AYER_The Foundations Of Empirical Knowledge_I_The Argument From Ilusion.pdf

7/29/2019 AYER_The Foundations Of Empirical Knowledge_I_The Argument From Ilusion.pdf

http://slidepdf.com/reader/full/ayerthe-foundations-of-empirical-knowledgeithe-argument-from-ilusionpdf 1/33

By tfu samc uthor*

PHILOSOPHICAL ESSAYS

THE PROBLEM OF KNOWLEDGE

THE REVOLUTION IN PHILOSOPHY (withothers)

THE CONCEPT OF A PERSON,AND OTHER ESSAYS

THE FOUNDATIONSOF

EMPIRICAL KNOWLEDGB

BY

ALFREDJ. AYER,M.A.,F.ts.A.FEI.LOW OF NEW COLI,EOB AND

WYRTHAM PROFHI{IOR OT LOGIC IN TIIE T'NIVERSITY

oa oxFoRD

LONDON

MACMILLAN g CO LTD

NE'W YORtr . ST MARTIN,S PRESS

tg63



This booh&t copytight in all cowilries whichare signatoriesn the Bene Contetttion

Fkd Etlition gpRcFinM t91?, t95r, 1953, t955, r95E, tg6z, tg6g

MACMII,L.ITN AND COMPANY LIMITED

St Ma*in's Street Loadon WCz

also Bombay Calattt4 Ma&u Melboma

TIIE MACMILI.AN OOMPANY OF CANADA LIMITED

Torcnto

6T'MARTIN'S PRESS TNC

New Yorh

PBINTED IN ONEAT ERITAIN

TO

VALERIE AYER



PREFACETun title of thisbook covers widerangeof subjects

and I do not wish to claim that I have investigated

them all. My main purposehas been o resolve hephilosophical roblemswhich arecommonlybrought

under he headingof " our knowledge f the external

world ". But I havealso ound occasiono dealwith

ruch further questionsas hose of our knowledgeof

other minds, of the characterof causal aws, of thernelysisof meaning,and of the natureof propositions

md their relation o facts.

Where I believe that I am indebted to other

luthors I have made my acknowledgementsn the

h*t. But I should like,here to pay a tribute to

rProfcasorH. H. Price'swgrk on Perception,o which

I m'o considerablymore han my frequent criticisms

It would suggest.My thanksare due also to my friend Mr. C. E.

for reading he proofs.

A. J. AYER

ol Gurnoe Drrdt,

CAttrxru, Sunnrr,March, g4o

YU



CONTENTS

I. Tnr AncuvrrNr rnou llr'usroN ' ' '"^T

r. ExPositionof the Argument 'r

z. Evaluationof the Argument ' rr

3. The Introductionof Sense-data rg

i. Mirrrr., of the Argument from Illusion ' z8

i. Theoriet of Perception as Alternative-L*g.r"g"t."46

II. THe CnenactrnrzATloN or SeNse-oerA'' ' 58

6. Acts and Objects n Sensation ' ' 58

7.ttEsseestPerciPi" ' ' ' '65

i. Sensingand Knowing 78

9. The Errors of Formalism '84

ro. Sentences,ropositions, nd Facts' ' 92

rr. The Nature of the " Given " ' ' rr3

III. TnB Ecocentnrc Pnnorc'nrvrnrlr ' '! '136

rz.The Privacy of PersonalExperience

'46

13. Public and Private Languages r.46

ti. Corr""*ing the Privacy of Sense-data nd

the Publicity of the Material World r53

15. The Hypothesis of the Existenceof Other

People'sExPeriences ' ' t6z

IV. Cluser,rrY AI'IDPrncrpuoN ' ' 'r7r

16. The CausalTheory of Perception ' ' r7r

17. Formulation of " the Principle of Deter-

' minism" .ix

' ' ' r79



v.

FOUNDATIONS OF EMPIRICAL KNOWLEDGE

r8. The Animistic Idea of Necessary Con-PAGB

nexion rg319. Criticism of the Rationalist Interpretation

of CausalLawsryg

eo. Evaluation of .. the principle of Deter_minism,, . zo7

zr. The Causationof Sense-data . . zzo

Tnn CoxsururroN or Mlrrnrer- TnrNcs . zzgzz. ConcerningPhenomenalism zzg23. Elementary Construction of the Material

World z4S24. Appearanceand Reality . . z6g

INnsr. 275

I

THE ARGUMENT FROM ILLUSION

r. ExposrtroN oF THEARGUMENT

I'r does not normally occur to us that there is any

rrccd for us to justify our belief in the existenceof

matcrial hings. At the presentmoment, for example,

I have no doubt whatsoever hat I really am per-

cciving the familiar objects, he chairs and table, the

pictures and books and flowerswith which my room

ir furnished ; and I am therefore satisfied hat they

cxist. I recognize ndeed that people are sometimes

tlcccived by their senses; ut this does not lead me

to suspect hat my own,sense-perceptions annot in

gcneralbe trusted, or even hat they may be deceiving

mc now. And this is not, I believe, an exceptional

nttitude. I believe that, in practice, most peoplengree with John Locke that " the certainty of things

cxisting 'in rerum natura, when we have the testimony

of our senses or it, is not only as great as our frame

can attain to, but as our condition needs." t

When, however, one turns to the writings of

tlrosc philosophers who have recently concerned

' An Essay concerning Human Underctandhry, Book IV, ch. a,

rcction viii.



2 FOUNDATIONS OF EMPIRICAL KNOWLEDGE r

themselveswith the subject of perception, one maybegin to wonder whether this matter is quite sosimple. It is true that they do, in general, allowthat our belief in the existenceof material things iswell founded ; someof them, indeed,would say hatthere were occasionson which we knew for certainthe truth of such propositions as" this is a cigarette ',or " this is a pen ". But even so they are not, forthe most part, prepared to admit that such objectsas pens or cigarettes are ever directly perceived.What, in their opinion, we directly perceive s alwaysan object of a different kind from these; one towhich it is now customary to give the name of

" sense-datum . These sense-data re said to havethe " presentative unction " r of making us consciousof material things. But how they perform thisfunction, and what is their relation to the materialthings which they present, are questions about whichthere is much dispute. There is dispute also aboutthe properties of sense-data apart rom their relation-ship to material things : whether, for example, they

are each of them private to a single observer;whether they can appear to have qualities that theydo not really have, or have qualities that they do notappear to have; whether they are in any sense" within " the percipient's mind or brain. I shallshow later on that these are not empirical questions.They are to be settled by making it clear how theterm " sense-datum is intended to be used. But

first I must explain why it is thought necessary ot Cf. H. H. Price, Paceftion, p. ro4.

