TO
BURT ORE BEN
im iend ad
constuctive critic
down the decades
opyight C 192, 10 by the Prsidnt and Fellows of Havard olege
All ights rv ed
Pnted in the Uited Stats of Amia
10 9 8 7 6 5 4 3 2 1
ist Havard Uiversity Prss
paerback ediion, 192
Libray of Congrtss Cataloging-in-wbliltion ata
I.
Quine, w. V. (Wllard Vn Omn)
Pursit of tuth I W. V. Quine. - Rev. d.
p. cm.
Includes bibliograpical referencs nd index.
ISBN -674-73951-5 (pbk. : lk. paer)
Meaning (Plosophy) 2. Refernce (Phlosophy) 3. Knowledge,
Thoy o. 4. Semnics (Plosophy) I. Title.
B94 5·Q5 3P87 192
UI-dc20
92-506
CP
Preface to the Revised Edition
irst
appeared, I was in th e g alan t ittle Republic of San Marino
for a week-long international colloquium on my philoso­
phy. Six months later I was in medieval Girona, in Cata­
loia, giving the Josep Ferrater Mora Lecturs-ifteen
hours of them and ive of discussion. Donald Davidson,
Bur to n Dreben, Daginn F0Uesdal, and Roger Gibson
were ll 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 n ges on thse occasions sparked thoughts that
would have made for a better book if the chronology had
b een inverted. I am approimang such an invers io n as
best I can by tis ear ly re vis ed edit io n
Old §13 , "Ontological r ela ti vit y,
has become more
emphaically "Ontology defused," and incorporats bits
from my responses in the projected San Maino volume.
My treatment of domestic meaning n §22 is utterly
In May 1990,
a
mere four
months ater
this book
.
"
changed, and so also, thanks to Davidson's and F0Uesdal's
a b et ting
,
are
March 192
§§28-29 on propositional
attitudes.
W.V.Q.
Preface to the First Edition
In these pages
c l a ri fy
I have undertaken to upda te, sum up, and
my variously interseng iews on cognitive mean­
ing, objective
reference, and the grounds of kn owledge.
S ome of the progress is expository and some substantive.
The substnce has ben precipitating sporadicaly over the
past ten years, and some of it has surfaced in lecturs, nfor­
mal d iscuss ions, and scattered paragraphs. In interrelating
these thoughts I have occasionlly found a faulty joint and
ha ve irmed it up to my satisfaction.
I intend tis ittle book no less for my past readers than for
my new ons, so I have curbed my exposiion of things
already belabored n my other books. I do retrace fa miliar
ground where 1 s e an impro ve ment in the idea or its pre­
sentation, and also where the new reader needs a ittle
brieing 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 iied 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 im l" y ielded earlier
parts of Chapter IV. That lecture was the third of four
Immanuel Kant Lecures that I gave at Stanford in 1980.
The tide of the seies offour was "Science and Sens ibilia , a
ake of of John Austin 's takeof of J ane Austen. The four
lecurs appeard as a little b ook in Ilian, a 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 Indeteminaies, " prsented
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,
Prpectives on Quine
( Oxford: Black­
well) . Another overlapping publication in the oing is
"Tuth," witten at the request of the Institut International
de Philosophie nd slated for
Philosophical roblems Today
( The Hague: Njho) . 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 reiication with
observation gave him
una sona di crampo mentale.
After two
letters I began to feel the cramp mysel. Result: a substantial
revision ofChapters I and II. A letter from Felix Mihlholzer
n Munich prompted me to insert a couple of paragraphs
recognzing the untidy side of scientiic method .
A
diiculty spotted by Lars Bergstrom of Stockholm is now
noted and dealt with in the text, and my indebtedness to
Donald D aidson, Daginn F011esdal, and Roger Gibson is
noted at appropriate points. I am much indebted to Buton
Drebn, who has read earlier drafts with care and insight
nd has made many helpful suggestions.
W.V .Q.
CONTENTS
I . EV I D EN C E
I
I.
Stimulation and prediction 1
2. Obsrvation sentences 2
) . Theoy-ladn? 6
4· Observation categoricals 9
5. Test and reutation 12
6. Holism I)
7. Empiri4l content 1 6
8. Noms and aims 19
II . REF ER EN C E
23
9. Bodies 2)
10. V.lues of variables 25
11. Utility of reication 29
12. Indferce ofontology )1
I). Ontology deosed ))
I I I . ME A N I N G
37
'4. Thefold linguist's ntering wedge )7
15. Stimulation again 40
16. To each his own 42
17. Translation resumed 44
18. Indeteminacy of translation 47
19· Syntax 49
x
C O NTE NTS
20. Indeteminacy of refrence 50
21. hither meanings?
