunction
and predication (§ 2.) , can gradually dawn on the child in its
own time.
By virtue of its narrowed focus, however, the focal ob­
servation categorical-unlike the free one-has decidedly
the air of general discourse about bodies : willows in the one
example, ravens in the other. This is where I see bodies
materializing, ontologically speaking: as ideal nodes at the
foci of intersecting observation sentences. Here, I suggest,
is the root of reification.
For the very young child, who has not got beyond obser­
vation sentences, the recurrent presentation of a body is
much on a par with similarities of stimulation that clearly
do not prompt reification. Recurrent confrontation of a ball
is on a par at first with mere recurrent exposure to sunshine
or cool air: the question whether it is the same old ball or
one like it makes no more sense than whether it is the same
old sunbeam, the same old breeze. Experience is in its
feature-placing stage, in Strawson's phrase. Individuation
comes only later.
True, an infant is observed to expect a steadily moving
object to reappear after it passes behind a screen; but this all
happens within the specious present, and reflects rather the
expectation of continuity of a present feature than the reifi­
cation of an intermittently absent object. Again a dog's rec­
ognition of a recurrent individual is beside the point; the
dog is responding to a distinctive odor or other trait, un­
available in the case of qualitatively indistinguishable balls.
To us the question whether we are seeing the same old
ball or just a similar one is meaningful even in cases where it
 REFERENCE 25

remains unanswered. It is here that the reification of bodies

is full blown. Our venerable theory of the persistence and
recurrence of bodies is characteristic of the use of reification
in integrating our system of the world. If I were to try to
decide whether the penny now in my pocket is the one that
was there last week, orj ust another one like it, I would have
to explore quite varied aspects of my overall scheme of
things, so as to reconstruct the simplest, most plausible
account of my interim movements, costumes, and expen­
ditures.
Perhaps such indirect equating and distinguishing of
bodies is achieved by some other animals to some extent.
Perhaps a dog seeking a ball that disappeared fairly recently
in one quarter will not settle for a similar ball at an unlikely
distance. However that may be, it seems clear that such
reification of bodies across time is beyond the reach of ob­
servation sentences and categoricals. Substantial reification
is theoretical.

10. Values of variables

Even our sophisticated conception of enduring and recur­
rent bodies, so characteristic of our human ontology, is for
us little more than a beginning. With our progressive sys­
tematization of science we have gone on to reify liquids and
the invisible air, and we have integrated these things with
bodies b y reckoning them as aggregates of bodies too small
to be detected. Nor have we stopped here. Abstract objects
have long since proved indispensable to natural science­
thus numbers, functions, classes.
At this level a question arises of what to count as
reification. and what to count rather as j ust a useful but
 PURSUIT OF TRUTH
ontologically noncommittal tum of phrase; for the idea that
seemed to mark so decisively the reification of bodies,
namely persistence between exposures, makes no sense for
abstract objects. I have urged elsewhere that the most deci­
sive general marks of reification in our language and kin­
dred ones are the pronouns, and indeed it was 'it' in (1) of§4
that signaled those early rumblings of reification in the focal
observation categoricals. The theme is taken up in full by
the relative pronouns and their auxiliaries. 1 When a lan­
guage is regimented in the logical notation of the predicate
calculus. the role of such pronouns is played by bound vari­
ables.
Observation sentences are to be taken holophrastically
from the standpoint of evidence, I urged (§ 3 ). and analyt­
ically word by word from the retrospective standpoint of
theory. From the latter standpoint a focal observation
categorical is an outright quantification. 'Ravens are black'
becomes
Vx(x is a raven ' -+ . x is black) .
Free observational categoricals would be construed simi­
larly, usually by quantifying over times or places.
So I have insisted down the years that to be is to be the
value of a variable. More precisely. what one takes there to
be are what one admits as values of one's bound variables.
The point has been recognized as obvious and trivial, but it
has also been deemed unacceptable. even by readers who
share my general philosophical outlook. Let me sort out
some of the considerations.
The artificial notation '3x' of existential quantification is
explained merely as a symbolic rendering of the words
1 See Thtorits lind Things. pp. S - 6 .
 REFERENCE 27

'there is something x such that'. So, whatever more one

may care to say about being or existence, what there are
taken to be are assuredly j ust what are taken to qualify as
values of 'x ' in quantifications. The point is thus trivial and
ob vious.
It has been objected that what there is is a question of fact
and not of language. True enough. Saying or implying
what there is , however, is a matter of language; and this is
the place of the bound variables.
It has been objected that the logical notation of
quantification is an arbitrary and parochial standard to
adopt for ontological commitment. The answer is that the
standard is transferable to any alternative language, insofar
as we are agreed on how to translate quantification into it.
For predicate-functor logic, thus, the equivalent principle is
that what one takes there to be are what one takes one's
monadic predicates (complements included) to be true of.
For ordinary English what one takes there to be are what
one takes one's relative pronouns to refer to. Ordinary dis­
course is indeed seldom meticulous about ontology, and
consequently an assessment based on the relative pronouns
of ordinary discourse is apt to bespeak a pretty untidy
world; but ontological clarity and economy can be pro­
moted by paraphrase, if one so desires, in terms still of
relative clauses and pronouns rather than quantifiers and
bound variables. The notation of quantification is what is
most usual and familiar, currently, where one is expressly
concerned with ontological niceties; hence my choice of it
as paradigm.
One thinks of reference, first and foremost, as relating
nam es and other singular terms to their obj ects . Yet singu­
lar terms often fail to refer to anything. Conversely, also,
 PURSUIT OF TRUTH
set theory teaches that there are bound to be individually
unspecifiable objects-unspecifiable irrational numbers,
notably-no matter how rich our notation and cumber­
some our expressions. Variables, on the other hand, take all
objects as values, irrespective of specifiability.
Once our language is regimented to fit the predicate cal­
culus, moreover, it is easy and instructive to dispense with
singular terms altogether, leaving variables as the only link
to objects. The underlying principle here is the equivalence
of'3x(Fx and a = x)' to 'Fa'; for this enables us to maneuver
every occurrence of , a' into the context 'a = " and then to
treat that context as an indissoluble predicate ' A ' , absorbing
the singular term. Singular terms can still be recovered
afterward as a convenient shorthand, by introducing singu­
lar description in Russell's way and defining 'a' as
' (U)Ax'.2
I f in some language w e are at a loss t o arrive a t a satisfac­
tory contextual translation of 'there is', and hence of exis­
tential quantification, then we are at a loss to assess the
ontology of the speakers of that language. Some languages
are perhaps so unlike ours that any translation of 'there is'
or '3x', however cunningly contextual, would be too far­
fetched and Procrustean to rest with. To entertain the na­
tion of an ontology at all, known or unknown, for the
speakers of such a language would be an unwarranted pra­
jection on our part of a parochial category appropriate only
to our own linguistic circle. Thus I do recognize that the
question of ontological commitment is parochial, though
within a much broader parish than that of the speakers and
writers of symbolic logic.

2 See Word and Object, pp. 1 76- 1 90 .

 R E FE R E N C E 2.9
J 1. Utility of reijication
We detected the first hint of reification in the predicational
compounding of observation sentences, as contrasted with
simple co nj unction . Predication is a stronger connection
than conj unction; it requires immersion of the pebble in the
blue (§ 2.) and the raven in the black, while mere conjunction
allows the features to go their separate ways.
At its in c e p tion, thus, we find reification contributing to
the logical connections between observation and theory by
tightenin g up on truth functions. Elsewhere I have made
the point more emphatically by a four-part example:
(1) A white cat is facing a dog and bristling.
Four simple observation sentences underlie this. One is
'Cat', or, on the analogy of the ontologically innocent ' It's
raining', 'It's catting' . The others are 'White', 'Dog-facing',
and ' B ristl ing ' . But ( 1 ) cannot be rendered as a mere con­
junction of these four, because the conjunction is too loose.
It tells us only that the four things are going on in the same
scene. We want them all in the same part of the scene,
superi m p osed . It is this tightening that is achieved by sub­
jecting the four-fold conj unction to exis tential quantifica­
tion, thus :
So m ething is catting and is white and is
dog-facing and is bristling .
which is to say ( 1 ) . An object h a s been posited, a cat. 3
For al l its comple xity . ( 1 ) is an observation sentence. It
could be acq uired by direct conditioning to the complex

l My approach here was inspired by Davidson 's logic of adverbs, in his

F.ssays on Action and Events , pp. 1 66 . See my " Events and Rcification . "
30 P U R SU IT O F TRU T H
situation that it report s, i f this situat ion were to recur and be
reported oftener th an one is prepared to expect. But it i s
illustrative o f an unlimited lot of equ ally complex and un­
like ly ob servation sentences. There i s no hope of dir ect
acquisition of each ; systematic construct ion from element s
is mandatory. Reificatio n, we see, to the rescu e.
For purpo ses of that context, a cat of the moment would
su ffice; no need of an enduring cat. To illustrate t he need
of an enduring c at I must go beyond observation sentences
and suppose that we ha ve somehow worked o ur way far
enough up into scientific theory to treat of time ; earlier and
later. Suppose then we wa nt to convey t his t hought:

(2.) If a cat eats a spoi led fi sh and sickens , then she

will thereafter avoid fish.
We cannot treat this as a simple "if-then" compound of two
self-sufficient component sente nces. Like the " and" of the
preceding example. the "if-then" connection i s too weak. It
has to be the same cat in both sentence s, and hence an endu r­
ing cat. O ur sente nce i s really a univer sal ly quantified con­
ditional:
Everything is such that if it is a cat and it eats a
spoiled fi sh and it sicke ns then it will thereafter
avoid fish.
Hilary Putnam and C harles P arso ns have both rem arked
on ways of economizing on abstract ob jects by recourse to a
modal operator of po ssibility. 4 We have ju st observed the
ot her side of the same coin: the po siting of objects can serve
to rei nforce the weak truth functions without recourse to
mod al operators. Where there are such trade-offs to choose
4 Putnam, pp. 47-49: Parsons, pp . 44-47.
 REFERENCE 31
between, I am for positing the obj ects. I posit abstract ones
grudgingly on the whole, but gratefully where the alterna­
ti ve course would call for modal operators. (Cf. §30.)
My examples offer a crude notion of how it may be that
reification and reference contribute to the elaborate struc­
ture that relates science to its sensory evidence. At its most
rudimentary level, reification is a device for focusing obser­
vation sentences convergently; thus ( I ) . Anaphora, clinch­
ing of cross-reference, continues to be its business also at
more sophisticated levels, as in (2) . It is no coincidence that
this is precisely the business also of pronouns, or bound
variables. To be is to be the value of a variable.

12. Indiffirence of ontology

Reference and ontology recede thus to the status of mere
auxiliaries. True sentences, observational and theoretical,
are the alpha and omega of the scientific enterprise. They
are related by structure, and objects figure as mere nodes of
the structure. What particular obj ects there may be is indif­
ferent to the truth of observation sentences, indifferent to
the support they lend to the theoretical sentences, indiffer­
ent to the success of the theory in its predictions.
The point can be accentuated by invoking what I have
called proxy jUnctions. A proxy function is any explicit one­
to-one transformation, J, defined over the objects in our
purported universe. By 'explicit' I mean that for any object
x, specified in an acceptable notation, we can specify fx.

Suppose now we shift our ontology by reinterpreting each

of our predicates as true rather of the correlates Jx of the
objects x that it had been true of. Thus, where 'Px' origi­
nally meant that x was a P, we reinterpret 'Px' as meaning
32 P U R S U I T OF T R U T H

that x is f of a P. Correspondingly for two-place predicates

and higher. Singular terms can be passed over in view of
§ IO. We leave all the sentences as they were, letter for letter,
merely reinterpreting . The observation sentences remain
associated with the same sensory stimulations as before,
and the logical interconnections remain intact. Yet the ob­
jects of the theory have been supplanted as drastically as you
please. s
Sometimes w e can waive the requirement that the proxy
function be one to one. Thus consider Godel 's numbering
of expressions. in the course of his proof of his famous
incompleteness theorem. In one's global theory of things it
would be unnatural to say that the expressions are identical
with those numbers, but still there might be no call to dis­
tinguish them . In that event a proxy function might just as
well treat them alike, assigning the same proxies to the
expressions as to the numbers.
However, one-to-one proxy functions were all I needed
for my present purpose, namely, to show the indifference
of ontology. A more radical case for the indifference of
ontology is afforded by the Lowenheim-Skolem theorem,
in a strengthened form due to Hilbert and Bemays . 6 When
applied to a theory that has been fitted to predicate logic,
cleared o f singular terms, and encompassed in a fmite lot of
axioms, this theorem enables us to express a truth­
preserving reinterpretation of the predicates that makes the
universe come to consist merely of natural numbers 0, 1 , 2,
. . . . This theorem does not, like proxy functions, carry each
of the old obj ects into a definite new one, a particular num­
ber. This was not to be hoped for, since some infinite do-

5 For more see Ontologietli Relativity, pp. 55- S 8 .

6 See Mnhods of Log ie, 4th ed . , pp. 209-2 1 1 .
 REFERENCE 33
mains-notably that of the irrational numbers-are of too
high a cardinality to be exhausted by correlation with natu­
ral numbers . Despite this limitation, however, the rein­
terpretations leave all observation sentences associated with
the same old stimulations and all logical links undisturbed.
Once we have appropriately regimented our system of
the world or part of it, we can so reinterpret it as to get by
with only the slender ontology of the whole numbers; such
is the strengthened Lowenheim-Skolem theore m . But we
could not have arrived at our science in the first place under
that interpretation, since the numbers do not correspond
one by one to the reifications that were our stepping stones.
Practically, heuristically, we must presumably pursue sci­
ence in the old way or within the reach , at least, of proxy
functions .