'THE ARGUMENT FROM ILLUSION 3

introduce such a term at all. Why may we not say

that we are directly aware of material things ?

The answer is provided by what is known as the

argument from illusion. This argument, as it is

ordinarily stated, is based on the fact that material

things may present different apPearanceso different

observers,or to the same observer in different con- 1ditions, and that the characterof theseaPPearancet\ ,,is to some extent causally determined by the state i V

of the conditions and the observer. For instance,

it is remarked that a coin which looks circular from

one point of view may look elliptical from another;

or that a stick which normally appearsstraight looks

bcnt when it is seen n water; or that to people who

ttke drugs such as mescal, hings appear to change

their colours. The familiar casesof mirror images,

and double vision, and complete hallucinations, such

irs the mirage, provide further examples. Nor is

this a peculiarity of visual appearances. The same

thing occurs in the domains' of the other senses,

including the sense of touch. It may be pointed

out, for example, that the taste that a thing appearsto have may vary with the condition of the palate;

or that a liquid will seem o have a different temPera-

ture according as the hand that is feeling it is itself

hot or cold ; or that a coin seems arger when it is

placedon the tonguethan when it is held in the palm

of the hand ; or, to take a caseof complete hallucina-

tion, that people who have had limbs amputated may

still continue to feel pain in them.f,et us now consider one of these examples, say




that of the stick which is refracted n water, and seewhat is to be inferred. For the present it must beassumedthat the stick does not really change itsshapewhen it is placed n water. I shall discuss hemeaning and validity of this assumption later on.

Then it follows that at leastone of the visual appear-ancesof the stick is delusive for it cannot be bothcrookedand straight. Nevertheless,even n the casewhere what we see s not the real quality of a material

thing, it is supposedhat we arestill seeingsomething;

and that it is convenient o give this a name. Andit is for this purpose hat philosophershave recourseto the term " sense-datum . By using it they areable to give what seems o them a satisfactoryanswer

to the question: What is the object of which weare directly aware, n perception, f it is not part ofany material thing ? Thus, when a man sees amirage in the desert, he is not thereby perceiving

any material thing; for the oasiswhich he thinkshe is perceiving does not exist. At the same ime,it is argued, his experience s not an experienceofnothing ; it has a definite content. Accordingly, it

is said that he is experiencingsense-data,which aresimilar in character to what he would be experiencing

if he were seeinga real oasis,but are delusive n thesensethat the material thing which they appear topresent is not actually there.. Or again, when Ilook at myself in the glassmy body appears o besome distancebehind the glass but other observa-tions indicate that it is in front of it. Since it isimpossible for my body to be in both these places

r THE ARGUMENT FROM ILLUSION 5

ir t or)ce, heseperceptionscannot all be veridical. I

lx'licve, in fact, that the ones that are delusive are

tlrosc in which my body appears o be behind the

gllss. But can it be denied that when one looks at

oncsclf n the glass one is seeingsomething? And

il', in this case, here really is no such material thing:rs rny body in the placewhere it appears o be, what

is it that I am seeing? Once again the answerwe

:rrc nvited to give is that it is a sense-datum. And

tlrc sameconclusion may be reached by taking any

otlrcr of my examples.

If anphing is establishedby this, it can be only

ttrat there are some cases n which the character of

()lrrperceptionsmakes t necessaryor us to say that

what we are directly experiencing s not a material

thing but a sense-datum. It has not been shown

ttrat this is so in all cases. It has not been denied,

lrrrt rather assumed, hat there are someperceptions

tlrat do present material things to us as they really

rrrc and in their case here seemsat first sight to

bc no ground for saying that we directly experience

scnse-data ather than material things. But, as I

lrave already remarked, there is general agreementamong the philosopherswho make use of the term

" scnse-datum , or someequivalent erm' that what

wc immediately experience s always a sense-datum

rundnever a material thing. And for this they give

lirrther argumentswhich I shall now examine.

In the first place it is pointed out that there is

no intrinsic difference in kind between those of our

pcrceptions that are veridical in their presentation



6 FOUNDATIONS OF EMPTRICAL KNOWLEDGE r

of material things and those that are delusive.,When I look at a straight stick, which is refractedin water and so appears crooked, my experience isqualitatively the sameas if I were looking at a stick

that really was crooked. When, as the result of myputting on green spectacles,he white walls of myroom appear to me to be green, my experience squalitatively the same as if I were perceiving wallsthat really were green. When people whose legshave been amputated continue to feel pressure.rponthem, their experience is qualitatively the ,u*" u.if pressure really were being exerted upon their

1.g.. But, it is argued, if, when our perceptionswere delusive, we were always perceiving somethingof a different kind from what we perceivedwhen theywereveridical, we should expectour experienceo bequalitatively different in the two cases. We shouldexpect to be able to tell from the intrinsic characterof a perception whether it was a perception of asense-datumor of a material thing. But this is not

possible, as the examples that I have given haveshown. fn some cases here is indeed a distinctionwith respect to the beliefs to which the experiencesgive rise, ascan be illustrated by my original example.For when, in normal conditions, we have the ex-perience of seeing a. straight stick, we believe thatthere really is a straight stick there ; but when thestick appears crooked, through being refracted in

water, we do not believe that it really is crooked;we do not regard the fact that it looks crooked in


w:rlcras evidenceagainst ts being really straight. It

rrrrrst,however, be remarked that this difference n

tlrc beliefswhich accompanyour perceptions s not

;,1r'oundedn the natureof the perceptionshemselves,

lrrrt dependsupon our past experience. We do notlrt'licve hat the stick which appearscrookedwhen it

nrlrrds n water really is crooked becausewe know

l'r'ornpast experience hat in normal conditions it

kroksstraight. But a child who had not learned hat

rcl'ractionwas a meansof distortion would naturally

hclicve hat the stick really was crookedas he saw t.

'l'hc fact, therefore, that there is this distinction

lrt:twcen the beliefs that accompany veridical andrk'ltrsiveperceptionsdoes not justify the view that

tlrcscare perceptionsof genericallydifferent objects,

cspcciallyas the distinction by no meansappliesto

nl l cases. For it sometimeshappens hat a delusive

cxperience s not only qualitatively indistinguishable

lirrm one that is veridical but is also itself believed

ro be veridical, as n the exampleof the mirage ; and,

t'onversely, here are casesn which experienceshatrurcactually veridical are believedto be delusive,as

whcn we see something so strange or unexpected

rhat we say to ourselves hat we must be dreaming.