52
22. Domestic meaning 53
23. Lexicography 56
I V. I N TE N S I O N
61
24. Prcption and obsevation sentnces
25. Percption extended 63
26. Percption of things 64
27. Belief and prcption 65
28. ropositioal attitudes 67
29. Anomalous monism 71
30. Malities 7 3
31. A mentalistic heritage 74
V. TRUTH
77
32. Vehicles of tu th 77
33. Tuth as disquotation 79
34. Paradox 82
35. Tarski's constuction 84
36. Paradox skited 86
37. Interlocked hierarchies 88
38. Excluded middle 90
39. Tuth versus waranted belief 93
40. Tuth in mathematics 94
41. Equivalent theories 95
42. Iresoluble ivaly 98
43. Two indeteminacies 101
References 105
Credits 109
Index 111
61
PURSUIT OF TRUTH
'W'�.I Tl x.I6�Q.
PLATO
Save the surface and you save all.
SHERWIN-WILLIAMS
I
EVIDE NCE
1.
Stimulation ad p rediction
From impacts on our sensory surfaces, we in our collective
and cumulative creaivity down the generations have pro­
jected our systematic theory of the extenal world. Our
system is proving successfl in predicting subsequent sen­
sory input. How have we done it?
Neurology is opening strange new vistas into what goes
on between stimulaion and erception. Psychology and
more particlarly psycholinguisics may be loked to for
something to say about the passage from percepion to ex­
pectaion, generlization, and systematization. Evolution­
ay genetics throws further light on the latter matters, ac­
countng for the standards of simlaity that nderlie our
generalizations and hence our expectations. The heuristic of
scientiic creativity is illuminated also, anecdotally, by the
history of science.
Within tis baling tangle of relations betwen our sen­
sory stimulation and our sientiic theory of the world,
there is a segment that we can gratefully separate out and
clarify without pursuing neurology, psychology, psycho­
linguistics, genetics, or istoy. 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
thn logical analysis .
Not that prediction is the main purpose of science . One
major purpose is understanding . Another is control and
modiication 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
observaion, and that what we predict are observations . But
the notion of observation is awkward to analyze. Clariica­
tion has ben sought by a shift to observable objects and
events. But a gulf yawns between the m and our immediate
input from the e xtenl world, which is rather the trigger­
ing of our sensory receptors. I have cut through ll this by
settling for the triggering or stimulaion iself and hence
speing, oddly perhaps, of the predicion of stimulation .
By the stimulation undergone by a subj ect on a given occa­
sion I j ust mean the temorally 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
quesion of evidence for science without help of 'evidence '
s
a tecnicl term . We can make do instead with the notion
of observaion sentences .
2.
Obsrvation
sentences
We were ndertaking to examine the evidential support of
sience. That support, by whatever name, comes now to e
seen as a relaion of stimlation to s cieniic theory. Theory
consists of sentences, or is couched
n them; and logic
con­
nects sentences to sentencs. hat we need, then, as iitial
EV I D EN C E
3
links in those conecng chains, are some sentences that a re
direcdy and irly associated with our stimulations. Each
should be assoiated airmaively 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 occsion of a simulation in the appropiate range, with­
out further nvestigation and independendy of what he may
have been engaged in at the ime. A urther requirement is
intersubj ectivity: unlike a report of a feeling, the sentence
must command the same verdict from al linguistically
competent witnesses of the occasion.
I cal them
obsevation sentences.
Exampls are 'It's rain­
ing' , ' It's getng cold', ' That's a rabbit' . Unlike ' Men are
mortal' , they are occasion sentences: true on some occsions,
false on others . Sometimes it is raining, sometimes not.
Biely 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.
Observaionity is vague at the edges . There are grada­
ions in an individul 's readiness to assent. What had passed
for an obser vation sentence, say 'That's a swan', may to the
subject's own surpise leave him undecided when he en­
counters a black specimen. He may have to resort to con­
vention to settle his usage. We shal need now and again to
remind ourselves thus of the untidiness of human behavior,
but meanwhle we foster perspiuity by fancying bound­
aries.
he range of simulations assoiated with an observation
sentence, airmatively or negatively, I cal its airmative or
negative
stimulus meaning for the given speaker.