1J . Ontology de}Used
We found that two ontologies, if explicitly correlated one to
one, are empirically on a par; there is no empirical ground
for choosing the one rather than the other. What is empiri­
cally significant in an ontology is just its contribution of
neutral nodes to the structure of the theory. We could rein­
terpret 'Tabitha' as designating no longer the cat, but the
whole cosmos minus the cat; or, again, as designating the
eat's singleton, or unit class. Reinterpreting the rest of our
terms for bodies in corresponding fashion, we come out
with an ontology interchangeable with our familiar one . As
wholes they are empirically indistinguishable. Bodies still
continue, under each interpretation, to be distinct from
their cosmic complements and fro m their singletons; they
are distinguished in a relativistic way, by their roles relative
34 PURSUIT OF TRUTH
to one another and to the rest of the ontology. Hence my
watchword ontological relativity. But see further §20.
The importance of the distinction between term and ob­
servation sentence shone forth in §§ 3 and 9 , and it does so
again here. 'There's a rabbit' remains keyed to the sensory
stimulations by which we learned it, even if we reinterpret
the term 'rabbit' as denoting cosmic complements or sin­
gletons of rabbits. The term does continue to conjure up
visions appropriate to the observation sentence through
which the term was learned, and so be it; but there is no
empirical bar to the reinterpretation. The original sensory
associations were indispensable genetically in generating
the nodes by which we structure our theory of the world.
But all that matters by way of evidence for the theory is
the stimulatory basis of the observation sentences plus the
structure that the neutral nodes serve to implement. The
stimulation remains as rabbity as ever, but the corre­
sponding node or obj ect goes neutral and is up for grabs.
Bodies were our primordial reifications, rooted in innate
perceptual similarities. It would be gratuitous to swap them
for proxies; the point was just that one could. But our on­
tological preconceptions have a less tenacious grip on the
deliberate refinements of sophisticated science. Physicists
did first picture elementary particles and light waves in anal­
ogy to familiar things, but they have gone on to sap the
analogies. The particles are less and less like bodies, and
the waves seem more like pulsations of energy in the void.
When we get to the positing of numbers and other abstract
objects, I have conj ectured in Roots oj ReJerence that we are
indebted to some fruitful confusions along the way. Lan­
guage and science are rooted in what good scientific lan-
 REFERENCE 35
guage eschews. In Wittgenstein's figure, we climb the lad­
der and kick it away.
Some findings known as the Bose-Einstein and Fermi­
Einstein statistics suggest how we might be led actually to
repudiate even the more traditional elementary particles as
values of variables, rather than retaining them and just ac­
quiescing provisionally in their mysterious ways . Those
results seem to show that there is no difference even in
principle between saying of two elementary particles of a
given kind that they are in the respective places Q and b and
that they are oppositely placed, in b and Q. It would seem
then not merely that elementary particles are unlike bodies;
it would seem that there are no such denizens of space-time
at all, and that we should speak of places Q and b merely as
being in certain states, indeed the same state, rather than as
being occupied by two things.
Perhaps physicists will accommodate this quandary in
another way. But I p rize the example as illustrating the kind
of consideration that could prompt one to repudiate some
hypothetical objects. The consideration is not based on
positivistic misgivings over theoretical entities . It is based
on tensions internal to theory .
Theories can take yet more drastic turns: such not merely
as to threaten a c herished ontology of elementary particles,
but to threaten the very sense of the ontological question,
the question what there is . What I have been taking as the
standard idiom for existential purposes, namely quantifica­
tion, can serve as standard only when embedded in the
standard form of regimented language that we have been
picturing : one whose further apparatus consists only of
truth functions and predicates . I f there is any deviation in
 P U RSU I T OF TRUTH

this further apparatus, then there arises a question offoreign

exchange: we cannot judge what existential content may be
added by these foreign intrusions until we have settled on
how to translate it all into our standard form. Notoriously,
in particular, quantum mechanics invites logical deviations
whose reduction to the old standard is by no means evident.
On one rendering these deviations take the form of proba­
bilistic predications. On an alternative rendering they call
for basic departures from the logic of truth functions. When
the dust has settled, we may find that the very notion of
existence, the old one, has had its day. A kindred notion
may then stand forth that seems sufficiently akin to warrant
application of the same word; such is the way of terminol­
ogy. Whether to say at that point that we have gained new
insight into existence, or that we have outgrown the notion
and reapplied the term, is a question ofterminology as well.
The obj ectivity of our knowledge of the external world
remains rooted in our contact with the external world,
hence in our neural intake and the observation sentences
that respond to it. We begin with the monolithic sentence,
not the term. A lesson of proxy functions is that our ontol­
ogy, like grammar, is part of our own conceptual contribu­
tion to our theory of the world. Man proposes; the world
disposes, but only by holophrastic yes-or-no verdicts on
the observation sentences that embody man's predictions .
 III

ME ANIN G

14. The field linguist's entering wedge

Philosophers in ancient India disputed over whether sen­
tences or words were the primary vehicles of meaning. The
argument in favor of words is that they are limited in num­
ber and can be learned once for all. Sentences are unlimited
in number; we can fully master them only by learning how
to construct them, as needed, from words learned in ad­
vance. Despite this situation, however, words can still be
said to owe their meaning to their roles in sentences . We
learn short sentences as wholes, we learn their component
words from their use in those sentences, and we build fur­
ther sentences from words thus learned. See §23 .
The quest for a clear and substantial notion of meanings
then should begin with an examination of sentences. The
meaning of a sentence of one language is what it shares with
its translations in another language, so I propounded my
thought experiment of radical translation. It led to a nega­
tive conclusion, a thesis of indeterminacy of translation.
Critics have said that the thesis is a consequence of my
behaviorism . Some have said that it is a reductio ad absurdum
of my behaviorism. I disagree with this second point, but I
agree with the first. I hold further that the behaviorist ap­
proach is mandatory. In psychology one m ay or may not be
 PURSUIT OF T R U TH
a behaviorist, but in linguistics one has no choice. Each of us
learns his language by observing other people's verbal be­
havior and having his own faltering verbal behavior ob­
served and reinfo rced or corrected by others . We depend
strictly on o vert behavior in observable situations. As long
as our command of our language fits all external check­
points, where our utterance or our reaction to someone's
utterance can be appraised in the light of some shared situa­
tion, so long all is well. Our mental life between check­
points is indifferent to our rating as a master of the lan­
guage. There is nothing in linguistic meaning beyond what
is to be gleaned from overt behavior in observable circum­
stances.
In my thought experiment the "source language. " as the
j argon has it, is Jungle; the " target language" is English .
Jungle i s inaccessible through a n y known languages as way
stations, so our only data are native utterances and their
outwardly observable circumstances. It is a meager basis,
but the native speaker himself has had no other.
Our linguist would construct his manual of translation
by conj ectural e x trapolation of such data. but the confirma­
tions would be sparse. Usually the concurrent publicly ob­
servable situation does not enable us to predict what a
speaker even of our own language will say, for utterances
commonly bear little relevance to the circumstances out­
wardly observable at the time; there are ongoing projects
and unshared past experiences . It is only thus, indeed, that
language serves any useful communicative purpose; pre­
dicted utterances convey no news .
There are sentences, however. that d o h i nge pretty
strictly on the concurrent publicly observable situation,
 MEANING 39

namely the observation sentences. We saw these in Chapter

I as the primary register of evidence about the external
world, and also as the child's entering wedge into cogni­
tive language. They are likewise the field linguist's entering
wedge into the jungle language. Other utterances-greet­
ings, commands, questions-will figure among the early
acquisitions too, but the first declarative sentences to be
mastered are bound to be observation sentences, and usu­
ally one word long. The linguist tentatively associates a
native's utterance with the observed concurrent situation,
hoping that it might be simply an observation sentence
linked to that situation. To check this he takes the initiative,
when the situation recurs, and volunteers the sentence him­
self for the native's assent or dissent.
This expedient of query and assent or dissent embodies,
in miniature, the advantage of an experimental science such
as physics over a purely observational science such as as­
tronomy. To apply it the linguist must be able to recognize,
if only conjecturally, the signs of assent and dissent in
Jungle society. If he is wrong in guessing those signs, his
further research will languish and he will try again. But
there is a good deal to go on in identifying those signs. For
one thing, a speaker will assent to an utterance in any cur­
cumstance in which he would volunteer it.
What the native's observation sentence and the linguist's
translation have in common, by this account, is the concur­
rent observable situation to which they are linked. But the
notion of a situation has seemed too vague to rest with. In
earlier writings I have accordingly represented the linguist
as trying to match observation sentences of the jungle lan­
guage with observation sentences of ms own that have the
 PURSUIT OF TRUTH

same stimulus meanings. That is to say, assent to the two

sentences should be prompted by the same stimulations;
likewise dissent.

J5 . Stimulation a gain
It would seem then that this matching of observation sen­
tences hinges on s amenes s of stimulation of both parties,
the linguist and the informant . But an event of stimulation,
as I use the term (§ I ) , is the activation of some subset of the
subject's sensory receptors . Since the linguist and his infor­
mant share no receptors, how can they be said to s ha re a
stimulation? We mi g h t say rather that they unde rg o similar
stimulation, but that would assume still an approximate
homology of nerve endings from one individual to another.
Surely such anatomi cal minutiae ought not to matter here.
I was expressing t h i s discomfort as early as 1 965 . 1 By
1 98 1 it prompted me to readjust my de fi n i tio n of observa­
tion sentence. In my original definition I had appe aled to
sameness of s ti m ulus meaning between speakers . 2 but in
1 98 I I defined it rather for the sin gl e speaker, by the follow­
ing condition:
If querying the s enten ce elicits assent from the given speaker
on one occasion. it will el icit assent likewise on any other
occasion when the same total set of receptors is triggered ;
and similarly for dissent. J

Then I accounted a
sentence observational for a whole com­
munity when it was observational for each member. In this
I E . g . in a lecture " Propositional Obj ects, " published in Onto/i1gictJ/
Relll t ivity aM Othtr Essays .

2 Thus Word and Object, p. 4 3 .

) Theories tJnd Things, p. 2 5 .

 MEANING 41

way the question of intersubjective sameness of stimulation

could be bypassed in studies of scientific method, I felt, and
deferred to studies of translation . There it continued to
rankle.
The question was much discussed in the course of a
closed conference with Davidson, Dreben, and Fellesdal at
Stanford in 1 986. 4 Two years later, at the St. Louis confer­
ence on my philosophy, S Lars Bergstrom observed that
even my bypassing of the question within studies of scien­
tific method was unsuccessful, since a sentence could be
observational for each of various speakers without their
being disposed to assent to it in the same situations. [t is odd
that [ overlooked this, for already in a lecture of 1 974 [ had
remarked in effect that the fisherman's sentence '[ just felt a
nibble' qualifies as observational for all individuals and not
for the group. 6
At the Stanford conference, Davidson proposed provid­
ing for intersubj ective likeness of stimulation by locating
the stimulus not at the bodily surface but farther out, in the
nearest shared cause of the pertinent behavior of the two
subjects. Failing a rabbit or other body to the purpose,
perhaps the stimulus would be a shared situation, if on­
tological sense can be made of situations. But I remain un­
swerved in locating stimulation at the neural input, for my
interest is epistemological, however naturalized. I am inter­
ested in the flow of evidence from the triggering of the
senses to the pronouncements of science. My naturalism

4July 1 4- 1 7, supported by Stanford's Center for the Study of Lan­

guage and Information .

5 " Perspectives on Quine , " Washington University, April 9- 1 3 , 1 98 8 .

6 " The Nature of Natural Knowled g e, " p. 7 1, .

 PURSUIT OF TRUTH
does allow me free reference to nerve endings, rabbits, and
other physical objects, so I could place th e stimulus out
where Davidson does without finessing any reification on
the subject's part. But I am put ofT by the vagueness of
shared situations.

16. To each his own

The view that I have come to, regarding intersubjective
likeness of stimulation, is rather that we can simply do
without it. The observation sentence 'Rabbit' has its stim­
ulus meaning for the linguist and 'Gavagai' has its for the
native, but the affinity of the two sentences is to be sought
in the externals of communication. The linguist notes the
native's utterance of 'Gavagai' where he, in the native's
position, might have said 'Rabbit'. So he tries bandying
'Gavagai' on occasions that would have p rompted 'Rabbit',
and looks to natives for approval. Encouraged, he tenta­
tively adopts 'Rabbit' as translation.
Empathy dominates the learning of language, both by
child and by field linguist. In the child's case it is the parent's
empathy. The parent assesses the appropriateness of the
child's observation sentence by noting the child's orienta­
tion and how the scene would look from there. In the field
linguist's case it is empathy on his own part when he makes
his first conjecture about 'Gavagai' on the strength of the
native's utterance and orientation, and again when he
queries 'Gavagai' for the native's assent in a promising sub­
sequent situation. We all have an uncanny knack for em­
pathizing another's peceptual situation, however ignorant
of the physiological or optical mechanism of his perception.
 MEANING 43
The knack is comparable, almost, to our ability to recog­
nize faces while unable to sketch or descr ibe them.
Empathy guides the linguist still as he rises above obser­
vation sentences through his analytical hypotheses (§ I 7),
though there he i s trying to p roj e ct into the native's associa­
tions and grammatical trends rather than his perceptions.
And much the same must be true of the growing child.
As for the lacuna that Bergstrom noted, my definition of
observation sentence in §2 reflects the correction in a rough
and ready form. More fully: I retain my 1 9 8 1 definition of
observation sentence for the single spe ake r, and then ac­
count a sentence observational for a group if it is observa­
tional for each member and if each would agree in assenting
to it, or dissenting, on witnessing the occasion of utterance.
We judge what counts as witnessing the occasion, as in the
translation case, by projecting ourselves into the witness's
position.
A pioneer manual of translation has its utility as an aid to
negotiation with the native communi ty Success in co m
. ­

munication is judged by smoothness of conversation, by

frequent predictability of verbal and nonverbal reactions,
and by coherence and plausibility ofnative testimony. It is a
matter of better and worse manuals rather than flatly right
and wrong ones. Observation sentences continue to be the
entering wedge for child and field linguist, and they con­
tinue to command the firmest agreement between rival
manuals of translation; but their distinctive factuality is
blurred now by the disavowal of shared stimulus meaning.
What is utterly factual isjust the fluency of conversation and
the effectiveness of negotiation that one or another manual
of translation serves to induce.
 PURSUIT OF TRUTH
In Word and Objec, (p. 8 ) I pointed out that communica­
tion presupposes no similarity in nerve nets. Such was my
parable of the trimmed bushes, alike in outward form but
wildly unlike in their inward twigs and branches. The out­
ward uniformity is imposed by society, in inculcating
language and pressing for smooth communication. In a
computer figure, we are dissimilar machines similarly pro­
grammed. Performance is mandated. implement it how
one may. Such is the privacy of the nerve net. Dreben has
likened it to the traditional privacy of other minds. Now in
my new move I give the subject yet wider berth, allowing
him the privacy even of his sensory receptors.
Unlike Davidson, I still locate the stimulations at the
subject's surface. and private stimulus meanings with them.
But they may be as idiosyncratic. for all I care. as the sub­
ject's internal wiring itself. What floats in the open air is our
common language, which each of us is free to internalize in
his p eculi a r neural way. Language is where intersubjectiv­
ity sets in. Communication is well named.
Obervation sentences are stimulus-synonymous for a
speaker if their stimulus meanings are the same for him. But
whereas one's stimulations and their ranges are a private
affair. stimulus synonymy makes sense socially. Sentences
are stimulus-synonymous for the community if stimulus­
synonymous for each member. This still does not work
between languages, unless the community is bilingual.