'l'he fact is that from the characterof a perception

consideredby itself, that is, apart from its relation

to further sense-experience,t is not possible o tell

whcther it is veridical or delusive. But whether we

:rrc entitled to infer from this that what we immedi-:rtcly experience is always a sense-datumremains



8 FOUNDATIONS OF EMPIRTCAL KNOWLEDGE r

Another fact which is supposed o show that evenin the caseof veridicalperceptionswe arenot directlyawareof material things is that veridical and delusivlperceptions may form a continuous series,both withrespect to their qualities and with respect o the con_ditions in which they are obtained.' Thus, if Igradually approach an object from a distance I maybegin by having a series of perceptions which arldelusive in the sensethat the object appears to besmaller than it really is. Let us assume that thisseries terminates in a veridical perception. Thenthe difference in quality between this perceptionand its immediate predecessorwill be of the same

order as the difference between any two delusiveperceptions that are next to one another in the seriesand, on the assumption hat I am walking at a uniformpace, the same will be true of the difference in theconditions on which the generation of the seriesdepends. A similar example would be that of thecontinuous alteration in the apparent colour of anobject .i'tti"tt was seen ., a grui.rally changing light.Here again the relation between a

veridical f.r".p-tion and the delusive perception that comes next toit in the series is the same as that which obtainsbetween neighbouring delusive perceptions, bothwith respect to the difference in quality and withrespect to the change in the conditions ; and theseare differences of degreeand not of kind. But this,it is argued, is not what we should expect if theveridical perception were a perception of an object

t Cf. price, op. cit. p. 32.

r TI{E ARGUMENT FROM ILLUSION e

,.rfa difierent sort, a material thing as opposed to a

scnse-datum. Does not the fact that veridical and

and nevera material thing.

The final argument that has to be considered in

this context is based upon the fact that all our Per-

for example, that the stick looks crooked because t

is seen n water ; that the white walls appear green

to me because am wearing greenspectacles; that

the water feels cool becausemy hand is hot; that

the murderer sees he ghost of his victim becauseof

his bad conscience or because he has been taking

drugs. In the case of perceptions that we take to

be veridical we are apt not to notice such causal

dependencies, ince asa rule it is only the occurrence

of the unexpected or the abnormal that induces us

tolookforacause.But inth ismatteralsothereis

no essentialdifference between veridical and delusive

perceptions. When, for example, I look at the piece



ro ITOUNDATIONS OF EMPIRICAL KNOWLEDGE r

of paper on which I am writing, I may claim that Iam seeing it as it really is. But I must admit thatin order that I should have this experience t is notsufficient that there should actually be such a pieceof paper there. Many other factors are necessary,

such as the condition of the light, the distance atwhich I am from the paper, the nature of the back-ground, the stateof my nervous systemand my eyes.A proof that they are necessarys that if I vary themI find that I have altered the character of my per-ception. Thus, if I screw up my eyes I see twopieces of paper instead of one ; if I grow dtzzy theappearance f the paper becomesblurred ; if I altermy position sufficiently it appears o have a differentshape and size; if the light is extinguished, oranother object is interposed, f cease o see t alto-gether. On the other hand, the converse does nothold. If the paper is removed I shall ceaseo see t ;but the state of the light or of my nervous systemor any other of the factors that were relevant to theoccurrtnce of my perception may still remain thesame. From this it may be inferred that the relation

between my perception and theseaccompanyingcon-ditions is such that, while they are not causally de-pendentupon it, it is causallydependent upon them.And the same would apply to any other instanceof a veridical percepti6n that one cared to choose.

This point being estabfished, the argumenr pro-ceedsas follows. It is held to be characteristic ofmaterial things that their existenceand their essential

properties are ndependent of any particular observer.

1 THE ARGUMENT FROM ILLUSION tI

lior they are supposed o continue the same'whether

tlrcy are observedby one Personor another' or not

,,lr*crved at all. But this, it is argued' has been

rlrown not to be true of the objects we immedi-

rrtcly experience. And so the conclusion is reached

rl,ai what weimmediately experience is in no case

u materi"l thing. Accord,ing o this way of reasoning'

il'some p.r".ptio,,s are rightly held to be veridical'

rurd others d'elusive, it is because of the different

rclations in which their objects stand to material

things, and it is a philosoPhi:l problem to discover

*hol ih.r" relatio-nsare. we may be allowed to

lrrve indirect knowledge of the properties of material

things. But this knowledge, it is held' must be

obtalned through the medium of sense-data' since

thcy are the only objects of which' in sense-perceP-

tion, we are immediatelYaware'

z. EveruerloN oF THE ARGUMENT RoM Ilr'usroN

With this I complete my exposition of the so-

cllled argument from illusion' In considering its

validity ii is important first to determinewhether

the question it raises concerning the nature of the

,,bjccts that we directly perceive is to be regarded

".'"qrr".aion of language or as a question of fact'

In most cases he philosophers who have made use

of this argument have taken it to prove a rnatter of

fact. They have inferred from it' not merely that it

is linguistically inconvenient' but that it is false to

sav thatwe are ever d'irectly aware of a material



12 ITOUNDATIONS OF EMPIRICAL KNOWLEDGE r

thing. But if the argument is interpreted in thisway it is evidently not conclusive. In the first place,when one examines the subsidiary arguments whichare supposed to prove that what we perceive whenour perceptions are veridical cannot be generically

different from what we perceive when they arldelusive, one finds that each of them rests upon apremise that is open to question. It is taken forgranted that if veridical and delusive perceptionswere perceptions of objects of different types, theywould always be qualitatively distinguishable ; oithat they would not, in respectof their qualities andthe conditions of their occurrence, be capable of

being ranged n a continuousseries or thirdly, thatmaterial things can exist and havepropertieswithoutbeing causallydependenton any observer. But eachof these assumptionscould be denied without self_contradiction. Nor do the first two appear to admitof any empirical proof. As for the third, it is truethat we find reason to believe a number of hypo-thetical propositions about the experiencesthai we

should be having if we were in certain situations inwhich we actually are not. And if the propositionthat material things are causally independeni of ourobservation of them is understood to imply no morethan that some such hypothetical propositions aretrue even though their protases are never fulfilled,we may consider ourselves ustified in maintainingit on inductive grounds. But though this is perhaps

the most natural way to interpret this proposition,and the one that I shall adopt myself, it is not the