Each of the
stimulations, by my deinition, 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 eliit behavior. Hence the stimulations encompased in a
stimulus meaning wll difer wildly from one another in
their inefective irings, but in their efective core they are
ound to be similar to one another in some rspect, 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 eicited behavior.
An obse rvation sntence may consist of a single noun or
adjetive, thought of as a sentence; thus ' Rain' , 'Cold', or
'Rabbit' , for ' It's raning ' , 'It's cold ' , ' It's a rabbit'. Obser­
vaion sentences also may
e compounded to
form further
obervaion sentences, for example by simple
conjunction:
'The sun is rising and birs 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 mean g . But they are not equivalent to the mere
conjuntion 'Lo , a pebble, and 10, blue'. Their c onnection is
tighter. The conj unction is ulilled so long as the stimula­
ion shows each of the component obsevation sentences to
e fullled somewhere in the scene-thus a white pebble
here, a blue lower over there. On the other hand the pre­
cation fouses the two fulllm ents, requirng them to coin­
ide or a mply overlap . The blue must encompass the
e­
ble. It may also extend beyond; the construction is not
symmetic.
What brought us to an examination of observation sen­
tences was our q uest of the nk between observation and
theory. The obervation sntence is the means of verbaliz­
ing he prediction that checks a theory. The requirement
1
Hnce ercepally similar, not recepually. Ros
pp. 1-18.
of Rermce,
EV ID EN C E
s
that it command a verdict outight is what makes it a fml
checkpoint. The requirement of intersubjectivity is what
makes science objetive.
Observation sentences are thus the vehicle of scientiic
evidence, we might say-though without venturing a dei­
nition of 'ev idence' itself. But also they are the entering
wedge n the leaning of language. The infant's irst acqui­
sitions in cognitive language are rudimentary observation
sentences, including 'Mama', 'Milk', and the ke s one­
word observation sentences. They become associated with
stimulations by the conditioning of rsponses. Their direct
association with concurrent stimulation is ssential 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 sientiic evidnce and as enterng wedge into lan­
guage-is no cause for wonder. Observation sentences are
the link between language, scientiic or not, and the rl
world that lanuage is all about.
Observation sentences as I have deined them far exceed
the pimitive ones that are the chld's entering wedge.
Many of them are leaned not by simple condiioning or
imitaion, but by subsequent construction from sophis­
ticated vocablary. The requirement of direct correspon­
dence to ranges of stimulation can e met either way.
Which ones are leaned directly by conditioning, and which
ones indiretly through higher language, will vary from
person to erson. But the two requirements, ntersubjectiv­
ity and corresondence to simulation. assure us that any
observation sentence could e leaned n the direct way. We
6
P U R S U I T OF T R U T H
hear our fellow speakes aiming and denying the sentence
on just the occasions when we are simulated in the charac­
teistic ways, and we join in.
J.
Theoy-laden?
My deiition 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 ecome fashionable to
question the notions, and to claim that the purportedly o­
servable is theory laden in varying degrees. It is ponted out
-
that when scientists marshal and check their own data or
one another's, they press no fa ther 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 slide is escaping are observational enough for
'
any of them, and more recondite reports are obse vatio nal
enough for some. I agre that the pracical notion of obser­
vaion is thus relaive to one or another limited community,
rather than to the whole speech community. An obseva­
tion sentence for a commnity is an occasion sentence on
which members of the community can agree outright on
itssing the occasion.
For philosophical purposs we can probe deeper how­
,
ever, and reach a single standard for the whole seech com­
munity. Obsevable in this sense is whatever would be at­
tested to on the spot by any witnss n command of the
language and is ive senses.
If scientists were perversely to
persist in demanding further evidence beyond what suiced
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 indescibable smell of some uncommon gas,
would resist reduction.
But w hat has all this to do with a sentence's being theoy­
la den or theory-free? My deiiion dstinguishes obser­
vation sentences from others, whether relative to speial
communiies or to the general one, without reference to
theory-freedom. There is a sense, as we shal now see, in
which they are al theory-laden, even the most primitive
ons, and there is a sense in which none are, even the most
p rofessional ones.
Think irst of primitive ones, the enteng wedge in lan­
guage leaning. They are associated as wholes to appropri­
ate ranges of stimulation, by conditioning. Component
words are there merely as component sylables, theory­
free. But these words recur in thoretical contexts n the
fullness of time. It is precisely this s haring of words, by
observaion sentences and theoretical sentencs, that pr­
vides logical conections between the two kinds of sen­
tences and makes observaion relevant to scientiic 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 theoreical sentences containing terms as technical as
'H20' . Seen holop hrastically, as conditioned to stimulatory
situations, the sentence is thoy-free; seen analyticaly,
word by word, it is theory-laden. Insofar as observation
sentences bear on science at all, afording evidence and tsts,
there has to be this retrospective theory -lading along with
the prisine holophrastic freedom from theoy. To impugn
their observaionality thus retrospecively is to commit
what Firth (p. 10) called the fallacy of conceptual retrojec­
tion.