1 7. Translation resumed
Our linguist then goes on tentatively identifying and trans­
lating observation sentences. Some of them are perhaps
compounded of others of them. in ways hinting of our
 MEANING 45
logical particles 'and', 'or', 'but', 'not'. By collating the
situations that command the natives' assent to the com­
pounds with the situations that command assent to the
components, and similarly for dissent, the linguist gets a
plausible line on such connectives.
Unlike observation sentences, most utterances resist cor­
relation with concurrent stimulations. Taking the initia­
tive, the linguist may volunteer and query such a sentence
for assent or dissent in various situations, but no correlation
with concurrent stimulation is forthcoming. What next?
He can keep a record of these unconstrued sentences and
dissect them. Some of the segments will have occurred also
in the already construed observation sentences. He will
treat them as words, and try pairing them off with English
expressions in ways suggested by those observation sen­
tences. Such are what I have called analytical hypotheses.
There is guesswork here, and more extravagrant guess­
work to follow. The linguist will turn to the unconstrued,
nonobservational sentences in which these same words oc­
curred, and he will project conjectural interpretations of
some of those sentences on the strength of these sporadic
fragments. He will accumulate a tentative Jungle vocabu­
lary, with English translations. and a tentative apparatus of
grammatical constructions. Recursion then sets in, deter­
mining tentative translations of a potential infinity of sen­
tences. Our linguist keeps testing his system for its efficacy
in dealing with natives, and he goes on tinkering with it and
guessing again. The routine of query and assent that had
been his standby in construing observation sentences con­
tinues to be invaluable at these higher and more conjectural
levels.
Clearly the task is formidable and the freedom for conjec-
 PURSUIT OF TRUTH
ture is enormous. Linguists can usually avoid radical trans­
lation by finding someone who can interpret the language,
however haltingly, into a somewhat familiar one. But it is
only radical translation that exposes the poverty of ultimate
data for the identification of meanings.
Let us consider, then, what constraints our radical trans­
lator can bring to bear to help guide his conjectures. Con­
tinuity is helpful: successive utterances may be expected to
have some bearing on one another. When several such have
been tentatively interpreted, moreover, their interrelation
itself may suggest the translation of a linking word that will
be helpful in spotting similar connections elsewhere.
The translator will depend early and late on psychologi­
cal conjectures as to what the native is likely to believe. This
policy already governed his translations of observation sen­
tences. It will continue to operate beyond the observational
level, deterring him from translating a native assertion into
too glaring a falsehood. He will favor translations that as­
cribe beliefs to the native that stand to reason or are conso­
nant with the native's observed way of life. But he will not
cultivate these values at the cost of unduly complicating the
structure to be ascribed to the native's grammar and se­
mantics, for this again would be bad psychology; the lan­
guage must have been simple enough for acquisition by the
natives, whose minds, failing evidence to the contrary, are
presumed to be pretty much like our own. Practical psy­
chology is what sustains our radical translator all along the
way, and the method of his psychology is empathy: he
imagines himself in the native's situation as best he can.
Our radical translator would put his developing manual
of translation continually to use, and go on revising it in the
light of his successes and failures of communication. The
 MEANING 47
successes consist-to repeat-in successful negotiation and
smooth conversation. Reactions of astonishment or bewil­
derment on a native's part, or seemingly irrelevant re­
sponses, tend to suggest that the manual has gone wrong.
We readily imagine the translator's ups and downs . Per­
haps he has tentatively translated two native sentences into
English ones that are akin to each other in some semantic
way, and he finds this same kinship reflected in a native's
use of the two native sentences. This encourages him in his
pair of tentative translations. So he goes on blithely suppos­
ing that he is communicating, only to be caught up short.
This may persuade him that his pair of translations was
wrong after all. He wonders how far back, in the smooth­
flowing antecedent conversation, he got off the beam .

1 8. Indeterminacy of translation
Considerations of the sort we have been surveying are all
that the radical translator has to go on. This is not because
the meanings of sentences are elusive or inscrutable; it is
because there is nothing to them, beyond what these fum­
bling procedures can come up with. Nor is there hope even
of codifying these procedures and then defining what counts
as translation by citing the procedures; for the procedures
involve weighing incommensurable values. How much
grotesqueness may we allow to the native's beliefs, for in­
stance, in order to avoid how much grotesqueness in his
grammar or semantics?
These reflections leave us little reason to expect that two
radical translators, working independently on Jungle.
would come out with interchangeable manuals. Their
manuals might be indistinguishable in terms of any native
 PURSUIT OF TRUTH
behavior that t he y give reason to expect, and yet each
manual m ig h t prescribe some translations that the other
translator would rej ect Such is t he thesis of indete r min acy
.

of translation.
A manual of Jungle-to-English translation constitutes a
recursive, or inductive, definition of a translation relation
toget her with a claim that it correlates sentences comp a tibl y
with the behavior of all concerned. The thesis of indetermi­
nacy of translation is that these claims on the part of two
manuals m i g h t both be true and yet the two translation
relations might not be usable in alternation, from s entence
to sentence, without issuing in incoherent sequences . Or, to
put it another way, the English sentences prescribed as
translation o f a given Jun gle sentence by two rival manuals
m igh t not be interchangeable in English contexts.
The use of one or the other manual might i n deed cause
differences in speech afterward, as remarked by Robert
Kirk in connection with the idioms of propositional at­
titude; but the two would do equal j ustice to the s tatu s quo.
I have directed my indeter m ina cy thesis on a radically
exotic lan guage for the sake of pla us ib il i t y but in prin ciple
,

it applies even to the home language. For gi ven the rival

manuals of translation between Jungle and English, we can
translate English perv ersel y into Engl ish by translating it
into Jungle by one manual and then back by the other.
The indeterminacy of translation is unlikely to obtrude in
p racti ce , even in radical translation. There is good reason
why it should not. The l in g uis t assumes that the na ti ve s '

attitudes and ways of thinki ng are like his own, up to the

po int where there is contrary evidence. He accordingl y im­
poses his own ontology and lin gui sti c patterns on the native
wherever compati b le with the native's speech and other
 MEANING 49
behavior, unless a contrary course offers striking simplifi­
cations. We could not wish otherwise. What the indetermi­
nacy thesis is meant to bring out is that the radical translator
is bound to impose about as much as he discovers.

19. Syntax
Readers have supposed that I extended my indeterminacy
thesis to syntax. This puzzled me until I became aware,
recently, of a subtle cause of the misconception. In Word and
Object (pp. 5 5 , 68- 7 2) I claimed that our distinctive ap­
paratus of reification and reference is subject to indetermi­
nacy of translation. The apparatus includes pronouns, = " •

plural endings, indeed whatever serves the logical purposes

of quantifiers and variables. But these devices, some of my
readers have reasoned, are part of what syntax is about. So
indeterminacy, they have supposed, extends to syntax.
The business of syntax is the demarcation of strings of
phonemes proper to the language. More than one battery of
grammatical constructions and vocabulary will probably be
capable of generating the same total output of strings, but in
this freedom there is no indeterminacy analogous to that of
translation. Indeterminacy of translation consists rather in
conflict in the outputs themselves. 7
What misled those readers was the indeterminacy of
translation of pronouns and other referential devices. But
that indeterminacy was only over whether to equate certain
Jungle locutions to these devices or to something else. The
translator will accommodate those locutions anyway,
7 The syntactician may indeed exercise some freedom in setting the
limits of the lan guage. but only marginally. See From a Logical Point of
View, pp. 53-55.
so PURSUIT OF TRUTH

whatever his translations. He may or may not call them

pronouns, plurals, quantifiers, and so on, according as he
thinks in terms of one or another manual of translation. The
difference will be only verbal or, at most, a choice of one
syntactic structure rather than another for generating one
and the same total output ofJungle strings.

20 . Indeterminacy of reference
The difference between taking a sentence holophrastically
as a seamless whole and taking it analyticall y term by term
proved crucial in earlier matters (§§3 , 9 , 1 3 ) . It is crucial also
to translation. Taken analytically, the indeterminacy of
translation is trivial and indisputable. It was factually illus­
trated in Ontological Relativity (pp. 3 5-36) by the Japanese
classifiers, and more abstractly above by proxy functions
(§ 1 3). It is the unsurprising reflection that divergent inter­
pretations of the words in a sentence can so offset one an­
other as to sustain an identical translation of the sentence as a
whole. It is what I have called inscrutability of reference;
'indeterminacy of reference' would have been better. The
serious and controversial thesis of indeterminacy of transla­
tion is not that; it is rather the holophrastic thesis, which is
stronger. It declares for divergences that remain unrecon­
ciled even at the level of the whole sentence, and are com­
pensated for only by divergences in the translations of other
whole sentences.
Unlike indeterminacy of reference, which is so readily
illustrated by mutually compensatory adjustments within
the limits of a single sentence, the full or holophrastic inde­
terminacy of translation draws too broadly on a language to
admit of factual illustration. Radical translation is a rare
achievement, and it is not going to be undertaken success-
 MEANING 51
fully twice for the same language. But see Levy for a plausi­
ble artificial example, based on measurement in deviant
geometries. Also there is Massey's sweeping example based
on the duality of affirmation to negation, conjunction to
alternation, and universal quantification to existential. His
rival translations, the homophonic and its dual, conflict on
every sentence. A weakness of this construction is that the
dual manual depends on viewing the natives' volunteered
sentences as denied rather than affirmed-a gratuitous re­
versal of the translator's conventional orientation. Still, in
view of Levy's construction if not Massey's, one can
scarcely question the holophrastic indeterminacy thesis.
A thick and imposing periodical on the philosophy of
language is published twice a year in the Canary Islands
under the title Gavagai. A book by David Premack, on his
language experiments with chimpanzees, came out lately
under the title Gavagai. Hubert Dreyfus has California van­
ity plates on his Volkswagen Rabbit that spell 'GA V AGAr.
The word has become the logo of my thesis of indetermi­
nacy of translation, and now it is making its way in a wider
world. Ironically, indeterminacy of translation in the
strong sense was not what I coined the word to illustrate. It
did not illustrate that, for 'Gavagai' is an observation sen­
tence, firmly translatable holophrastically as ' (Lo, a) rab­
bit'. But this translation is insufficient to fix the reference of
'gavagai' as a term; that was the point ofthe example. It is an
extreme example of the indetenninacy of reference, the
contained term being the whole of the sentence. No room
is left here for compensatory adjustments, and none are
needed.
Kindly readers have sought a technical distinction be­
tween my phrases 'inscrutability of reference' and 'ontolog­
ical relativity' that was never clear in my own mind. But I
S2 PURSUIT OF TRUTH
can now say what o n tolo g ica l r el ati v i t y
is relative to, m ore
s uc cin ctl y than I did i n the lectures, paper, and book of that
title. It is relative to a manual of t ran s la tion . To say that
' gavagai ' denotes rabbits is to opt for a ma n u a l of translation
in which 'ga v ag ai ' is translated as ' rabbit' , instead of opting
for any of the alternative m anu als .
And does the inde t erm ina cy or relativity extend also
somehow to the home language? In " On tolo g ica l Relativ­
ity" I said it did, for the home language can be translated
into itself by permutations that depart materially from the
mere identity transformation, as proxy functions bear o ut .
But if we choose as our manual of translation the identity
transformation, thus taking the home language at face
value, the relativity is resolved . Reference is then ex pli ca ted
in disquotational paradigms anal o g ou s to Tarski's truth
paradigm (§3 3); thus 'rabbit' denotes rabbits, w ha tever they
are, and 'Boston' designates Boston.

21 . Whither meanings?
If we could contrive an acceptable relation of s ameness of
meaning, it would be a sho rt step to an acceptable definition
of me anings . For, as mo re than one philosopher has noted,
we could define the meaning of an expression as the class of
all expressions like it in m eaning . Conversely, if we had the
meanings to begin with, they and identity would provide
sameness of meaning, there being no entity without iden­
tity. 8 In short, meanings and sameness of meaning present
one and the s a m e pro ble m.

8 This platitude has latdy been obscured b y a confusion over the axiom
of extensionality . which individuates sets . or classes. and has been sus­
pended by some set theorists in an exploratory spirit. Might we not
likewise rec o gniz e meanings without identity ? No. Dropping extension-
 MEANING 53
Translation does enjoy reasonable determinacy up
through observation categoricals and into the logical con­
nectives. Thus one could make a stab at the interlinguistic
equating of empirical content (§7) . even in radical transla­
tion. But empirical content pertains only to testable sen­
tences and sets of sentences. We are still left with no general
concept of the meanings of sentences of less than critical
semantic mass.
It is not a conclusion that one readily j umps to or rests
with. One is tempted to suppose that we might define
meanings for sentences of less than critical mass, and even
for terms, by substitutivity. If we can interchange two ex­
pressions without disturbing the empirical content of any
testable context. are they not alike in meaning? Well, the
plan collapses between languages. Interchanging expres­
sions would tum the context into nonsense if the expres­
sions belong to different languages. So the plan offers no
relief from the indeterminacy of translation.

22 . Domestic meaning
Lowering our sights. then, and giving up on "proposi­
tions" as language-transcendent sentence meanings, we
might still look to the substitutivity expedient for a strictly
domestic, interlinguistically inapplicable notion of same­
ness of meaning. Sentences are cognitively equivalent, we
might say, if putting one for the other does not affect the
empirical content of any set of sentences. This sounds right
in principle. For the most part it resists decisive application,

allty does not exempt sets from identity either. It only tables the question
of sufficient conditions for their identity . The notation 'x = y' stays on,
with sets as values of the variables. There is still no entity-no set,
nothing- without identity.
54 PURSUIT OF TRUTH
however, because of the rather visionary status of empirical
content (§7) .
Another approach looks to belief: to the speaker's assent
or dissent when the sentences are queried. This would not
do for standing sentences; it would equate all his beliefs.
But it works for occasion sentences, where we can check
each pair for concomitance over varying occasions. Two
occasion sentences may be accounted cognitively equiva­
lent for a given speaker ifhe is disposed on every occasion to
assent to both or dissent from both or abstain. Derivatively,
then, they may be accounted cognitively equivalent for the
community if cognitively equivalent for each member.
When in particular the sentences are observation sentences,
we are back to stimulus synonymy (§ I 6) .
Cognitive equivalence s o defined then extends immedi­
ately to terms, or predicates. They are cognitively equiva­
lent, or we might now say cognitively synonymous, if their
predications-'It's an F', 'It's a G'-are cognitively equiva­
lent. In view of our definition of cognitive equivalence of
occasion sentences, this boils down to saying that terms are
cognitively synonymous for a speaker if he believes them
to be coextensive, that is, true of the same things; and syn­
onymous for the community if synonymous for each
member.
Some slight progress can then be made toward cognitive
equivalence of standing sentences. Certainly they should be
rated cognitively equivalent if one can be got from the other
by supplanting a component term by a cognitive synonym.
But this does not cover all the pairs of standing sentences
that we would want to regard as cognitively equivalent.
There is a third approach in a,udyticity . Once we have
analyticity, cognitive equivalence is forthcoming; for two
 MEANING 55

sentences are cognitively equivalent i f and only if their

truth-functional biconditional is analytic. Now a sentence
is analytic, in mentalistic semantics, when it is true by virtue
of the meanings of its words. In Roots of Reftrence (pp.
70-80) I suggested externalizing the criterion: a sentence is
analytic if the native speaker learns to assent to it by learning
one or more ofits words. This accounts for such paradigms
of analyticity as 'No bachelor is married', and also for the
analyticity of many elementary logical truths. The concept
can be adjusted to cover also the truths derivable from ana­
lytic truths by analytic steps.
I think this definition does some justice to the intuitive
notion of tautology, the notion that comes into play when
we protest that someone's assertion comes down to '0 = 0'
and is an empty matter of words. But the definition gives
no clue to the demarcation between analytic and synthetic
sentences that has exercised philosophers, out beyond
where anyone either remembers or cares how he learned the
pertinent words. And it gives no clue, certainly, to a general
concept of cognitive equivalence.
Why was it important? Where metaphysics had sought
the essence of things, analytical philosophy as of G. E.
Moore and after settled for the meanings of words; but still
it was as if there were intrinsic meanings to be teased out
rather than just fluctuant usage to be averaged out. Analy­
ticity, then, reflected the meanings of words as metaphysi­
cal necessity had reflected the essences of things. In later
years analyticity served Carnap in his philosophy of mathe­
matics, explaining how mathematics could be meaningful
despite lacking empirical content, and why it is necessarily
true. However, holism settles both questions without ap­
peal to analyticity. Holism lets mathematics share empirical
 P U R S U I T OF T R U T H
content where it is applied, and it accounts for mathematical
necessity by freedom of selection and the maxim of mini­
mum mutilation (§6) .