I THE ARGUMENT FROM ILLUSION 13

intcrpretation hat is required by the argument from

illusion. For these hypothetical propositions con-

ccrning what we should observe n certain unfulfilled

conditions might very well be true of objects whose

rnanifestationsproved, when we actually did observe

thcm, to be causally dependent upon our observa-

tion. If, therefore, the proposition that material

things are independent of our observation of them

fncans no more than the certain hypothetical pro-

positions about our sense-experiences ay be both

tme and unfulfilled, it does not entail the con-

clrrsion that the objects we directly perceive, being

causallydependent uPon the state of the observer

rrnd the accompanyingconditions, are not material

things. But if we have to interpret the proposition

in such a way that it doesentail this conclusion, hen

it no longer appears capable of being justified on

cmpirical grounds. It might indeed be made a

rnzrtterof definition that material things should be

t:ausally ndependent of any Processof observation,

in some sensefrom which it would follow, in con-

jrrnction with the facts I have mentioned about theL,,rr.", of our sense-exPeriences,hat the objects of

which we were immediately aware could not be

rnaterial things. But the consequenceof this would

bc that such objects as pens and books and tables,

rrs hey are ordinarily conceived, could not in virtue

of this definition any longer be counted as material

things. And it was with objects such as these that

thc argument which the definition is supposed to{:rvewas primarily concerned.




shown that the ground on which it is maintainedthat there are at any rate some occasionson whichwe perceive sense-datawhich are not parts of anymaterial things is that someperceptions are delusiveland the ground on which it is maintained that someof our

perceptions must be delusive is that if wetake them all to be veridical we shall involve ourselvesin contradictions, since we shall have to attributeto material things such mutually incompatible pro_perties as being at the same time both green andyellow, or both elliptical and round. nut here it

f see as green, the proposition that our perceptions


are both veridical is not contradictory in itself. The

contradiction arisesonly when one adds the assump-

tion that \ryeare both seeing he samematerial thing.

Nor again, in the case where one puts both one's

handssimultaneously into a vesselof water and finds

that it seenrscold to one hand and hot to the other,

is it self-contradictory to hold that both these

cxperiencesare veridical. To obtain the contradic-

tion it must also be assumed that the temperature

of the water with which the vessel is filled is really

thc same throughout. And so it goes with all the

examples, ncluding those in which the experience

of onesenses supposed o contradictthat of another,

and those in which our perceptions are held to becompletely hallucinatory. For instance, if a stick

that looks crooked at the same time feels straight,

it is not logically impossible that both these percep-

tions should be veridical. It becomesso only if one

makessuitable assumptions about the correlation of

the domains of sight and touch. And in the case

of the mirage, it is only if we make suitable assumP-

tions about the duration of materialthings and their

accessibility o different observers hat we are obliged

to hold that the treesand the water which the observer

in question thinks he is perceiving do not really

cxist.

But now let us supPose hat instead of recognizing

thcse alleged contradictions and attempting to evade

thcm by saying that we immediately experience

ncnse-datawhich are not parts of material things,

wc choose ather to deny the supplementary assumP-



t6 FOUNDATIONS OF EMPIRICAL KNOIVLEDGE r

tions which are required for the contradictions toresult. None of--these assumptions is logicallynecessary,so that if they are to be validated iimusibe on empirical grounds. What then is the natureof the empirical evidence upon which they rest ?

The answer s that it is in"u"ry

instance a matter ofour being able to establish a certain order among ourexperiences. We

laythat an object seen n a looilirrg_

glas,ss not really in the place in which it"pp..r.

iobe, because,when we go to that place, we find thatthere is no such object there to be seenor touched.We say that a penny which appears o havea difierentshape when it is seen from a different angle has notreally

changed its shape, because, when we returnto our original point of view, we find that it looks thesameshapeas t did before. We say that a man whoseesasred what we seeas green s neverthelessseeingthe sameobject asourselves,becauseof the obrerv"ltions that we make concerning his behaviour; for

|*"TOt.,we may have the experienceof hearing him

descnbe it, apart from its colour, in the ,"*" *auas we should ourselves,or if we ask him to touch

iiwe may seewhat appears o be his finger coincidingwith an appearanceof the object in our own visualfields. We say that the temperature of the waterthat feels hot to one hand and cold to the other isreally uniform, because it is observed to yield auniform reading on a thermometer. We say, in theinstance of the mirage, that the trees do nlt reallvexist, becausewe believe that people who *"r" i.

whatwe shouldcall anormalphysiologicalstatewould


rrot perceive them and because hey cannot sub-

nccluently e perceived by the observerhimself.

Now, having described the nature of the evidence

that is ordinarily thought to be sufficient to establish

thcsevarious assumptions, wish to considerwhatwould be the position of onewho, though he acknow-

lcdgcd the particular facts about our experiences hat

constitute this evidence, still chose o deny the pro-

positions about material things that these facts are

supposed o prove. One may maginehis saying, or

cxample, that the fact that the shapeof the penny

still appears the same when the observer returns to

his original point of view doesnot prove that its real

shapehas been unchanged; for it might be the case

that the shape hat it originally appeared o have was

in reality altered and then regained. Or again, he

might say that the fact that I observe another person

behaving in the way I should expect him to behave

if he were seeing the same object as I am myself

doesnot prove that he reallyis seeing he sameobject;

it may show that the structure of his world is in a

large measure similar to that of my own, but it doesnot follow that its contents are the same. And I

have no doubt that by postulating a greater number

of material things and regarding them as being more

variable and evanescent than we normally do, it

would be possible to deal with all the other cases

in a similar way. How then is one who holds this

position to be refuted ? The answer is that so long

as we persist in regarding the issue as one concerninga matter of fact it is impossible for us to refute him.