8
P U R SU IT OF TRU TH
More sophisticated observation sentences, inc lud ng
those of specia li zed sientiic communities , are sim larly
tw -faced, even tho ugh learned by composition rather than
direct condiio ng. What qu aliies them as obse rvation
sentences is still her holophras ic 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 jecive" ones.
When epistemology rounded the linguistic um, talk of
observable objects ga ve way to tal k of observation terms. It
was a g od mo ve, but not good enough. Observation sen­
ten cs w ere distinguished from theoret ical ones only
deivati vely, as containing observation terms to the exclu­
sion of theory-laden or theoreical terms. Consequently
Rei chenbach and others felt a need for "bridg e p rincip les "
to re late the two kin s of sentences. No bridge is wanted,
we now see, and bridging is the w rong igure. 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 bon of
freeing the deinition of observation sentence from any de­
endence on the dist nction 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 late on the nat ure of reiication and its utility for
scientiic theory -a topic for Chap ter II. Taking terms as
starting point wou ld h ave me ant inssing reiication and
EVI D ENC E
9
con ceding objetive reference out of hand, without consid­
e ing what it is for or what goes into it.
4.
Observation categoricals
The suppot of a theory by obsevation stands forth most
expl icitly in expeiment, 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 efect should be observed. So he set s up the situation
in qu estion. If the p redi ct ed efect fail s to appear, he aban­
don s hi s hypothesis. If the efet 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 i eld minerlogist s have tu ned
up an unfam liar rystaline minerl of a distinc ively p in k­
ish cast. They speak of it provisionally as litholite, for want
of a better name. One of them conjeures its ch emical
co mposition. Tis 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 sulide 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 coleagues know litholite when
they see it and hydrogen sul ide when they smel it, and
t hey can read a the rmometer.
The t st of a hypothesis thu s hin gs on a lo gical rel ation
of i mpli caion. 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 ave an implied gneraity that the experimenter
can directly test, directly challenge-in ths case by heating
some of the pink suf and siing.
A generality that is compounded of observables in this
way-'Whenever this, hat'-is what I call an obsevation
ategoial. 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 efect
that the cirumstances described in the one observation sen­
tence are invaiably accompanied by hose dscibed in the
other.2
Though compounded of two occasion sentences, the ob­
sevation categorical is iself a standing sentence, and hence
fair game for implication by scientiic thory. It thus solves
the problem of inking theory logically to observation, as
wel as epitomizing the experimental situaion.
That situaion is where a hypothesis is being tested by
an experiment. An opposite situation is equally familiar: a
chance observation may prompt us to conjecure a new
obsevation categorical, and we may nvent a theoretical
hyothsis to explain it. For example, we might notice wil­
lows leaing over
a
stream. This suggests the observation
categorical:
(I)
When a willow grows at the water's edge, it leans
over the water.
2
he obsevaion categorical is not to e confud with the obseva­
ion colitioz, a less fruitful notion that I ventured in
1975. he observa­
tion conitional is formd from two stnding sentences each of which has
enbuilt upon an observation sentence with help of theory.
al hings, pp. 6-,7.
ee heoies
EVI D E N C E
II
Tis suggests, n um, a theoretical hyothsis : 'A willow
root nourishes maly its own side of the tree' . Taken t ­
geth er with prior bits o f theory , such a s that roots get more
nourishment from wetter ground, nd that nourishment
p ro m ote s the growth of branches , the hypothesis is found
to imply the obser vation categorical. Other ob se vation
cate go icals will be imp lied too, and the continud testin g
of th e hyothesis would proceed b y t estng various of
them , a long with further testing of the one that happened to
sugge st the hypothesis.
The o bservation categoricl (1) exceeds my deinition n a
sub tle way : it is not compounded of two self-su icient o b­
serv ation sentences. It cannot be read 'When a w llow
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 ree 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 ng the one feature w ll p resen t the other somewhere
about. The focal observation categorical generalizes ra ther
on a pre dica tiona l observation sentence. (1) gener aizes o n
the predicaion 'This riverine willow leans over the w ater'
t o say that they all do.