23 . Lexicography
To question the old notion of meanings of words and sen­
tences is not to repudiate semantics. Much good work has
been done regarding the manner, circumstances, and devel­
opment of the use of words. Lexicography is its conspicu­
ous manifestation. But I would not seek a scientific rehabili­
tation of something like the old notion of separate and
distinct meanings; that notion is better seen as a stumbling
block cleared away. In later years indeed it has been more of
a stumbling block for philosophers than for scientific lin­
guists, who, understandably, have simply found it not
technically useful.
Dictionaries are reputedly occupied with explaining the
meanings of words, and the work is neither myth-bound
nor capricious. How does it proceed? I hold that it is not
directed at cognitive equivalence of sentences, nor at syn­
onymy of terms, and that it presupposes no notion of mean­
ing at all. Let us consider then what the business of dic­
tionaries really is.
Sometimes the dictionary explains a word by supplying
another expression that can replace it salva veritate at least in
positions uncontaminated by quotation or idioms of prop­
ositional attitude. Sometimes, instead, a selection of in­
formation is set down regarding the object or objects to
which the word refers. There is no pretense here of a dis­
tinction between essential and accidental traits. It is a matter
 MEANING 57
purely of pedagogy: the lexicographer wants to improve his
reader's chances of successful communication as best he can
in a small compass. Often, moreover, a dictionary entry
neither paraphrases the word nor describes its objects, but
describes, rather, the use of the word in sentences. This is
usually the way with grammatical particles, and it is often
the way also with terms. It is bound to be the way with a
term that neither refers to concrete objects nor admits of a
separable, self-contained paraphrase.
Behind this seeming disorder there is a unifying princi­
ple. The goal may be seen always as the sentence. The
lexicographer is out to help his reader profit by the sen­
tences that he sees or hears, and to help him react to them in
expected ways, and to help him emit sentences usefully.
But sentences are unlimited in their variety, so the lexicog­
rapher organizes his teaching of sentences word by word,
teaching how to use each word in making sentences. One
way of teaching this, which is convenient when available, is
by citing a substitute expression; for the lexicographer thus
exploits the reader's presumed knowledge of how to use
that substitute expression in making sentences. And the
other sorts of dictionary entry likewise aim, in their differ­
ent ways, at the same end: teaching the use of sentences.
When from semantics as pursued by philosophers we
move to lexicography, we shift our focus from likeness of
meaning to knowledge of meaning, so to speak; from syn­
onymy of expressions, anyway, to the understanding of ex­
pressions. The lexicographer's job is to inculcate under­
standing of expressions, that is, to teach how to use them.
He can be wholly successful in teaching the use of sentences
without considering in what sense they might be said to be
58 P U R S U I T OF T R U T H
equivalent. Nothing, apparently, could be more remote
than meanings from the lexicographer's concern. Why
should they be less remote from ours?
So we might try looking to the understanding of expres­
sions, rather than to synonymy, as the operationally basic
notion of semantics. What sense can we make of it? In prac­
tice we credit someone with understanding a sentence if we
are not surprised by the circumstances of his uttering it or
by his reaction to hearing it-provided further that his reac­
tion is not one of visible bewilderment. We suspect that he
does not understand it if the event is drastically at variance
with those conditions. Still no boundary is evident, no gen­
eral criterion for deciding whether he actually misunder­
stands the sentence or merely holds some unusual theory
regarding its subject matter.
We can be more confident in imputing misunderstanding
of a word than of a sentence, for we can then observe some­
one's use of numerous sentences, or his response to them,
aU of which contain the word. We can control our experi­
ment, choosing and querying sentences ourselves. We may
find that he responds otherwise than would generally be
expected when the sentences contain the word in question,
and that he responds in the more usual ways to many sen­
tences that lack that word but are much the same in other
respects.
In this matter of understanding language there is thus a
subtle interplay between word and sentence. In one way the
sentence is fundamental: understanding a word consists in
knowing how to use it in sentences and how to react to such
sentences. Yet if we would test someone's understanding of
a sentence, we do best to focus on a word, ringing changes
on its sentential contexts . Once we have thus satisfied our-
 MEANING 59

selves, through a multiplicity of such sentences, that he

misunderstands the word, we are justified at last in conclud­
ing that his odd response to the original sentence containing
it was due to a misunderstanding of the word and not to
some odd opinion regarding matters of fact.
Understanding, behaviorall y viewed, is thus a statistical
effect: it resides in multiplicities. The nucleus is the word,
and the mass is made up of the countless sentences in which
the word occurs. A predominantly healthy or unhealthy
coloring of this mass is what counts as understanding or
misunderstanding of the word and the sentences; and a
sharp boundary need not be sought. A forest presents a
sharp boundary to the airborne observer but not to the man
on the ground; an ink blot presents a sharp outline to the
unaided eye and not under a magnifying glass; and we can
acquiesce in a similar attitude toward the distinction be­
tween understanding and misunderstanding an expression.
Lexicography has no need of synonymy, we saw, and it has
no need of a sharp distinction between understanding and
misunderstanding either. The lexicographer's job is to im­
prove his reader's understanding of expressions, but he can
get on with that without drawing a boundary . He does
what he can, within a limited compass, to adjust the reader's
verbal behavior to that of the community as a whole, or of
some preferred quarter of it. The adjustment is a matter of
degree, and a vague one: a matter of fluency and effec­
tiveness of dialogue.
 IV

INTENSION

24 . Percep tion and observation sentences

Observation sentences, typically , are reports of events or
situations in the external world. Some are mentalistic, how­
ever. and they can play an important role. Thus consider, to
begin with. the observation sentence 'It's raining' . Tom is
learning it from Martha by ostension. Martha's business is
to encourage Tom in uttering the sentence, or in assenting
to it, when she sees that he is noticing appropriate phenom­
ena, and to discourage him otherwise. Thus Tom 's mastery
of the physicalistic sentence ' It ' s raining ' hinges on Martha's
mastery, virtual if not literal, of the mentalistic sentence
'Tom percei v es that it's raining ' .
Observation sentences, learned ostensively. are where
our command of language begins, and our learning them
from our elders depends heavily on the ability of our elders
to guess that we are getting the appropriate perception . The
handing down of language is thus implemented by a con­
tinuing command, tacit at least, of the idiom 'x perceives
t hat p ' where 'p' stands for an observation sentence. Com­
mand of this mentalistic notion would seem therefore to be
about as old as language. It is remarkable that the bifurca­
tion between physicalistic and mental i stic talk is fore­
shadowed already at the level of observation sentences, as
62 PURSUIT OF TRUTH
between 'It is raining' and 'Tom perceives that it is raining'.
Man is indeed a forked animal.
Each perception that it is raining is a fleeting neural event.
Two perceptions by Tom that it is raining are apt to differ,
moreover, not only in time of occurrence but neurally,
because there are varied indicators of rain. Tom's percep­
tions of its raining constitute a class of events that is perhaps
too complex and heterogeneous neurally to be practically
describable in neurological terms even given full knowl­
edge of the facts. Yet there is also, we may be sure, some
neural trait that unites these neural events as a class; for it
was by stimulus generalization, or subjective similarity,
that Tom eventually learned to make the observation sen­
tence 'It's raining' do for all of them.
So much for Tom. The class of the whole population's
perce ptions of its raining will be much more forbidding
still, since people's nerve nets differ-certainly in conse­
quence of different histories oflearning, and perhaps genet­
ically as well. Yet the idiom 'perceives that it's raining' cuts
through all that hopeless neurological complexity and en­
capsulates all perception that it is raining-not just on
Tom's part, but on everyone's.
It does so by citing a symptom rather than a neural mech­
anism. And what a remarkable sort of symptom! We detect
it by empathetic observation of the subject's facial expres­
sions and what is happening in front of him, perhaps, and
we specify it by a content clause consisting of a vicarious
observation sentence.
Martha empathizes Tom's perception that it is raining
j ust as the field linguist empathizes the native's perception
that a rabbit has appeared (§ I 6) . Learning a language in the
field and teaching it in the nursery are much the same at the
 INTENSION
level of observation sentences: a matter of perceiving that
the subject is perceiving that p .

25 . Perception extended
Observationality varies with the group of speakers con­
cerned, and is also, within the group, somewhat a matter of
degree (§2) . Consequently the construction 'perceives that
'
p continues to ftourish when the content clause is not obser­
vational, or not very. We even hear 'Tom perceives that the
train is late'.
Consider how one would get on to using that sentence.
People have ways of showing that they perceive that the
train is late, and these ways run to type. One way is by
saying that the train is late. Also people pace impatiently,
they look at the clock, they look along the track. Along
with acquiring such habits ourselves, we have learned to
observe similar manifestations on the part of others. We are
ready to see our own ways replicated in another person.
This readiness was what enabled us to teach observation
sentences to other persons, and to learn when to affirm ' x
'
perceives that p in observational cases; and the ability ex­
tends beyond observation sentences to sentences like 'The
train is late' .
The evidence is not assembled deliberately. One em­
pathizes, projecting oneself into Tom's situation and Tom's
behavior pattern, and finds thereby that the sentence 'The
train is late' is what comes naturally. Such is the somewhat
haphazard basis for saying that Tom perceives that the train
is late. The basis becomes more conclusive if the observed
behavior on Tom's part includes a statement of his own that
the train is late.
 PURSUIT OF TRUTH
An occasion sentence of the form 'x perceives that p ' can
be true even when the content clause is a standing sentence,
such as 'Randy is a dog', rather than an occasion sentence.
What is then required, however, is not only that the perci­
pient be prepared on that occasion to assent to that clause,
but also that he be just then becoming aware that it is true.
Ascriptions of perceptions call increasingly for background
knowledge and conjecture on the ascriptor's part as we
move away from observation sentences.

26. Perception of things

'
Alongside the construction 'x perceives that p , where the
perception is described by a content clause, we have the
construction 'x perceives y', where the perception is de­
scribed by a term as objective complement. The term desig­
nates an object that incites sensory receptors that arouse the
percipient's attention. If the object is a bowl, the force may
be the light that it reflects to the eye. But that light is coming
also from the sun, or from a lamp, via the bowl. What
distinguishes the perceived object is perhaps that the force
comes from it directly? No, this will not do; we want also to
allow the bowl to be perceived by reflection in a mirror.
There is an easy solution: focus. Between perceiving the
bowl reflected in the glass and perceiving the glass itself
there is a difference in the tension of the eye muscles; for the
focal distance of the bowl is the total distance from the eye
to the glass and thence to the bowl. The same criterion of
focus serves to distinguish between seeing something
through a glass and seeing the glass.
But focal distance and causality do not suffice to single
out the perceived object. A bit of the surface of the bowl
 INTENSION
would meet those conditions as well as the bowl itself.
Anything of which that patch of surface is a part would
likewise qualify-thus the bowl, or the nearer half of the
bowl, or any sector of the environment that includes the
bowl, taken over any period of time that includes the
stimulatory event. Which of these objects to count as per­
ceived would be settled by the percipient's observation sen­
tence, if he were to volunteer one. I
We noted earlier (§§ 3 , 9 , 1 3 ) that observation terms, ret­
rospectively seen as designating objects, are best viewed at
their inception rather as one-word observation sentences.
The same attitude best befits the ascription of perceptions:
think of ' x perceives y' in the image rather of ' x perceives
that p'. We say 'Tom perceives the bowl' because in em­
pathizing Tom's situation we fancy ourselves volunteering
the observation sentence 'Bowl' rather than 'Surface of a
bowl', 'front half of a bowl', 'Bowl and background'.
Adherence to content clauses, in preference to perceived
objects, imposes no real restraint on our day-to-day ascrip­
tions of perception. When we ask 'What did he perceive?'
we are content with an answer of the form 'He perceived
'
that p . When we say 'They perceived the same thing'
'
a fuller explanation in the form 'They perceived that p af­
fords satisfaction. The stubbornly substantival 'What' and
'same thing' intrude only for lack of words to stand for
clauses.

27. Belief and perception

'
The idiom ' x perceives that p applies beyond observation
sentences, we saw, and even beyond occasion sentences
I For further quandaries about perceived objects see Chisholm. ch 10.
68 PURS UIT OF TRUTH
' '
whether p , and indeed 'says that p . Empathy figures in
most ascriptions of these kinds, to subjects other than one­
'
sel£ This is true even of 'says that p : the allowable depar­
tures from direct quotation depend on what the ascriber
deems the quoted subject to have had in mind. Whether to
p a raphrase 'the commissioner' as 'that scoundrel', in an in­
direct quotation, is a question not of the com missioner's
character, but of the quoted speaker's view of it.
Empathy is why we ascribe a propositional attitude by a
content clause. We saw (§26) that content clauses were
more to the point than terms as grammatical objects even in
the case of per ception. The content clause purports to refle�t
the subj ect ' s state of mind rather than the state of things.
From the ascriber's point of vi ew it figures holophrastically;
its component terms do not necessarily refer, here, as he
means them to when he speaks for himself.
The obj ects of propositional attitudes-what are be­
lieved, regretted, etc.-have commonly been taken to be
propositions , or s entence meanings; but these have gone by
the board (§2 1 ) . 1 take them simply as sentences, namely the
content clauses themselves, thus treating 'that' as a quota­
tion mark initiating a name of w h at comes after it. Obvious
adjustments are to be understood in cases like 'He believes
he is Napoleon'; the belief is 'I am Napoleon. '
In thus ascribing pro p ositional attitudes to men �nd
beasts by quo tation I do not ascribe a command of the
quoted language, or of any. A cat can believe 'A mouse is
in there'. The language is that of the ascriber of the attitude.
tho u gh he p r ojects it empathetically to the creature in the
attitude. The cat is purportedly in a state of mind in which
the ascriber would say ' A mouse is in there'. The q uotational
 INTENSION
account reflects the empathy that invests the idioms of
propositional attitude from 'perceives that' onward.
If we were to activate this account by actually rewriting
the 'that' as quotation, rather than j ust noting it as tacit in­
tent, we would confound indirect quotation with direct.
But we could easily resolve the ambiguity by agreeing to
distinguish two verbs: 'say' for indirect quotation, 'utter'
for direct, and quotation marks for both.
We are familiar with the failure of substitutivity of iden­
tity in content clauses of propositional attitude. It fails be­
cause the person in the attitude may be unaware of the per­
tinent identity. This failure was a concern of Frege's.
Likewise we must beware of quantifying into such a clause,
for the values of the variable of our outlying quantifier are
the things of our real world, and might not fit the attitudi­
nist's ontology. Such is the referential opacity of the propo­
sitional attitudes. The quotational account nicely drama­
tizes it, for the quotation designates a mere string of
phonemes or signs, whose syntax and semantics, if any, are
a strictly internal affair.
Not that we must acquiesce in the quotation as a syntacti­
cally indigestible mass relative to the broader context. It is
digestible by spelling. We adopt names for all single signs,
finite in number, and then generate a name of any string of
signs by intercalating a sign of concatenation. Thus ' 11'av' is
pi-alpha-nu, and this is as straightforward in its syntax as
arithmetical addition or a polynomial.
Spelling dissolves the syntax and lexicon of the content
clause and blends it with that of the ascriber's language. So
long as we rest with the unanalyzed quotational form, on
the other hand, the inverted commas mark an opaque inter-
70 P U R S U I T OF T R U T H
face between two ontologies, two worlds: that of the man
in the attitude, however benighted, and that of our respon­
sible ascriber of the attitude.
The interface is sometimes breached. Like an actor step­
ping out of his part and speaking for himself, the ascriber is
heard to say of the real people of his world that
(I) There are some whom Ralph believes to be spies,
not just that
(1) Ralph believes '3x(x is a spy)' .
I f rendered quotationally, ( I ) goes incoherent.
(3) 3x(Ralph believes 'x is a spy') .
The quotation in (3) is just a name of a string of seven letters
and three spaces; its 'x' has nothing to do with the outlying
'3x'. (I) ascribed belief de rei quotation ascribes it de dicto.
Between ( I) and (1) we sense the vital difference between
spotting a suspect and merely believing, like all of us, that
there are spies. In affirming (I) we dissociate Ralph's suspi­
cions from the world as he conceives it and train them upon
denizens of our real world; we ride roughshod over failures
of identification on his part. Ralph suspects a man whom
he has seen lurking about a certain sensitive installation;
meanwhile he esteems Bernard). Ortcutt as a pillar ofso�­
ety. He is unaware that they are the same man. Does he then
both suspect Ortcutt and think him innocent? That would
be impossible or, at best, unfair to Ralph. 2
Propositional attitudes de re presuppose a relation of inten­
tion, between thoughts and things intended, for which I
2 Accommodation of ( I ) by singular descriptions was refuted by
Sleigh, q. v .
 INTENSION 71

conceive of no adequate guidelines . To garner empirical

content for ( I ) we wou l d have to in terro g ate Ralph and
compile some of his pertinent beliefs de dicto .
I conclude that the p r opositional attitudes de re resist an­
nexation to scientific l an gua ge , as propositional attitu des de
dicto do not. At best the ascri p tio ns de re are s ig nal s pointing
a direction in whi ch to look fo r inform ativ e as cri p ti o n s de
dicto .