18 FOUNDATIoNS oF EMPIRICAL KNoWLEDGE I

We cannot refute him, because, s ar as he factsareconcerned, here is really no dispute betweenus. Ithas been assumed hat he agreeswith us about thenature ofthe sensibleappearances; and no evidenceof any other kind is or can be available. In what

then doesour disagreementconsist It consists nthe fact that he refuses o describe he phenomenain the way in which we describe hem. Where we

conunon. But the facts to which these expressionsare ntended to refer are n either case he same. Inother words, we are not disputing about the validityof two conflicting sets of hypotheses,but about thlchoice of two different languages. The disputeappears o be concernedwith a matter of fact becauseour opponentuses he same orm of words as we useourselves; but since he empirical evidencewhich we

regard as ustifying the employment of thesewords isnot so regardedby him, it is to be inferred that heis assigning to the words a different meaning fromthat which we have given them. Assuming that hislanguage is self-consistent, we may still criticize iton the ground that it is cumbersome and incon-venient. And we may argue in support of someother terminology, such as the terminology of sense-data, that it involves a less radical departure

from


rr,,rmalusage. But at the same ime we must recog-

rrizc that the superiority of the language that we

lrrvour, if it is superior, does not consist in the

cxprcssionof any truth that our opponent disregards

or tlcnies. For if there is here to be any question of

trrrth or falsehood, here must be somedisagreement

llruut the nature of the empirical facts. And in this

('rscno such disagreement xists.

3. THn lNtnooucuoN oF SnNsr-Plte

'I'he conclusion that I have now reached is that

irr order to account for our perceptualexperience, t

is not necessary o maintain that any of our perceP-tions are delusive. And in saying this I am not

ruttcmpting to deny the empirical facts which the

proposition that someof our perceptionsare delusive

inrordinarily understood to exPress. I am pointing

orrt mcrely that this is not the only way of describing

tlrcrn. I have argued that these acts do not render

it irnpossible o hold that what is perceived is always

n rnaterial thing, provided that one makes suitable

nssumptions bout the constitution and behaviour of

tlrc material things in question. But since these

lssumptions are not factual but linguistic in charac-

Ir:r, the effect of making them is to give to the

cxprcssion " material thing " a different meaning

Ironr that which is ordinarily given to it. I have,

tlrcrcfore, not yet answered the question that I

originally set out to discuss. For I have not yet

rrrcccededn showing hat it is logically possibleboth




to admit the facts that are ordinarily expressedbysaying that some perceptions are delusive, and alsoto maintain that, in the caseof such perceptions, wedirectly perceive material things, in the ordinarysense. It may be granted that the argument fromillusion doesnot prove that it is necessary,n describ-

ing our perceptual experience, to use any word inprecisely the sameway as philosophershaveproposed

to use the word " sense-datum . But does it notprove tJratwe need to make at any rate some altera-tion in our ordinary way of speaking if we are to beable to describe all the empirical facts ?

I do not think that even this limited claim can bejustified. It is indeed true that, if we restrict our-

selves o using words in such a way that to say of anobject that it is seen or touched or otherwise per-ceived entails saying that it really exists and thatsomething really has the character that the objectappears to have, we shall be obliged either to deny

,that any perceptions are delusive or else to admitthat it is a mistake to speakas if the objects that weperceived were always material things. But the fact

is that in our ordinary usagewe are not so restricted.Thus, to return to the familiar examples, if I saythat I am seeinga stick which looks crooked, I do notimply that anylhing really is crooked, or if I say thatsomeone is feeling pressure on his leg, I do notnecessarily exclude the possibility that his leg hasbeenamputated, or if, being subject o an illusion ofdouble vision, I say that I am perceiving two pieces

of paper, I need not be implying that there really

r TI{E ARGUMENT FROM flf.USfON 2r

ilr'e wo piecesof paper there. But surely, it may be

nnirl, f the two pieces of paper really are perceived

llrcy must both exist in some sense,even if not as

rrrntcrial hings. The answer o this objection s that

it is basedon a misunderstandingof the way in which

I lrn using the word " perceive . I am using it

lrt'rc n such a way that to say of an object that it is

lrcrcciveddoesnot entail sayingthat it exists n any

xerrscat all. And this is a perfectly correct and

lnrrriliar sageof the word.

lf there is thought to be a difficulty here, it is

pe:rhaps ecause here is also a correct and familiar

rrnugc f the word " perceive", in which to sayof an

olrjcct that it is perceiveddoescarry the implication

t utt t exists. If I decide o use he word " perceivein this sense, cannot then describemy illusion of

rlorrblevision by saying " I perceived wo pieces of

l)irpor but there was really only one piece there ".

lrrstcad may say " f thought I perceived wo pieces

ol'paper but I was really perceiving only one ". At

tlris point someonemay be tempted to ask " Which

wa.*rt that you really perceived ? Two pieces of

l)lpcr or one ? " and to imagine that he is raising atlrrcstionof fact. But this would be a mistake. For

tlrc fact which I am purporting to describe is the

Hrnrc, whichever method of description I choose.

Assuming hat the proposition which I am intending

to express s true, all that is needed to settle the

tprrcstion hether I did or did not really perceive wo

picccs of paper is that we should decide what we

rrrcan,by perceiving ". If the word is used n one



22 F-OUNDATIONS OF EMPIRICAL KNOWLEDGE I

familiar sense,t can be said that I really did perceivetwo piecesof paper. If it is used in another sense,which is alsosanctionedby convention, hen it mustbe said that I perceivedonly one. In the senseof" perceiving" in which it is impossible to perceive

what does not exist the word is not applicable toexperiences hat are existentially delusive. In thissense did not perceive wo piecesof paper ; I onlythought that I perceived hem. What I really per-ceived was one piece of paper, if only one pieceexisted or if it be assumed hat my experiencewastotally hallucinatory, I did not really perceive any-thing at all. I was ndeed having an experience hat

could properly be described as perceptual in onesenseof the word. But in this sense here is noobjection to my saying hat I really did perceive wopiecesof paper,even f they did not both exist. Thereis no problem so long as one keeps he two usagesdistinct.

The sameambiguity is to be found in the way inwhidh we describeperceptions hat are qualitativelydelusive. For example,a

man will say that he seesa distant star which has an extension greater thanthat of the earth ; but if he is asked o describewhatit is that he is actually seeing,he may say that it is asilvery speckno bigger than a sixpence. Now, sinceit is impossible that the star should both be biggerthan the earth and at the same ime no bigger thana sixpence,one is tempted to conclude that one atleastof theseassertionss false. And, in fact, manyphilosophers would say that the man was mistaken


irr tsscrting that he saw the star. But this is to

iylrrorche fact that the word " see , like the word" pt'rceive , is commonly used n a variety of senses.