A more sucnct 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
ad reotation
An observation categorical is tested by pairs of observa­
ions. It is not conclusively veriied by obsevations that are
conformable to it, but it is refuted by a pair of observaions,
one airmative and one negative-thus observation of
lithoite at
I80Ge but absence of hydrogen sulide, or obser­
vation of riveine wilows leang away from the water.
he fre observation categoical 'When the sun comes up
the birds sing' is refuted by observing sunrise among silent
birds.
The obsevational test of scientiic hypotheses, in turn,
and nded of sentences generally, consists in testing obser­
vation categoicals that they imply. Here again, as in the
cae of the observation categoical itself, there is no conclu­
sive veriiation, but only reutation. Reute an obsevaion
categorical, by an airmative and a negaive observation,
and you have refuted whatever implied it.
Traditional epistemology sought grounds in sensoy ex­
peince capable of implying our theoies 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 oly to refute theory and
not to supot it. We have now been seeing in a schematic
way why this is so.
But again we mst bear in mind, as in §2, that we are
schematizng: positing shap boundaries where none can e
drawn. The pair of observations in purported refutaion of
an observaion categorical may be indecisive ecause of un­
foreseen indeision over the stimulus meaning of one of the
pair of observation sentences, as
n 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 e refuted by the discovery of
the odd specimen, depending on our own decision regard­
ing the vague simulus meang 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 contnually reasons
not only in refutation of hy pothess but in support of them.
This, however, is a matter of argung logically or probabl­
isticaly from other beliefs aready held. It is where the tech­
nology of probability and mathematical statisics is brought
to bear. Some of those supportng beliefs may be observa­
tional, but they contibute support only in company with
others that are thoretical . Pure observation lends oly
negative evidence, by refuting an observation categoical
that a proposed theory implies.
6.
Holism
Let us recall that the hypothesis regarding the chemicl
composition of itholite did not im p ly its observation
categoical single-handed. It implied it with the help of a
backlog of accepted sientiic theory. In order to deduce an
obsevation categoical from a given hy pothesis, we may
have to enlist the aid of other theoretical sentencs and of
many common-sense platitudes that go without saying,
and perhaps the aid even of aithmeic 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 neded to imply the
observation categoical. In order to retract that conjuncion
14
P URS U IT OF TRU T H
we do not ha ve to retract the hypothesis in qusion; we
could retract some other sentence of the conj unction in­
stead. This is the important nsight called holism. Pierre
Duhem made muc h of it early in this century , but not too
m uch.
The sientist thinks of his experiment as a tst spe iically
of his new hypothesis , but o ly because t his was the sen­
tence he was wonde ng about and is prepared to re ject.
Moreo ver, there are also the situations where he has no
preconcei ved hyothesis, but just happens upon an anoma­
lo us phenomenon . It is a case of his happening upon a
co unter-instance of n observation categoical which, ac­
cording to his c urrent theory s a whole, ought to have been
true. So he loo ks to his theory with a c ritical eye.
Over-Iogii 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 flse categori cl. Implication may be taken here
simply as deducibility by the logic of t uth nctio ns,
quniicaion, and identity. (We can always pro vide for
more substan ial 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 is thre at on determining
that the fateful implication still holds witho ut their help .
Any p urely logical tuth 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 ers of S, we rsind one that seems
most suspect, or least c ucial to our o veral theory. We heed
a maim of mi nimum m utilation. If the remaiing mem­
bers of S s ill conspire to imply the fa se categoricl, we try
EVI DENCE
15
res inding another and restoring the irst. 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 begining. We must also track down
sets of s entences elsewhere, in our overal th eo y, that im­
pl y th ese newly rescinded beliefs ; for thos e must be defused
too. We continue thus untl co sistency sems to be re­
s tored. Such is th e mutilation that the ma im of min imum
muti lation is meant to minim ze.
In p articular the maxim constrains us, n our choice of
what sentences of S to resind, to safeguard any purely
mathematicl t uth ; for mathemat ics in ltrates all branches
of our system of the world, and its disruption would r e­
verb erate intolerably. If sked why he spare s mahematics,
the scientist wil perhaps say that its laws are necessaril y
tue; 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 ng mathematics by exercising our freedom to
re ject other beliefs instead.
So the choice of wich of th e b eiefs to r ject is i ndifer ent
only so far as the failed observation categorical is con­
cened, 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 ltimat e objective is so to choos e the r evision a s
to m aimi ze future success i n pr ediction: future coverage of
tue 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
n future predi ions.