29 . Anomalous monism
We reflected in §24 that a neurological rendering of ' Tom
perceives that it i s raining' , appl icabl e to all su ch occa si on s
merely on Tom's part, would already be pretty formidable
even if Tom's neural make-up were known in detail. We
reflec ted further that a neuro lo gical renderin g of ' perceiv es
that it is raining', applicable to all comers, would be out of
the question.
Yet each percepti on is a sin gle occurrence in a particul ar
brain, and is fully specifiabl e in neuro l o gi cal terms once the
details are known. We cannot say the same for a belief,
which can be publicly shared. but we can say somewhat the
same for the instance of the belief in a si n gle believer. The
period during which I go on believing that the earth rotates
is distin guished from my earlier stages by at least some
verbal dispositions, which must reside in some distinctive
quirks in my nervous system.
Per ceptions are neural reali t ies . and so are the individual
instances of beliefs and other propositional attitudes insofar
as these do not fade out into irreality altogether (§27) . Phys­

icalistic explanation of neural events and states goes blithely

forward with no intrusion of mental laws or intensional
7 1. PURSUIT OF TRUTH
concepts. What are irreducibly mental are ways of grouping
them: grouping a lot of respectably physical perceptions as
perceptions that p, and grouping a lot of respectably physi­
cal belief instances as the belief that p . I acquiesce in what
Davidson calls anomalous monism, also known as token
physicalism: there is no mental substance, but there are
irreducibly mental ways of grouping physical states and
events.
At first the problem of mind was ontological and linguis­
tic. With the passing of mind as substance, there remained
a twofold problem of mentalistic language: syntactic and
semantic. The distinctive syntactic trait of mentalistic �s­
course was the content clause, 'that p'. This obstructed ex­
tensiorudity: that is, the substitutivity of identity and more
generall y the interchangeability of all coextensive terms and
clauses salva veritate. It obstructed classical predicate logic
as a universal theoretical framework. Now this quarter of
the mind problem is in a fair way to dissolution. Quota­
tional treatment of propositional attitudes de dicto delivers
them to the extensional domain of predicate logic, thanks
to the reduction of quotation to spelling. Propositional atti­
tudes de re, on the other hand, we downgraded.
So we see the attitudes de dicto reconciled syntactically
with extensional logic. A single language, regimented in
predicate logic, can take them in stride along with natural,
science. The residual oddity of these mentalistic predicates
de dicto is purely semantic: they do not interlock produc­
tively with the self-sufficient concepts and causal laws of
natural science.
Still the mentalistic predicates. for all their vagueness
(§1. 7) . have long interacted with one another, engendering
age-old strategies for predicting and explaining human ac-
 INTENSION 73
tion. They complement natural science in their incommen­
surable way, and are indispensable both to the social sci­
ences and to our everyday dealings. Read Dennett and
Davidson.

30 . Modalities
The modalities of necessity and possibility are not overtly
mentalistic, but still they are intensional, in the sense of
resisting substitutivity of i de�tity Here again we have the
.

interplay between de dieto and de reo Thus 'nee (7 < the

,
number of the planets) is true de re, since nec (7 < 9). but
false de dicto.
In respect of utility there is less to be said for necessity
than for the propositional attitudes. The expression does
serve a purpose in daily discourse. but of a shallow sort. We
modify a sentence with the adverb 'neces saril y when it is a
'

sentence presumed accep table to our interlocutor and stated

only as a step toward the consideration of moot ones. Or we
write 'necessarily' to identify something that follows from
generalities already expounded. as over against new conjec­
tures or hy po t heses Such utility is local. transitory, and
.

unproblematic. like the utility of indexical expressions. The

sublimity of necessary truth turns thus not quite to dust, but
to pretty common clay.
The subjunctive or contrary-to-fact conditional has had
close associations with the ne ces s ity idiom. and a similar
account of it exp resses almost a commonsense view. The
conditio nal holds ifits consequent fo llows logically from its
antecedent in conjunction with b a ck g ro un d sentences that
one's interlocutor is prep ared to grant, or sentences that one
has already set down or im plicitl y assumed in one's exposi-
74 PURSUIT OF TRUTH
tory piece. The consequent of the conditional follows from
the antecedent ceteris paribus, and those supporting sen­
tences are the cetera paria .
Chatting of sublimity and common clay, I might pause
for a word on essence. Champions of modal logic mean
necessity to have an obj ective sense, as if to say metaphys­
ical necessity or physical necessity. But then it must make
sense to speak of a thing's essence , comprising those prop­
erties that it has necessarily. For, 'x necessarily has F' is
simply ' nec Fx ' . The essence has to be de re, inherent in the
thing independently of how referred to, since the thing can
figure simply as the value of a neutral variable as here.
In its everyday use as I described it, 'necessarily' is a
second-order annotation to the effect that its sentence is
deemed true by all concerned, at least for the sake and space
of the argument. A similar second-order role is cut out,
then, for ' possibly ' . Since it simply means 'not necessarily
not' , 'possibly' marks its sentence as one that the beliefs or
working assumptions of concerned parties do not exclude
as false. Thanks to our overwhelming ignorance, the realm
of possibility thus conceived is vaster far than that of neces­
sity. It is the domain of all our plans and conj ectures, all our
hopes and fears.

31 . A mentalistic heritage
Appreciation of one another's perceptions is fundamental,
we saw (§24) , to the handing down of language. The men­
talistic strain is thus archaic. We see it in animism, the
primitive ascription of minds to bodies on an excessive
scale. Perhaps there was a vestige of animism in Aristotle's
theory of the natural motion of substances: earth down-
 INTENSION 75

ward, fire upward, stars around and around. We see the

archaic dominance of mentalism in a preference for final
cause over efficient cause as a mode of explanation. It is
evident in the Middle Ages. The bestiaries accounted for the
supposed traits or practices of various animals as God's way
of setting moral examples for man to emulate. This predi­
lection for explanation by final cause is evident still today in
people who seek the meaning of life. They want to explain
life by finding its purpose.
Purpose is one of various mentalistic notions drawn from
introspection of one's mental life. Others are disposition
and capability. All three reflect one's sense of will, one's
sense of freedom to choose and act. The modality of possi­
bility is perhaps a depersonalized projection of the subjec­
tive sense of capability, a projection reminiscent of the
animists' projection of spirits into the rocks and trees.
Necessity, then, would be a projection of the subjective
sense of constraint, or abridgment of capability.
I suppose the idea even of efficient cause was mentalistic
in origin, being a projection of the subjective sense of effort.
Anyway it gained the upper hand over final cause with the
rise of physics in the Renaissance. Concomitantly, matter
gained the upper hand over mind. Mind was selflessly do­
ing itself in. Matter and efficient cause were a formidable
combination, vindicated in waves of successful prediction.
Final cause still had its explanatory duties too, not only in
relation to the mind of man but also in biology, where it
became an embarrassment, depriving biology of the aus­
terely scientific status that physics had come to enjoy. Dar­
win at length settled that matter, reducing final cause in
biology to efficient cause through his theory of natural
selection.
 PURSUIT OF TRUTH
Efficient cause figures conspicuously still in fairly austere
science. It is not clearly intensional, in the sense of resisting
substitutivity of identity , but it is like the intensional idioms
in lodging sentences indigestibly within sentences. We can­
'
not resolve 'p because q into predicates, quantifiers, and
truth functions, nor do we have as clear a notion of cause as
we could wish. Science at i ts mo re auste re bypasses the
notion and settles for concomitances.
Disposition is like cause in admitting substitutivity of
identity but resisting the predicate calculus. Also it is like
cause in its want of clarity . It seems to rest on an uncomfort­
able notion of potentiality. But these discomforts can be
quickly dissipated, for there is no need to invest the disposi-"
tional suffixes '-ble' and '-ile' wi th any theoretical content.
' Fragile ' and 'soluble' are physical p redi cat e s on a par with

others, and the dispositional form of the words is just a

laconic encoding of a relatively dependabl e test or symp­
tom. Breaking on impact and dissolving on immersion a re
symptomatic of fragility and solubility. See Roots of Refer­
ence, §§3- 4 .
 v

TRUTH

32 . Vehicles of truth
What are true or false, it will be wi del y agreed, are proposi­
tions . But it would not be so widely agreed were it not for
ambiguity of ' prop o si t i on ' . Some unders t an d the word a s
referring to sentences m eeting certain specifications.
Others understand it as referring rather to the meanings of
such sentences. What looked like wide agreement thus re­
solves into two schools of thought: fo r the first school the
vehicles of truth and falsity are the sentences , and for the
second t hey are the meanings of the sentences.
A weakness o f this second position is the tenuousness of
the notion o f sentence meani ng s . The tenuousness reaches
the breaking p oin t if one is pursuaded of my thesis of the
indeterminacy of translation (§§ I 8, 2 1 ) . Even apart from that
thesis, it seems perverse to bypass the v isi b le or audible
sentences and to center upon sentence meanings as truth
vehicles; for it is only by recurring to the sentence that we
can say which sen tence meaning we
have in mind.
There was indeed a motive for p res sing to the sentence
meanings . Many sentences in the same or different lan­
guages are deemed to be alike in meanin g . and distinctions
among them are indifferent to truth; so one narrowed the
field by ascribing truth rather to the meanings. This motive
 PURSUIT O F TRUTH
would be excellent if the notion of sentence meani n g were
not so elusive. But as matters stand we fare better by treat­
ing direct l y of sentences. These we can get our teeth into.
There was also a second motive, eq ual and opposite to the
first, for pressing on to the sentence m eani n g s ; n a m el y, that
one and the same sentence can be true o n some o ccasion s
and false on others. Thus 'The Pope wi ll visit Boston' was
true but turned false after his last vis i t . ' I ha ve a headache' is
true or false depending on who says it and when. Ambi­
g u ity or vagueness of terms, also, can cause the truth value
of a sentence to depend in part on the sp eaker ' s in ten tio n .
Propositions, thought of as sentence m ea nin g s , were th e
meanings exclusively of sentences of a fi rm e r sort, not sub­
ject to such vacillations; w ha t we may call eternal sen ten ces . t
My obvious response, then, i s that those eternal sentences
themselves can serve as the t ru t h vehicles. Just think of T ,
'yo u ' , 'he', 'she', 'here', and ' t here ' as supplanted by names
and addresses or other i d enti fyin g particulars as needed .
Think of tenses as dropped; we can use dates, the predicate
'earlier than', and the like as needed. Think of ambiguities
and vaguenesses as resolved by paraphrase-not abso­
lut ely , but enough to immobilize the truth value of the
parti cul ar sentence. The truth values need not be known,
b u t the y must be stable.
The attitude is the one that is familiar in the teaching of.
l og ic . When we take illustrative sentences from everyday
l In my logic books O f 1 940, 1 94 1 , and 1 9 50, and revised editions down
the years, my word for them was ' state m en t '; but I became chary of it
because of its cu stomary use rather for an act . ' Eternal sentence', along
w ith 'standing sentence ' (§4) , d a t e s from Word and ObjtCl. ' St an di n g
sentence' is more inclusive. 'The Timts h as come' is a standing sentence,
for it can com m and assent all day independently of interim stimulation ;
but it is not eternal .
 TRUTH 79

l anguage and paraphrase them into the notation of truth

functions and quantifiers , we think of the reference of
demonstratives and personal pronouns as fixed-albeit
tacitly - and we never dream of reading '3x' as 'there was'
or 'there will be something x'.
Declarative sentences thus refined-eternal sentences­
are what I shall regard as truth vehicles in ensuing pages, for
the most part. On the whole it is the convenient line for
theoretical purposes. We must recognize, though, that it
bypasses most of what counts in daily discourse as true or
false, since our utterances are not for the most part thus
refined. The truth vehicles directly related to behavior are
not sentences as repeatable linguistic forms, but rather the
individual acts of uttering them. These are for the most part
univocal in truth v alue without benefit of paraphrase. There
are just occasional failures, perhaps because some name
turns out to be empty or because some vague term turns out
to be indeterminate j ust where it matters for the utterance in
question. Such utterances may be dismissed as neither true
nor false .
So much by way of coming to terms with the realities of
verbal behavior. Let us now retu rn to the more conve­
niently manageable domain of eternal sentences, whose
truth or falsity, known or unknown, is unchanging .

JJ . Truth as disquotation
Such being what admit of truth, then, w he rein does their
t rut h consist? They qualify as true, one is told, by corre­
sponding to reality. But correspondence word by word will
not do; it in vites the idle cluttering of reality with a bizarre
host of fancied obj ects, just for the sake of correspondence.
80 PURSUIT OF TRUTH
A neater plan is to posit facts, as correspondents of true
sentences as wholes; but this still is a put-up job. Objects in
abundance, concrete and abstract, are indeed needed for an
account of the world; but facts contribute nothing beyond
their specious support of a correspondence theory .
Yet there i s some underlying validity t o the correspon­
dence theory of truth, as Tarski has taught us. Instead of
saying that
'Snow is white' is true if and only if it is a fact
that snow is white
we can sim ply delete 'it is a fact that' as vacuous, and there­
with facts themselves:
' Snow is white' is true if and only if snow is
white.
To ascribe truth to the sentence is to ascribe whiteness to
snow; such is the correspondence, in this example. Ascrip­
tion of truth just cancels the quotation marks. Truth is dis­
quotation.
So the truth predicate is superfluous when ascribed to a
given sentence; you could j ust utter the sentence. But it is
needed for sentences that are not given. Thus we may want
to say that everything someone said on some occasion was
true, or that all consequences of true theories are true. Such
contexts, when analyzed logically, exhibit the truth predi­
cate in application not to a quotation but to a pronoun, or
bound variable.
The truth predicate proves invaluable when we want to
generalize along a dimension that cannot be swept out by a
general term. The easy sort of generalization is illustrated
by generalization on the term 'Socrates' in ' Socrates is mor-
 TRUTH 81
tal'; the sentence generalizes to 'All men are mortal'. The
general term ' man' has served to sweep out the desired
dimension of generality. The harder sort of generalization is
illustrated by generalization on the clause 'time flies' in 'If
time flies then time flies' . We want to say that this com­
pound continues true when the clause is supplanted by any
other; and we can do no better than to say just that in so
many words, including the word 'true'. We say "All sen­
tences of the form 'If p then p ' are true. " We could not
generalize as in ' All men are mortal' , 1?ecause 'time flies' is
not, like 'Socrates', a name of one of a range of objects
(men) over which to generalize. We cleared this obstacle by
semantic ascent: by ascending to a level where there were
indeed objects over which to generalize, namely lingui�tic
objects, sentences.
Semantic ascent serves also outside of logic. When Ein­
stein propounded relativity, disrupting our basic concep­
tions of distance and time, it was hard to assess it without
leaning on our basic conceptions and thus begging the ques­
tion. But by semantic ascent one could compare the new
and old theories as symbolic structures, and so appreciate
that the new theory organized the pertinent data more sim­
ply than the old. Simplicity of symbolic structures can be
appreciated independently of those basic conceptions.
As already hinted by the correspondence theory, the
truth predicate is an intermediary between words and the
world. What is true is the sentence, but its truth consists in
the world's being as the sentence says. Hence the use of the
truth predicate in accommodating semantic ascent.
The disquotational account of truth does not define the
truth predicate-not in the strict sense of 'definition'; for
definition in the strict sense tells how to eliminate the
 PURSUIT OF TRUTH
defined expression from every desired context in favor of
previously established notation. But in a looser sense the
disquotational account does define truth. It tells us what it is
for any sentence to be true, and it tells us this in terms just as
clear to us as the sentence in question itself. We understand
what it is for the sentence 'Snow is white' to be true as
clearly as we understand what it is for snow to be white.
Evidently one who puzzles over the adj ective 'true' should
puzzle rather over the sentences to which he ascribes it.
'True' is transparent.
For eternal sentences the disquotational account of truth
is neat, we see, and simple. It is readily extended, more­
over, to the workaday world of individual utterances; thus
an utterance of ' I have a headache' is true if and only if the
utterer has a headache while uttering it.