Ir r one sense, he sense n which the man ca n say

tr.rrly hat he sees he star, t is necessaryhat what

in sccn should really exist, but not necessary hat itr,lrorrld ave he qualities hat it appears o have. In

rrrrolhcr ense,which is that in which the man can

rlry truly that what he sees s no bigger than a six-

l)('ncc, t is not possible that anything should seem

to lurve qualities that it does not really have, but

rrlrurnot necessary hat what is seen should really

r,xist. If, in our example, t is assumed hat the man

is rrsing he word " see consistently , hen it doesli rlk w hat at leastone of his statementss empirically

l';rlst:. But it is also open to us to accept both his

rrtrrtt:rncntsf we assume,as indeed we ordinarily

rrlrorrld,ha t he has slipped from one to the other

uri:rgc f the word " see". The important point to

rrotit:c s that we do not require the two usages n

olrlt'r to describe he facts. Let us suppose hat, in

orrrcxample, he man decideso use he word " see

,,rrly n the sense n which he is justified in saying

tlrlt lrc seeshe star. How, then, s he to expresshe

l;rt:l that what he sees n the other sense s a speck

rr,r rigger than a sixpence Simply by using the

lrlrrirsc appears o be " insteadof " is ". And this

tcr.rrrinologyf " appearing is one that we do very

,ltt'n use in describing facts of this sort. That we

,rrc lccustomed also to describe them in another

rr';ry, vhich involves a different usageof words like




"-see" and " perceive ',is not ordinarily a source

of confusionto us, becausewe are able to tell fromthe context what is the sense n which such wordsare intended to be understood. But it has misledphilosophers, as their use of the argument from

illusion shows.

words in other, more familiar, senses. But it is thisusage hat leads them to the introduction of sense_

saying that this sense-datum s perceived,they saythat it is sensed. But if in the caseof a delusivlperception one is gging to say that one is sensinga sense-datum, hen, inasmuch as the distinctioibetween delusiveand veridical perceptions s not adistinction of quality, it is convenient o extend thisusage o all cases. That is to say, he contention hatif theseperceptionsare not qualitativelydistinguish_


rble the objects perceived must be of the same ype,

which I saw no reason for accepting when it was

trcatcd as a statement of fact, can reasonably be

lcceptedas a rule of language. And thus one arrives

rrt the conclusion hat in all casesof perception the

objectsof which one is directly aware are sense-data

irrrdnot material things.

This procedure is in itself legitimate; and for

ccrtain purposes t is useful. I shall indeed adopt it

rrryself. But one must not suppose hat it embodies

rrny factual discovery. The philosopher who says

tlrat he is seeinga sense-datum n a casewhere most

pcople would say that they were seeing a material

thing is not contradicting the receivedopinion on any

rluestionof fact. He is not putting forward a new

hypothesiswhich could be empirically verified or

confuted. What he is doing is simply to recommend

:r new verbal usage.r He is proposing to us that

insteadof speaking, or example,of seeinga straight

stick which looks crooked, or of seeing an oasis

when there is no oasis there, we should speak of

sccing a sense-datumwhich really has the qualityof being crooked, and which belongs to a straight

stick, or of seeing a sense-datumwhich really has

llrc qualities hat are characteristic f the appearance

ol'an oasis,but doesnot belong o any material hing.

ll ' we accept this recommendation it will not be

lrt'causeour ordinary language s defective, in the

s('r)sehat it does not furnish us with the means of

' This point has been clearly brought out by G. A. Paul in hisl)rfpcr on " Is there a Problem about Sense-data ?", AristotelianS oc e y Supltlementary Proceedings, 1936,



26 FOUNDATIONS OF EMPIRICAL KNOWLEDGE I

describingall the facts,or in the sense hat it obligesus to misdescribesomeof them; but simply becauseit is not so good an instrument as the sense-datumlanguage for our special purposes. For since inphilosophizing about perception our main object is

to analyse he relationshipof our sense-experiencesothe propositionswe put forward concerningmaterialthings, it is useful for us to have a terminology

that enables us to refer to the contents of ourexperiences ndependently of the material thingsthat they are taken to present. And this the sense-datum languageprovides.r It has also he advantage

of laying down an unambiguousconvention for theuse of words that stand for modesof perception,and

so freeing us from the verbal problemsthat develop,as we have seen,out of the ambiguoususe of suchwords n ordinary speech. We may admit, therefore,

that there are good grounds for adopting this tech-nical terminology. But in adopting it we mustrealize hat it does not in itself add to our knowledge

of empirical facts, or even make t possible or us toexpressanything that we could not have expressed

without it. At the best t enablesus onlyto refer tofamiliar facts in a clearerand more convenientway.

The main proposition which the argument fromillusion, as I have interpreted it , was supposed oestablishwas that what we see,or qtherwisedirectly

experience, renever material things, but only sense-data. This conclusion I accept, but not as a pro-

t Not fully, indeed, but to an extent that is sufficient for ourpurpose. Vide Pa:r' V of this book.

THE ARGUMENT FROM ILLUSION

position in the ordinary sense. A way of making

this distinction clear is to contrast the sentence " I

ncver seematerial things but only sense-data with

$ome sentence of similar appearance that does

cxpressa proposition about a matter of fact. Con-

sider, for example, the sentence I never see gold

sovereigns ut only Bank of England notes". The

proposition which this sentence expresses s one

whose validity is subject to empirical tests. My

prcsent experienceconfirms it, but at the same ime

I can imagine having perceptions that would con-

l\rte it. My experiencewould be different if it were

I'irlse. But when I say " I never see material things

lrrrt only sense-data I am saying something the

truth or falsehood of which makes no differencewhatsoevero the nature of my experience. Indeed

it is misleading o speakof truth or falsehoodat all

in this case. For if we allow ourselves o say that

thc sentence f never see material things but only

scnse-data expresses ither a true or a false pro-'grosition,we thereby suggest hat it is on a level with

scntencesike " f never seegold sovereigns ut only

llank of England notes "; that its validity depends

rrponempirical facts. But I haveshown hat between

:r philosopherwho says hat he seesonly sense-data

lnd one who says hat he seesmaterial things there

is no disagreement about any matter of fact. If they

:rppeal o the facts at all it can be only to show that

onc form of expression s more convenient han the

other. Thus, the sensen which my experience ives

rrrc ustification for saying that I seeonly sense-data

27



zB FOUNDATIONS OF EMPIRICAL KNOWLEDGE r

is quite different from the sense n which it givesme justification for saying that I see only nank ofEngland notes. In the latter case t is a questionofthe verification of an empirical proposition. In thecaseof the sense-datat is a question of there being

extraneous grounds for preferring one method ofdescription to another, which is equally true to thefacts. And this shows that there is a genericdifference in the meaning of the two sentences nspite of the similarity of their appearances. fn onecase we are expressing a proposition about anempirical matter of fact ; in the other casewe areexpressinga resolution about the usage of words.