34 . Paradox
It seems paradoxical that the truth predicate, for all its trans­
parency, should prove useful to the point of indispensabil­
ity. In the matter of paradox, moreover, this is scarcely the
beginning. Truth is notoriously enmeshed in paradox, to
the point of out-and-out antinomy.
An ancient form of the antinomy of truth is the Paradox
of the Liar: 'I am lying', or 'This sentence is not true'. A
looser and fancier version was the paradox of Epimenides
the Cretan, who said that all Cretans were liars . The under­
lying antinomy can be purified for logical purposes to read
thus:

(I ) 'yields a falsehood when appended to its own

quotation' yields a falsehood when appended to
its own quotation .
 TRUTH
Executing the instructions in (I) , we append the nine-word
expression to its quotation. The result is ( I ) itself. Thus ( I )
says that ( I ) itself i s a falsehood. It i s thus tantamount to ' I
a m lying', but more clean-cut. I t hinges only on the innocu­
ous operations of quoting and appending and the notion of
falsehood, which reduces to an innocent 'not' and true. The
truth predicate is clearly the trouble spot. The inevitable
conclusion is that the truth predicate, for all its transparency
and seeming triviality, is incoherent unless somehow re­
stricted.
For further explicitness a technical tum of phrase will be
convenient. The truth predicate will be said to disquote a
sentence S if the form
___ is true if and only if ___

comes out true when S is named in the first blank and

written in the second. Thus what the disquotational ac­
count of truth says is that the truth predicate dis quotes
every eternal sentence. But the lesson of the antinomy is
that if a language has at its disposal the innocent notations
for treating of quoting and appending, and also the nota­
tions of elementary logic, then it cannot contain also a truth
predicate that disquotes all its own eternal sentences-on
pain of inconsistency. Its truth predicate, or its best approx­
imation to one, must be incompletely disquotational.
Specifically, it must not dis quote all the sentences that con­
tain it. That was the trouble with ( I ) . And of course it must
not dis quote all the sentences containing terms by which
that predicate could be paraphrased. This, apart from its
special orientation to quoting and appending, is substan­
tially what has come to be known as Tarski's Theorem. He
has proved harder things.
 PURSUIT OF TRUTH
The truth predicate loses little in general utility thereby,
for it can still disquote all the eternal sentences that do not
themselves contain it or other expressions to the same ef­
fect. And even these excluded applications can be accom­
modated by a hierarchy of truth predicates . The hierarchy
begins with a predicate 'trueo', which disquotes all sen­
tences that contain no truth predicate or equivalent devices.
A predicate 'true t next, disquotes all sentences that contain
',

no truth predicate or equi valent devices beyond 'trueo' . And

so on up. It is a hierarchy of progressively more nearly
perfect truth predicates. The plan dates back in a way to the
early phase of Russell's theory of types ( 1908) , by which he
meant to obstruct the Paradox of the Liar among others.

35 . Tarski 's construction

We saw that dis quotation is loosely defmitive of truth We .

may now be thankful for the looseness seeing as we do that

,

definability of truth for a language within the language

would be an embarrassment. And thus it was that Tarski
undertook the perilous adventure of defining it for the lan­
guage within the language as nearly as possible, if only to
,

see what minimum obstacle saved the situation. This was

not his order of presentation, but it comes out the same.
The language chosen for the construction contains the
logical notations for quantification and the truth functions
and the set-theoretic notation 'x E y' for membership. 2 It
contains also a finite lexicon, as large as you please, of predi­
cates for natural science and daily life. Finally it contains the

2 Readers expecting a contrast between obj ect langU4lge and metalan­

guage should bear in mind that I am already addressing the aforesaid
perilous adventure.
 TRUTH 8S

means, in effect, of quoting and appending , as in ( I ) ; that is,

it can specify each of its single signs and it can express the
concatenation of e x p ressio ns .
Truth pert ain s t o closed sentences, that i s , sentences
without free variables. Its analogue for open sentences is the
two-place predicate of satisfoction . An assignment of objects
to variables slltisfits a sentence if the sentence is true for those
values of its free variables .
What sort of object is an assigrtlfltnt of objects to variables?
It is simply a function, or one-many rtlation , relating one
and only one object to each variable-that is, to each letter,
'x', ' y , ' z ' , ' w " , etc. A relation, in turn, is a set, or class,
'
'w',

or oNtrtd pairs. Ways are well known of defining the nota­

tion '(x,y)' of ordered pairs contextually by means of epsi­
lon and the logical particles .
Once satisfaction is defined, truth comes easily; for a
closed sentence, having no free variables, is vacuously
satisfied by all assignments or none according as it is true or
false. We can simply define
'
(2.) y is true' as 'Vx(x is assignment . ..... . x satisfies y) ' .

So Tarski's bigjob i s to define satisfaction . First he defines it

for atomic sentences, each of which consists ofjust a predi­
cate adjoined to one or more variables . For instance an as­
'
signment satisfies the atomic sentence 'x E y if and only if
what is assigned to the letter 'x' is a member of what is
' '
assigned to the letter y . C orres pon dingly for each of the
other predicates in the lexicon. An assignment satisfies an
alternation of sentences, next, if and only if it satisfies one
or both of them; it satisfies their conjunction if and only if it
satisfies both; and it satisfies a negation if and only if it does
not satisfy the sentence that is negated . Finally, an assign-
86 PURSUIT O F TRUTH
ment satisfies an existential quantification '3x ( . . . x . . . ) ' if
and only if some assignment, matching that one except
perhaps for what it assigns to ' x ' , satisfies . . . x . . . . .
'

Such is Tarski's recursive or inductive definition of satis­

faction. It explains satisfaction of atomic sentences out­
right, and it explains satisfaction of sentences of each higher
grade or complexity in terms of satisfaction of their compo­
nents. Universal quantification is passed over because it is
expressible in terms of existential quantification and nega­
tion in familiar fashion.

36. Paradox skirted

Clearly all the clauses of this inductive definition can be
formulated within the formal language itself, except for the
word 'satisfies' that is being defined. Thus we have appar­
andy defmed satisfaction for the language within the lan­
guage. Invoking (2) , then, we have done the same for truth.
This was supposed to spell contradiction.
We could even get contradiction directly from satisfac­
tion, without the detour through (2) , 'truth', and ( I) . We
have merely to ask whether assignment of the sentence 'not
(x satisfies x) ' to the variable 'x' satisfies the sentence 'not (x
satisfies x) ' itself Such is Grelling's so-called Heterological
Paradox. 3
What saves the situation is that the definition of satisfac­
tion is inductive rather than direct. The inductive definition
explains satisfaction of each specific sentence, but it does
'
not provide a translation of 'x satisfies y with variable 'y ' .

l See my Ways of Paradox , pp. 4-6.

 TRUTH
Consequently it does not translate the 'not (x satisfies x)' of
Grelling's paradox, and does not suppon the truth defi­
nition (2) for variable 'y'; it just explains truth of each spe­
cific closed sentence. It leaves the truth predicate in the same
state in which the disquotational account left it; namely,
fully explained in application to each specific sentence ofthe
given language but not in application to a variable.
It was a near miss, and I turn now to a nearer one. Treat­
ing relations again as classes of ordered pairs, we can write
'(x,y) E Z ' to mean that x bears the relation z to y. Now
imagine the above inductive definition of satisfaction writ­
ten out in our formal language, with the variable 'z' always
in place of 'satisfies' and so '(x, y) E z' in place of ' x satisfies
y' . Let the whole inductive definition, thus edited, be ab­
breviated as '�z'. It fixes z as the satisfaction relation. Evi­
dently we arrive thus at a direct definition:
(3) 3z(�z . (x , y) E z)
of 'x satisfies y ' strictly within the formal language itself.
Doesn't this spell contradiction?
No. The catch this time is that there might not be any
relation z such that �z. Indeed there better not be, on pain,
we see, of contradiction. The two-place predicate 'satisfies'
remains well defined in its inductive way, but a grasp of the
predicate and how to use it carries no assurance of the exis­
tence of a corresponding a bstract object, a corresponding
set of ordered pairs. And, failing such a pair set, (3) fails to
translate 'x satisfies y'. Though the satisfaction predicate is
well explained even within the formal language by the re­
cursion, it does not get reduced to the prior notation of that
language. Satisfaction, and truth along with it, retain the
88 P U R S U I T OF TR U T H

status that truth already enjoyed under the disquotation

account: clear intelligibility without full elimin a bility . 4

37. lnterloclud hierarchies

The inductive definition fully explains what it is for an
assignment to satisfy a sentence. That there is such a satis­
faction relation, then, or pair set, is sheer common sense .
The paradoxes of set theory , however-Russell's, Burali­
Foni's, Cantor's-have overruled the commonsense no­
tion that clear membership conditions · assure the existence
of a class, a set. All of those were paradoxes ultimately of the
membership predicate ' E ' ; what is striking about the present
case is jus t that we find set theory responding also to para­
doxes of truth and satisfaction.
Some mathematicians supplement the universe of
classes, or sets, with a layer of classes that are not eligible for
membership in any further classes. The hitheno inter­
changeable terms 'set' and 'class' are then used to mark the
distinction: sets are classes that '''� members of funher
classes . The added classes, members of nothing, came to be
known lamely as classes proptr. or "proper classes. " I have
called them ulti""'t� classes. Membership conditions that
failed to determine sets can be reinstated without fear of
contradiction as determining ultimate classes. Luxuriously,
thenceforward, every membership condition on sets deter­
mines a class; maybe it will be a set, maybe an ultimate class.
Parsons has shown (pp. 2 U-2I 4) that the satisfaction rela-

4 The foregoing aaulysis is adapted from my Philosophy oj Logic, pp.

] s-46. A somewhat different analysis, in my 19SZ paper "On an Appli­
cation ofTarski's Theory of Truth, " is called for when the set theory is of
the kind that admits both sets and ultimate classes.
 TRUTH
tion that had failed to exist as a set of pairs now comes to
exist as an ultimate class of pairs . A direct definition of
satisfaction, and so of truth, is thus achieved after all. But it
is achieved only for sentences of the old theory, unsupple­
men ted with ultimate classes.
Thus let us suppose that along with adding the ultimate
classes we introduced a new style of variables, to range over
classes generally-the old variables being limited to sets.
Then the point is that truth, as newly and directly defined,
is assured of disquotationality over all the old sentences,
but will fail of it for some sentences containing the new
variables.
But we can repeat the expedient, adding layer on layer of
new classes without end . ' Ultimate' ceases to be the word
now; I must submit to 'proper class' . The classes at each
level admit members freely from all and only lower levels.
For the ith level, for each i, the variables 'x;', 'y;" etc. range
over that level and lower ones; thus ' xo' , 'Yo' , etc. range only
over sets. Predicates 'trueo'. 'true. '. and so on are then all
forthcoming by direct definition. For each i, ' true;' is de­
pendably disquotational in application to sentences contain­
ing no bound variables beyond level i. We get a self­
contained language with a hierarchy of better and better
truth predicates but no best. Trutho is already good enough
for most purposes, including classical m .. thematics.
In his early version of his theory of types, mentioned at
the end of § 3 4. Russell sought to block the paradoes both of
truth and of membership by decreeing a single complex
hierarchy of predicates. The scheme was vague and cum­
bersome. He and others subsequently sharpened and
simplified it for purposes of set theory by dispensing with
the truth aspect as extraneous to set theory. And now we see
 PURSUIT OF TRUTH
a new interlocking o f the class hierarchy with the truth
hierarchy, along clean-cut lines and for clear but subtle rea­
sons that could not be foreseen in Russell's dOly.
I have pictured the hierarchy of class variables and the
hierarchy of truth predicates as embraced within a single
inclusive language. This is how I like it. But alternatively
we can pictUre a hierarchy of languages, each with single­
sorted variables and a unique truth predicate for the next
lower language. This approach has an important mathe­
matical application in establishing relative strengths of for­
mal systems. To prove that one system is stronger than
another, reinterpret its predicates in such a way as to be able
to define, within it, the truth predicate of the other system.

J8. Excluded middle

Let us look into some seemingly deviant notions about

truth. One such, traceable to Aristotle, is that a prediction is
neither true nor false until events have occurred that caus­
ally determine it. Theologians have favored the doctrine. If
contingent predictions were true now, they reason, the
events would be determined now by God's knowledge, and
hence would not be contingent. The consequent determin­
ism, it is felt, would leave no place for man's moral re­
sponsibility.
This doctrine, for all its bizarreness, is no repudiation of
the disquotational account of truth. If it is not yet true that
there will be a sea-fight tomorrow-to take Aristotle's ex­
ample-then it is a mistake to say now that there will be a
sea-fight tomorrow; for as of now the contingent sentence
is neither true nor false. The logic, granted, is deviant: the
law of excluded middle is suspended pending causal deter-
 TRUTH 91
mination. But the disquotational character o f truth re­
mains.
Bizarreness remains too. There is the abandonment of the
law of excluded middle; also the drastic narrowing of the
range of sentences with fixed truth values. Happily, how­
ever, the theological argument underlying this desperate
move is inconclusive on two points. One is the assumption
of an omniscient God. The other is the notion that universal
determinism precludes freedom of action. We are free and
responsible, it can be argued, in that we act as we choose to;
whether our choices are determined by prior causes is be­
side the point.
Some other apparent challenges to the law of excluded
middle are, in part, not what they seem. Let it be clear, to
begin with, that ignorance of the truth or falsity of a sen­
tence is par for the course, and quite in keeping with its
being true or false. Further, it commonly happens that a
sentence can be rendered eternal in divergent ways, reflect­
ing a speaker's intentions in different situations. Here it is
rather the respective utterances that are true or false, to­
gether with their full unambiguous elaborations if we care
to elaborate them. The original ambiguous sentence is in­
deed then neither true nor false, but this need not be seen as a
breach of the law of excluded middle; it is better seen as an
incompleteness that has still to be filled out in one Qr an­
other way. This line was not available in the case of the
theologians' strictures on contingent predictions, because
those sentences were meant still to become true or false,
without supplementation, once they stopped being con­
tingent.
There is another and stronger case that likewise threatens
the law of excluded middle. It is where a purported name or
91 PURSUIT OF TR UTH
singluar description fails to designate anything. When a
sentence contains such a term, one pos s i bl e line is to drop
the sentence from consideration; treat it as mea ning less .