We may appeal to experience o show that such aresolution is sensible or foolish, but we cannotproperly claim that it is either true or false.

4. NlrsusnsoF THEAncuurur FRoM r-rusroN

But is this al l that the argument from illusionyieldstrs ; a motive for adoptinga new terminology ?

It seemsa meagre result when one considershowmuch importance hasbeenattached o the argumentby writers on the theory of knowledge. fn one formor another it has been used to support such con-clusions as that the world of sensiblephenomenais self-contradictory;' that our ideas of secondaryqualities are not resemblancesof.any real qualitiesof material things ; which means, nter alia,ihut ,ro

material hing is literally coloured,or literally charac_terizedby any sensible emperature or weight; that


the primary qualitiesof solidity, extension,number,

ligure and motion are also " not real " ; that the

tcstimony of the sensess not to be relied on, and

that if we have any acquaintance with material

things as they really are,it is not through any act of

scnse-perception ut only through some " intuitionof the mind ". But all such uses of the argument

are invalid, as I shall now proceed o show.

I may begin by pointing out that if the assertion

that the world of sensible phenomena is self-

contradictory is taken literally, no argument can

possibly prove it, for the reason hat the notion of

sclf-contradiction s not applicable o the phenomena

thcmselves. It doesnot make sense o sayof differentphenomenaeither that they do or that they do not

contradict one another ; they simply occur. It is

only with regard to the propositions that we use to

dcscribe the phenomena that the question of con-

tradiction can legitimately be raised. Let us there-

fbre assume hat what the argument from illusion is

supposed o prove is that it is impossible o describe

whatever phenomena occur without sooner or laterfirlling into self-contradiction. But so f.ar from

proving this, the argument does not show even that

runy ontradiction s involved in the propositions hat

wc actually use to describe phenomena. It shows

that we speakof material things as appearing o exist

in some conditions and not in others, or to some

observersand not to others, and that we speak of

thcm as appearing o have incompatible qualities to




in all this there is no contradiction. It is not self_contradictory to say that a penny that looks roundfrom one standpoint looks elliptical from another,or that a curtain that looks green to me appears osomeother personto be blue. It is not self_contra_dictory to sayboth that a man feels

pain in an ampu-tated leg, and that the leg is not plrceptible to urryother observeror subsequently o the man himself.And the samewould hold good of any other examplethat could be produced.

But this, I may be told, is only because amtreating the appearances smere appearances. Treatthem as realities ; assume hat things really have thequalities hat they appear o have, rrd

"o.rirrdictionsrise at once. As Bradley says, ,. a thing must beself-consistentand self-dependent. It either has aquality or has not got it . And if it has t, it cannothave it only sometimes,and merely in this or thatrelation. But such a principle is the cond.emnationof secondaryqualities ,,,t and, as he later adds, ofprimafy qualities also.

If this argument is intended to apply not merelyto the language hat we

actually ur.-b,ri also to anyalternative language in which we might attempt todescribe he same acts, then I have alreadyreiutedit . For I have shown that a language n which nodistinction is made between things as they appearand things as they really are need not be ,.lf_"orrtr"-dictory. In such a language, a statement that amaterial thing, M, appearedto have a quality, e,

t Appeatance and Reality, pp. rr_r2.

THE ARGUMENT FROM ILLUSION 3r

would always entail that M really did have a, but

this would not lead to contradictions, provided that

the languagealso contained suitable criteria, which

would, of course, be different from the criteria that

we now employ, for determining when a thing

changed ts qualities and when two appearances ereappearances f the same hing. It is true, indeed, hat

if we abolished he distinction that we ordinarily make

between appearanceand reality, and at the same ime

refused o introduce any compensatory onventions,

we should be nvolved n self-contradiction. But why

should one be expected o do this ? The argument

which I am considering doesnot furnish any reason.

Furthermore, is it true even that we do avoid

contradictions only by treating all appearancesas

mere appearances Admittedly, we do not speak

as f every perception were veridical. We use such

sentences s " The curtains look green n this light,

but they are really blue," ; " The penny looks

clliptical from this angle, but it is really round ".

But this does not mean that we speak as if every

perception were delusive. We assume, n fact, that

someof our perceptionsare veridical and others not.And in this there is no contradiction. There would

be a contradiction only if the sameappearances ere

held to be both veridical and delusive in the same

sense and this is not the case. What we actually

do is to define the real qualities of a material thing

in terms of the qualities of certain privileged appear-

ances. Later on it will be shown that this is not an

arbitrary procedure that an appearance s held to




be privileged, n this sense, ecause f certain special

relations that it bears to other appearances, pro-

position which will be stated more clearly in the

terminology of sense-data.r But even if the pro-

cedure were arbitrary, which it is not, there would

still be no ground for saying that it led to any con-tradictions. There is no logical reason why, in

classifying appearances s veridical or delusive, we

should have o include them all in the samecategory.

A more moderate and common use of the argu-

ment from illusion is that in which it is taken to show,

not that the assumption that certain sensiblequalities

literally characterize material things involves any

contradiction, but that it is always false. For this,

recourse s had to what ProfessorPricehas called he

causal form of the argument,z with which I have

already dealt in a slightly different context.3 It is

pointed out that the appearance of these sensible

qualities to an observer depends, as I have shown,

upon other factors than the state of the material

thing which he supposes hem to characteize, among

these factors being the physiological and psycho-

logical condition of the observerhimself. And fromthis it is inferred that the sensiblequalities do not

really characterize the material thing in question,

or at least that no orre has any ground for believing

that they do. To quote Bradley again," A thing is

coloured, but not coloured n the sameway to every

eye ; and except to some eye, it seemsnot coloured

' V'ide Par:"V, Section a4, of this book.