This line is awkward when w e regi ment our sentences for

logical p urposes , for the existence of the object may be an
open questio n as in the case of C a melo t or Prester John or
-

the outermost satellite of Pluto. It is q ui t e in order for the

truth value of a sentence to remain an open ques t io n but it
,

is inconvenient to leave the very meaningfulness of a sen­

ten ce foreve r unset d ed.

One might accordingly re l in quish the law of excluded

middle and opt rather for a three-valued logic, recognizing
a limbo between truth and fals i t y as a thi rd truth value.
What then comes to hi ng e on existence of Camelot, or
whatever, is truth value r a th er than meaningfulness. and
that is as it should be. But a price is paid in the cumber­
someness of three-valued logic. A lon gsi de 'not', w hic h
sends truths into falsehoods , falsehoods i n t o truths, and
now limbo into limbo, there would be a truth function that
sends t ruths into limbo, l i m bo into falsehoods, and false­
hood s into truths; ako three more such one-place truth
functions, playing out the combinations-as contrasted
with a si n g le one , neg atio n , in two-valued lo gi c When we
.

move out to two-place truth fu n ctions (co nj u n ction alter­

,

nation, and their derivatives) , p roli fe r a tion runs amok. It

can still be handled , but there is a n evident pre m i um on our
si m ple streamlined t w o-valu ed logic.
We can adhere to the l atter in the face anyway of the
,

threat of empty s i n g ula r terms, by simply dispen s in g with

s in g ul a r terms as in § I O. 'Camelot is fair' becomes ' 3x(x is
Camelot and x is fair) ' . It does not go into li mbo ; it simply
goes false if it is false that 3x(x is Ca m elo t) . The pre di cate
 TRUTH 93
'is Camelot' is seen on a par with 'is fair' , as a predica te irre­
ducibly .

39 . Truth versus warranted belief

Pilate was p robably not the first to ask what truth is, and he
was by no means the last. Those who ask it seek something
deeper than disq uotation, which was the valid residue of the
correspondence theory of truth (§3 3 ) . Yet there is surely no
im pu gning the disquotation account; no disputi ng that
' Snow is white' is true if and only if snow is white. More­
over, it is a full account : it explicates clearly the truth or
falsity of every clear sentence. It i s even a more than full
account: it imposes a requirement on the truth predicate
that is too strong for any predicate within the language
concerned-on pain of contr adi ction (§3 4.) .
There are recurrent referen ces to a coherence theory of
truth, or a pragmatist theo ry of truth. The quest ion that
motivates this quest beyond disquotation can perhaps be
phrased thus: if to call a sentence true is simply to affirm it,
then how can we tell whether to affirm it?
The lazy answer is "That all depends on what the sen­
tence is. In the case of ' Snow is white' you just look at snow
and check the color. The more sympathetic answer is a
"

general analysis of the grounds of warranted belief, hence

scientific method-perhaps along the lines of §§2-7.
The moderately holistic considerations there set forth are
uncongenial to a line currently urged by Michael Dummett,
in which he contests the law of excluded middle on epis­
t emologi c al grounds. The attack was mounted in mathe­
matics by L. E. J. Brouwer ea rly in this century, and Dum­
mett adopts the attitude toward science in general . His
94 PURSUIT OF TRUTH
rou g h idea is to reckon a sen ten ce of natural science neither
true nor false if n o pro ced u re is known for making a strong
emp irical case for its truth or falsity.
Holistic considerations make it doubtful what sentences
should be retained, then, as eligib le for t ru th or fal s it y .

Clear candidates for retention are the observation categ o ri­

cals. Other sentences share empirical conten t i n va ry in g
deg rees by i mp lyin g ob serv ation categ ori cals j o in t l y It
.

seems vain to seek an invidious distinction between sen­

tences eligible for truth or fals it y and sentences in limbo,
unless we either draw that boundary at the observation
ca te goric a ls themselves or else draw it at the far extreme to

exclude just those sentences that never i m bi be any empirical

content b y patticipating in the joint implying of any obser­
vation categoricals.
Truth is one thing, warranted belief another. We can gain
cla rity and enjoy the sweet simplicity of two-valued logic
by heeding the dist in c tio n.

40 . Truth in mathematics
What now of those parts of mathematics that share no em­
pirical meaning, because of never getting a pplied in natural
science? What of the hi g her reaches of set th eo ry? We see
them as mean in gful because they are couched in the same
grammar and vocabulary that generate the applied parts of
mathematics . We are just sparing ourselves the unnatural
gerrymandering of grammar that would be needed to ex­
clude them. On our two-valued approach they then qualify
as true or false, albeit inscrutably.
They are not wholly inscrutable. The main axioms of set
theory are generalities operative already in the applicable
 TRUTH 95
part of the domain. Further sentences such as the con­
tinuum hypothesis and the axiom of choice, which are inde­
pendent of those axioms, can still be submitted to the con­
siderations of simplicity, economy, and naturalness that
contribute to the molding of scientific theories generall y .
Such considerations support COders axiom of construcribil­
ity, 'V = L'. 5 It inactivates the more gratuitous flights of
higher set theory, and incidentally it implies the axiom of
choice and the continuum hypothesis. More sweeping
economies have been envisioned by Hermann Weyl, Paul
Lorenzen, Errett Bishop, and currently Hao Wang and Sol­
omon Feferman, who would establish that all the mathe­
matical needs of science can be supplied on the meager basis
of what has come to be known as predicative set theory. 6
Such gains are of a piece with the simplifications and econo­
mies that are hailed as progress within natural science itself.
It is a matter of tightening and streamlining our global sys­
tem of the world.

41. Equivalent theories

I defined empirical content in § 7 only for testable theories,
and I went on to point out that much solid experimental
science fails of testability in the defined sense. This can
happen, we saw, because of vague and uncalibrated proba­
bilities in the backlog of theory. No doubt it happens also
in more complex ways, not clearly understood. I have no
definition of empirical content to offer for such theories,
but it still seems to make reasonable intuitive sense to speak
of empirical equivalence among them, since experimenta-
s See my St' Thtory and Irs Log ic, 2.d ed. , pp. 2.34-2.38.
6 See Qu iddi'its, pp. 34-36.
 PURS UIT OF TRUTH
tion i s still brought to bear. The idea is that whatever obser­
vation would be counted for or against the one theory
counts equally for or against the other. What I shall have to
say about empirically equivalent theories applies indiffer­
ently to testable ones and to theories that are empirically
equivalent in this ill-defined way.
Theories can differ utterly in their obj ects, over which
their variables of quantification range, and still be empiri­
cally equivalent, as proxy functions show (§ 1 2) . We hardly
seem warranted in calling them two theories; they are two
ways of expressing one and the same theory. It is inter­
esting, then, that a theory can thus vary its ontology.
Effort and paper have been wasted, by me among others,
over what to count as sameness of theory and what to count
as mere equivalence. It is a question of words; we can stop
speaking of theories and just speak of theory formulations. I
shall still write simply 'theory', but you may understand it
as 'theory formulation' if you will.
Theories (theory formulations) can be logically incom­
patible and still be empirically equivalent. A familiar ex­
ample is Riemannian and Euclidean geometry as applied to
the surface of a sphere. Riemannian geometry says that
straight lines always meet. Euclidean geometry says that
some do and some do not, and in particular that there are
none on a sphere. The conflict is resolved by reinterpreting
'straight line' in the Riemannian glossary as 'great circle' .
The next example, due t o Poincare (ch. 4), i s less trivial.
We have on the one hand our commonsense conception of
infinite space and rigid bodies that move freely without
shrinking or stretching, and on the other hand the concep­
tion of a finite spherical space in which those bodies shrink
uniformly as they move away from center. Both concep-
 TRUTH 97

tions can be reconciled with all possible observations; they

are empirically equivalent. Yet they differ, this time, more
deeply than in the mere choice of words . The theory with
the finite space makes crucial use of a theoretical term that
admits of no counterpart in the theory with the infinite
space-namely, 'center of space'.
Imagine now two theories, ours and another, such that
we are persuaded of their empirical equivalence but we see
no way of systematically converting one into the other by
reinterpretation sentence by sentence, as we did in the ex­
ample of the proxy function and that of the sphere. There
are three cases to consider.
Case I : The other theory is logically compatible with our
own and is expressed directly in our own terms . It differs
from ours in that it implies some theoretical sentences that
ours leaves unsettled, and vice versa. Yet the theories are
empirically equivalent. This case presents no problem. We
would simply accept the other theory and incorporate it
into our own as an enrichment, answering many theoretical
questions that ours left open.
Cast 2: Again the other theory is logically compatible
with ours, but, like Poincare's example, it hinges on some
theoretical terms not reducible to ours.
Case 3: The two theories are logically incompatible.
Donald Davidson showed me that this case can be reduced
to Case 2 by the following maneuver. Take any sentence 'S
that the one theory implies and the other denies. Since the
theories are empirically equivalent, S must hinge on some
theoretical term that is not firmly pinned down to observ­
able criteria. We may then exploit its empirical slack by
treating that term as two terms, distinctively spelled in the
two theories. S thus gives way to two mutually indepen-
 PURSUIT OF TRUTH
dent sentences S and S'. Continuing thus, we can make the
two theories logically compatible.

42 . Irresoluble rivalry
So we may limit our attention to Case 2. Let us limit it
further to global systems of the world, so that there is no
question of fitting the rival theories into a broader context.
So we are imagining a global system empirically equivalent
to our own and logically compatible with ours but hinging
on alien terms. It may seem that as staunch empiricists we
should reckon both theories as true. Still, this line is unat­
tractive if the other theory is less simple and natural than
ours; and indeed there is no limit to how grotesquely cum­
bersome a theory might be and still be empirically equiva­
lent to an elegant one. We do better, in such a case, to take
advantage of the presence of irreducibly alien terms. We can
simply bar them from our language as meaningless. After
all, they are not adding to what our own theory can predict,
any more than 'phlogiston' or 'entelechy' does, or indeed
'fate' , 'grace', ' ni rvana ' , ' mana'. We thus consign all con­
texts of the alien terms to the limbo of nonsentences.
We have here an encroachment of coherence considera­
tions upon standards of truth. Simplicity and naturalness
are making the difference between truth and meaningless­
ness.
We might still choose to enrich our original theory with
any novel findings of the other theory that do not use the
alien terms. It would be a matter of welcoming information
from a presumed dependable outside source, much as sup­
plementary truths of number theory are got by excursions
 TRUTH 99

through analysis, or as the four-color-map question was

setded by an elaborate computer program.
But now suppose rather that the rival theory is as neat and
natural as our own. Our empiricist scruples reawaken.
Should we incorporate that theory into our own, as in Case
I ? No, this would ill accord with the scientists' quest for
simplicity and economy; for the irreducible new terms im­
ported with the annexed theory have added no new cov­
erage of observables. The two theories were already em­
pirically equivalent to each other, and hence to their
conjunction. The two theories were streamlined and neck­
and-neck, but the tandem theory is loaded beyond neces­
sity: loaded with all the sentences containing the new terms.
One possible attitude to adopt toward the two theories is
a sectarian one, as I have called it: 7 treat the rival theory as in
the preceding case, by rejecting all the contexts of its alien
terms. We can no longer excuse this unequal treatment of
the two theories on the ground that our own is more ele­
gant, but still we can plead that we have no higher access to
truth than our evolving theory, however fallible. Dagfmn
Fellesdal and Roger Gibson abetted me in this sectarian
attitude. The opposing attitude is the ecum enical one, which
would count both theories true. Its appeal is empiricism:
reluctance to discriminate invidiously between empirically
equivalent and equally economical theories. The tandem
theory, which we found prohibitively uneconomical, was
one ecumenical line. But a different ecumenical line has
been urged by Donald Davidson: that we account both
theories separately true, the truth predicate being under-

' ''Reply to Gibson. "

100 PURSUIT OF TRUTH
stood now as disquotation in an inclusive and thcory­
neutral language in which both theories are couched. In
recent years I have vacillated among these alternatives,
which are now down to two: the sectarian line and the
ecumenical line a la Davidson.
The latter alternative raises questions regarding the inclu­
sive language. It would include all terms of both of the rival
systems of the world, and its variables would range over
both ontologies. Distinctive predicates would serve to de­
limit values of the variables to the one ontology or the other
as needed. How much more widely should the variables

range? And what of truth: will there be a h iera rch y of truth

predicates and a matching hierarchy of styles of variables, as
at the end of §3 7? We must call a halt. We sought only an
inclusive language, not a third theory.
What is to be gained is not evident, apart from the satis­
faction of conferring the cachet of truth evenhandedly. The
sectarian is no less capable than the ecumenist of appreciat­
ing the equal evidential claims of the two rival theories of
the world. He can still be evenhanded with the cachet of
warrantedness, if not of truth. Moreover he is as free as the
ecumenist to oscillate between the two theories for the sake
of added perspective from which to triangulate on prob­
lems. In his sectarian way he does deem the one theory true
and the alien terms of the other theory meaningless, but
only so long as he is entertaining the one theory rather than
the other. He can readily shift the shoe to the other foot.
The fantasy of irresolubly rival systems of the world is a
thought experiment out beyond where linguistic usage has
been crystallized by use. No wonder the cosmic question
whether to call two such world systems true should simmer
 TRUTH 101
down, batheti ca l l y , to a question of words. Hence also,
meanw hile , my vacill a tion.
Fare these conventions as they may, the rival theories
describe one and the same world. Limited to our h uman
terms and devices, we grasp the world variously. I think of
the disparate ways of getting at the dia meter of an impene­
trabl e sphere: we may pi n ion the s phere in calipers o r we
ma y girdle it with a tape measure and divide by pi, but there
is no getting ins ide.

43 . Two indeterminacies
There is an evident parallel between the empirical under­
deter mina tion of global science and the indeterminacy of
translation. In both cases the total ity of p o s s ible evidence is
insufficient to cli n ch the system uniquely. But the indeter­
minacy of translation is additional to the other. If we settle
upon one of the empirically equivalent systems of the
world, however arbitrarily, we still h a v e within it the inde­
terminacy of translation.
Another distinctive point about the indeterminacy of
tran slation is that it cle arly has nothing to do with inaccessi­
ble facts and human limitations. Dispositions to ob s e rvable
behavior are all there is for semantics to be rig h t or wrong
about (§ I 4) . In the case of sy s tems of the world, on the other
hand, one is prepared to belie ve that reality exceeds the
scope of the human app aratus in unspecifiable ways.
Let us now look more closely to parallels. On the one
hand we have the two incompatible but equally faithful
systems of translation; each propounds some translations
that the other rej e cts . On the other hand we have two in-
102 PURSUIT OF TRUTH
compatible but empirically equivalent systems of the
world. We noted in § 1 8 that we can reconcile the two sys­
tems of translation by recognizing them as defining differ­
ent relations, translation! and translation2' We noted in §4 I
that we can reconcile the two systems of the world by simi­
larly splitting one or more theoretical terms.
What the indeterminacy of translation shows is that the
notion of propositions as sentence meanings is untenable.
What the empirical under-determination of global science
shows is that there are various defensible ways of conceiv­
ing the world.
RE F E R E N C E S

CREDITS

INDEX
 REFERENCES

Barrett, R. B . , and R. F. Gibson, cds. Persptctivts Oil Quillt. Oxford:

Blackwell. In press.
Bergstrom, Lars. " Quine on Underdetermination. " In Barrett and
Gibson.
Carnap, Rudolf. Logischt SYlltax du Spracht. Vienna, 1 934.
Chisholm, Roderick. Puctivillg: A Philosophical Study . Ithaca: Cor­
nell University Press, 1 957.
Davidson, Donald. Essays o n Action and Evtnts. Oxford: Clarendon,
1 980.
Dennett, Daniel. Tht Intmlional Stanct. Cambridge: MIT Press,
1 987.
Duhem, Pierre. La thlorit physique: son objet tt sa structUrt. Paris, 1 906.
Dummett, Michael . Truth and Othu Enigmas . Cambridge: Harvard
University Press, 1 978.
Firth, Roderick. "Reply to Sellars. " Monist 64 ( 1 98 1 ), pp . 9 1 - 1 0 1 .
Griin baum, Adolf. "The Falsifiability of Theories. " Synthtse 1 4
( 1 96:1), pp. 1 7- 3 4 .
Kirk , Robert. "Quine's Indeterminacy Thesis. " Mind 78 ( 1 969) ,
pp . 607-608 .
Levy , Edwin . "Competing Radical Translations. " Boston Studits in
Philosophy ofScimct 8 ( 1 97 1 ) , pp. 590-605 .
Lewis, C. I. An Analysis of Knowltdgt and ValU4tion . La Salle: Open
Coun, 1 946.
Massey , G. J. "Indeterminacy, Inscrutability, and Ontological Rel­
ativity . " Amuican Philosophical Quarterly , Monograph u ( 1 978),

pp· 43 - 5 5 ·
Parsons. Charles. Mathematics and Philosophy . Ithaca: Cornell Uni­
versity Press, 1 98 3 .
1 06 REFERENCES
Poincare. Henri. Scimct and Hypothesis. New York. 1 905.
Popper. Sir Karl. The Logic oj Scientific Discover. New York: Basic
Books. 1 959.
Premack. David. eavagai. Cambridge: MIT Press. 1 986.
Pumam. Hilary . "Mathematics without Foundations . " Journal oj
P#ailosophy 64 ( 1 96 7) . pp . 5 -22 .
Quine. W. V. Mathemaliad Logic. New York. 1 940. Corrected ed.
(thanks to Wang) . Cambridge: Harvard University Press. 1 95 1 .
--- Elemmt4ry Logic. Bos to n. 1 94 1 . Rev. ed . • Harvard Univer­
sity Press. 1 96 5 .
-- Methods oJLogic. New York . 1 950. 3 d ed 1 972. 4th, Harvard
. •

University Press, 198 2 .

.
--- "On an Application ofTarski's Theory of Truth . . Proceedings
oJlhe National Academy oJ Sciences 3 8 ( 1 951). pp . 43Q-43 J . Rpt. in
Selected Logic Papm . New York: Random House, 1 966.
-- From a Logical Point 0J Vit1ll . Harvard University Press, 1 95 3 .
-- Word and Object . Cam bridge : MIT Press, 1960.
--- Set Theory a nd Its Logic. Harvard University Press, 1 96 3 . Rev.
cd 1969.
.•

--- TItt Ways oj Paradox and Other Essays. New York, 1 966.
Enlarged ed Harvard U niv ers i ty Press, 1 976.
.•

--- Ontological Relativity and Other Essays . New York : Columbia

University Press. 1969.
-- Philosophy oj Logic. Englewood Cliffs. N.J. , 1 970; Harvard
University P ress , 1 986.
-- Roots oJRefrrmce, La S alle : Open Court, 1 974.
-- "The Nature of Natural Knowledge. " In J . Guttenplan. ed.
Mind and Language (Oxford : Clarendon Press, ( 975). pp. 67-8 1 .
-- Theories and Things . Harvard U niversi ty Press. 1 98 1 .
..
--- "Events and Reification . In E . Lepore and B . McLaughlin.
cds. , Action and Evmls (Oxford: Blackwell, 1 98 5). pp. 1 61 - 1 7 1 .
-- "Reply t o Roger F . Gibson, Jr. " In L . E . Hahn and P . A .
Schilpp. cds. , TItt Philosophy oJ W. V. QuiM (laSalle : Open Court.
1 986), pp. 1 5 5 - 1 5 7.
-- Quiddities. Harvard University Press, 1 987.
Quine. W. V . , and J . S. Ullian. The Web oJBelieJ New York: Ran­
dom House. 1 970. Rev . ed. , 1 978.
 REFERENCES 1 07

Russen, Bertrand. "00 Denoting. " Mirul 1 4 ( 1 9OS), p p . 479-493 .

-- "Mathematical Logic as Based on the Theory of Types. "
AmtrictUl jou"",I ofMathmultics 3 0 ( 1 908), pp. 1U-161.
Sleigh, R. C. "On a Proposed System of Episternic Logic, " NoNs 1
( 1 96 8), pp . 39 1 - 398.
Tanki, Alfred. "The Concept of Truth in Formalized Languages. "
In his Logic, Smulrttics, Mttamathtmatics (Oxford: Clarendon Press,
1 9 S6) , pp. I Sl-178. Translated from the German of 1 936.
Ullian, J. S. Stt Quine and Ullian.
Vosida, Natuhiko. "Scientific Laws and Tools for Taxonomy. An­ II

Mis oftht japantst Association for tht Philosophy of Scittllt 6 ( 1 984),

pp . 107-1 1 8.
 CREDITS

Five and a half pages are fro m theJou",,' ojPlti'osophy, 1 98 ]- 1 987. as

follows. Part of pp. 499-500 of " Ontology and Ideology Revisited"
(vol. 80) appears in § 1 0; part of p. 6 of "States of Mind" (vol. b)
appears in §�; and parts of pp. 5-9 of "Indeterminacy of Transla­
tion Again" (vol. 84) appear in S§ 1 4. 1 7. 1 8. and 2. ] . Two pages
from pp. 1 39- 1 4 1 of "Cognitive Meaning" in the Monist, v ol 62.. .

1 979. appear in §2.... Most of page 16.t of my reply to daIla Chiara

and Toraldo di Francia in A",,'isis filos6.fico, vol. 2. (Buenos Aires,
1 982.) appears in S I ] . Most of p. 5 1 of "Sensory Support of Sci­
ence, " in Discursos de i""utidurQ "Doaor Ho"oris CQNSQ ", Granada,

Spain, 1 986, appears in S I I . I am grateful to all the copyright hold­

ers for their permission.
 INDEX

Abstract objects, 30, 34 Conditionals, 73

Analytical hypotheses. 43 , 4 5 Conjunction, .. 1 1
Analyticity. 1 6. 5 5 - 56 Content clause, 61-69, 72
Animism. 74-75 Correspondence theory, 79- 8 1 ,
Anomalous monism. 7 1 -72 93
Aristode. 74. 90 Critical semantic mass, 1 7, 5 3
Assignment, 8 5
Austin. John L . . vii
Axioms: of choice, continuum, Darwin, Charles, 7 S
constructibility, 95; of exten­ Davidson, Donald. viii , 19n, 4 1 ,
sionality, 52n 41, 44. 71, 7 3 , 97, 99- 1 00
Dt diao , ,0-73
DefiIUtion, 8 1 -82, 86-87
Behaviorism, 3 7- 3 8 Dennett, Daniel A . , 73
Belief, 66-67, 71; warran ted, 9 3 - Dt rt, 70-73
� , 1 00 Descrip tions, 28, 92
Bergstrom, Lars, viii. 4 1 , 43 Determinism, 90-9 1
Bernays, Paul, 3 2 [ijctionaries, 56- 59
Bishop, Erret, 95 [ijspositions, 76
Bodies, 24- 26, 34 [ijsquotation, 51, 80-84, 89-9 1 ,
Bose-Einstein statistic, 20 , 3S 93
Bridge principles, 8 Dreben, Burton S. , v, viii, 4 1 , 44
Brouwer, L. E. j . , 93 Dreyfus, Hubert L . , 5 1
Duhem, Pier re 1 4 ,

Dummeti. Michael, 93
Carnap, Rudolf, 56
Cause, 7 5-76
ennis paribus, 1 8. 74 Ecumenical line, 99- 1 00
Checkpoints, 10-21, 3 8, 43-47 Einstein, Albert. 20, 3 5 . 8 1
Chisholm, Roderick M . , 65n, 7 1 n Empathy. 42-43, 46. 61-63 ,
Classes, 88-90 68-69
Coherence, 93, 98 Empirical content, 1 6- 1 8. S J -54,
Communication, 44 94-95
Concatenation, 69, 8 5 Empirical equivalence. 1 7, 95- 1 0 1
112 I N D EX
Empi rici sm , 1 9 Indeterminacy : of reference.
Epistemology, 1 -2, 8, 1 2, 1 9- 20, 50- 5 2 ; of trans la tion . 37.
42 47-5 1 ; 53. 77 . 1 0 1 - 1 02
Equivalence, 1 7. 54- 5 5 , 95- 1 0 1 Indexicals. 78
Essence. 5 6 . 74 Intensionality. 68, 71 -73
Eternal sentences, 78-79
Evidence. 1 -2 . 5. 1 2- 1 3
Excluded middle. 90-92 Kant. Immanuel. 1 8
Existence, 27-28. 92 Kirk, Robert, 48
E x periment . 9- 1 0. 1 4
Exten si onal ity . 5 2n. 72
Learning. 5- 8 , 2 3 - 24, 29-30, 3 9.
44 ; empath y in. 42. 6 1 -62
Leon elli , Michele, viii
Facts, 80
Fallibility, 2 1
Levy, Edwin, 5 1
Feferman, Solomon, 95 Lewis, Clarence I. . 8
Lexicography. 56- 59
Firth. Roderick, 7
Logic, 1 4, 27-2. 8 . 3 5 - 36• 45 .
FsUesdal. Dagfinn. viii. 4 1 . 99
78-79. 8 1 . 92.
Freedom, 91
Frege. Gottlob, 69
Lorenzen. Pau l, 95
LOwenheim-Skolem Theorem. 3 2
Future. 90-9 1

Massey. G. J. . 5 1
Gavdgai, 4�, 5 1 Mathematics. 1 5 . 56. 94-95
Geometry. 5 I . 96 Meurings. 3 7, 52-56. 77. 1 02.
Gibson. Robert F . •
viii. 99
Mind, 6 1 , 72, 74-75
God, 90-9 1
Minimum mutilation. 14. I S. S6
GOdel, Kurt. 3 2 . 95 Modal operato rs , 3 0 , 73
GreUing. Kurt. 86-87
Moore, G. E 5 5 . •

Gninbaum. Adolf, 1 6 Miihlholzer, Felix, viii

Heterological. 86 Naturalism, 19-20, 4 1 -42

Hierarchies. 84. 81)-90 Necessity, 5 5- 56, 73
Hilbert, David. 3 2 Nerves, I . 4 1 - 4 2 , 44, 62, 7 1 -72
Holism. 1 4- 1 6. 56. 9 3 -94 Norms. 1 9-20
Holophrastic sense. 8. 2 3 . 26, 34,
3 7. 50. 68
Hypotheses. 9. 1 2- 1 4; analytical . O bservatio n , 2, 1 3
43 . 45 Observation categoricaJs , 1 0- 1 2,
1 8, 23 . 94
Observation sen tences . 2-8, 1 3 .
Identity. 52 29-30; translation of, 3 9-40.
Implication. 9. 1 4 42.-4 5
 INDEX 113

Occasion sentences, 3 , 1 0, 54, 6 1 , Satisfaction, 85- 87, 89

63 -64, 78 Science, 1 -2 , 20-- 2 1 , 7 1 -73 ,
Ontological relativity, 34, 5 1 - P . 1 5- 16; analogy in, 34- 3 5 ; cush­
Ontology, 27-28, 3 1 - 3 3 , 3 5- 3 6, ioning of, 1 1 - 1 9, 95; under­
96 detennination of, !)6- 1 020 Stt
Opacity, 69-70 also Hypotheses ami Theories
Ordered p airs, 8 5 Sectarian line, 99- 1 00
Semantic ascent, 8 1
Semantics, 56-59
Paradoxes, 82-84, 86-88 Sentence versus term, 8, 2 3 , 26,
Parsons, Charles D o , 30, 88 34, ]1; in lexicography, 5 1- 59;
Perception, 1, 6 1 -68, 7 1 -72 under translation, 50-- 5 I
Physical objects, 20-2 1 , 24-26, Set theory, 5 1.n , 8 8-90, 95
34- 3 5 She�in-Williams (paint makers),
Poincare, Henri, 96, 9 7 Xll

Popper, S ir Karl R . , 1 2 Similar ity, 4, 34, 62

Possibility , 30, 73-74 Simplicity, 20, 8 1 , 95 , 98, 99
Predicate-functor logic , 27 Singular terms, 28, 92
Predicate logic, 1 4, 28, 3 5-36, 72 Sleigh, R. C . , 1011
Predication, 4, I I , 2 3 -24, 29 SpeUing, 69, 12, 8 5
Predicativity, 95 Standing sentences, 1 0, 64, 78n
Prediction, 1 -2, 1 6, 20, 90-91 Stimulation, 1 -4, 40-- 4 1.
Premack, David, 5 I Stimulus eq uivalence, 1 6- 1 1. 44
Probability, 1 3 , 20, 3 6, 95 Stimulus meaning. 3 -4, 1 2- 1 3 ,
Pronouns, 26, 3 1 1 6, 40, 42 , 44
Propositional atti tu d es , 48, 56, Strawson, Sir Peter, 24
67-73 Substitutivity, 5 3 - 540 69, 12
Propositions, 5 3 . 68, 77-78, 1 02 Synonymy, 1 6- 1 7, 44, 5 4- 5 5
Proxy functions, 3 1 - 3 3 52, 96
. Syntax, 49- 50. 69 , 71.
Putnam, Hilary, 30 Synthetic, 1 6, 1 8

Quantification, 1 4, 26-27, 30, Tarski, Alfred, 5 2 , 80, 83 -86, 88n

69- 70 Telepathy, 1 9. 21
0

Query and assen t, 3 9-40 Tense, 78

Quotation, 68-70, 72, 80, 82- 8 3 Terms, Stt Reification ami Sen-
tence versus term
Testability, 1 6- 1 8, 95
Reference, 2 7 , 52, 6 8 Tests, 1 2 , 1 7
Reichenbach, Hans, 8 Theology, 75, 90-9 1
Reification, 8, 23-25, 29-30, 42 Theories, I , 4, 6-8 , 1 3 , 95- 1 0 1 .
Relations, 8 5 Stt also Hypotheses and Science
Russell, Bertrand, 1 8, 28, 84, Time 30 ,

88-90 Transl ation , 3 7-40, 4 3 - 5 1

 INDEX

Truth. 79-80. 93; as disquotation. Variables. 26-28. 3 1 . 6cr70. 89

80-84. 8cr9 1 . 93 ; definition of.
8 1 - 82. 8 5 ; paradoxes of, 82-84.
86-88; utility of. 80-8 1 ; vehi­
Wang Hao, 95
cles of. 77-79
Warranted belief. 93 -94. 1 00
Truth functions. 29. 45. 92
Weyl, Hermann . 95
Wittgenstein. Ludwig. 20, 3 5

Ullian. Joseph S. . zo
Under-determination. 96- 1 02
Understanding. 5 8-59 Yosida. Natuhiko. 1 8