. Perception, p. 27 fr. r p.9.


at all. Is it then colouredor not ? And the eye-

rclation to which appears somehow to make the

.1,rdiy - does that iJseH Possess olour ? Clearly

,ro, .o, unless there is another eye which sees t'

Nothing therefore is really coloured colour seems

only tolelongto what itself is colourless' And the

*o-" ,"..tlt holds again, with cold and heat' A

thing may be cold or hot according o different parts

"i; .ki.r; and without some relation to a skin' it

scems without any such quality' And' by a like

"rg.r^"rrt,the skin is proved not itself to own the

qu""tiy, which is hence possessed y nothing

'it" "tg"*ent

shows everywhere that things have

*".orrdJry qualities only for an orgalr; and that the

urg"r, itr"lf^h", thesequalities n no.other way'" r

This is, substantially, a recapitulation of the

argument by which Locke sought to prove that ideas

uf'.".ond"ry qualities were not resemblancesof any

rcal qualitl.t of material things ; and it has been a

.urrr". of surprise o subsequentphilosophers hat he

did not"*t.,,d

this conclusion o the ideasof primary

q.rutiti". utro. For they have pointed out that it is

"qrrally true of the sensible qualities of solidity'

"*r".rriot,figure, number and motion that " things

havethem orrly fo, an organ' and that the organ itself

has hem in no other *t "' This criticism is justi-

fied ; but we must not iherefore infer that the dis-

tinction which Locke drew between primaty-""1

secondaryqualities' is altogetherarbitrary' For it

' OP - cit- P' tz 'r Vide An Essay concerning Human (Jndnstanding' Book [It

ch . 8 an d ch.23.




is possible o account or it, if we regard t as ormingpart of a causal theory of perception. The under-lying assumption s that we are entitled to ascribe omaterial things only those qualities that are neces-sary and sufficient to causea given observer to havethe sensationshat he does and these, t is held, arethe so-calledprimary qualities. Thus, the view thatthe quality of colour doesnot literally characteize anymaterial thing was maintained, I suggest,not merelybecauseof the argument from illusion, but alsobecause t was thought that sensationsof colourcould be sufficiently accounted for in terms of themotion of particles which had, besidesmotion, theprimary qualities of solidity, number, figure and

extension, but were themselves colourless.r ft is,however, o be remarked hat even f this was a validline of argument it would not prove that materialthings were not literally coloured,but only that wehad no good reason o suppose hat they were; norwould it prove, what Locke appears o have held,that particular ideas of primary qualities were everexact resemblances f the real qualities of any given

material thing, but only that material things werecharacterized by some particular qualities of theprimary kind. It would entitle us to infer, forexample, that a material thing had some quality ofextension,but not that it had precisely hat qualityof which, to use Locke's terminology, we received

I Another, and perhaps the fundamental, source of the distinc-tion between primary and secondary qualities is the identificationof what is real with what is quantitatively measurable. I refer tothis point in the last section of this book.


on any particular occasion a simple idea. It is,

however, unnecessary o elaborate this point, be-

cause he argument s in fact invalid. For it will be

shown ater on I that the causal heory which it pre-

supposess itself untenable.

But let us return to the argument which has been

supposed by Bradley and others to be " the con-

tlcmnationof secondaryqualities ". The assumption

upon which it rests is that if the fact that a thing

appearso havea certainquality is causallydependent

rrpon he nature of the environment and the observer,

it follows that there is no ground for believing that

the thing really has that quality. But if I am right

in supposing that the real qualities of a material

thing are to be defined in terms of the qualities ofsomeof its appearances,he thing itself indeed being

nothing apart from its actual and possible appear-

ances,2hen this assumption s fallacious. For the

questionhow the appearances ome o be manifested

tloesnot arise n this contqxt. Let it be granted hat

the relevant appearanceswould not, in fact, be met

with, unlesscertain further conditionswere fulfilled ;

there s still no inconsistency n taking theirqualities

to be the real qualities of the material thing. It is

true indeed that when we speak of a thing's really

having certain qualitieswe imply that it retains hem

cven when, through the absence of the requisite

causalconditions, t is not actually appearing o have

tlrcm. But all that this involves is that the hypo-

thetical propositions, which assert hat the appear-

t Part [V. t Vi.dePart Y.



36 FOUNDATIONS OF EMPIRICAL KNOWLEDGE t

ances would be manifested if the conditions were

fulfilled, remain rue evenwhen their protases appen

not to be realized. It is true also, as Bradley says,

that " if the qualities mpart themselves ever except

under conditions ", we cannot say " what they are

when unconditioned ". But the answer to this isthat when we assert that a given quality really

characterizessome material thing we are not assert-

ing that it is unconditioned, n this sense. No doubt

the argument would be valid if we conceived of

material things as unobservable things-in-them-

selves. But we do not conceiveof them in this way,

and there is no good reasonwhy we should.

In considering the use of the argument fromillusion as a ground for distrusting the deliverances

of sense-perception may take as a text the famous

passagen the Meditations of Descartes n which he

explains how a wide experienceby degreessapped

the faith which he had reposed n his senses. " For

I frequently observed , he says, " that towers

whicli at a distance seemed ound, appearedsquare

when more closelyviewed, and that colossal igures,

raised on the summits of these towers, looked like

small statues, when viewed from the bottom of

them ; and in other instances without number, I

alsodiscoverederrors n judgements ounded on the

external senses; and not only in those founded on

the external, but even in those-that rested on the

internal senses; for is there aught more internal

than pain ? And yet I have sometimes been in-

formed by parties whose arm or leg had been


irrrrputatedhat they still occasionally-seemedo feel

1,,,i., ., that part of the body which they had lost' a

.'ircumstan.ethat led me to think that I could not

lrc quite certain even that any one of my members

was afiected when I felt pain in it' And to these

grounds of doubt I shortly afterwardsadded two

,,th.r. of very wide generality ; the first of them was

that I believed I never perceived anything when

irwakewhich I could not occasionally hink I also

pcrceivedwhen asleep,and as I do not believe that

thc ideas seem o perceive n my sleepproceed rom

olrjectsexternal to me' I did not any more observe

,,,',ygrot.d for believing this of such as I seemto

p"r.Jirr" when awake;the second was that since I

*o, u, yet ignorant of the author of my being' or

:r t least srrppot"d myself to be so, I saw nothing to

prevent my having been so constituted by nature as

ttrot I should be deceivedeven in matters that ap-

pcared o me to possesshe greatest ruth' ' ' ' And

,rlthough the perceptions of the senseswere not

tlcpenJent on my will, I did not think that I ought

on that ground to conclude hat they proceeded rom

rhings Jifierent from myself, since perhaps there

,nigit be found in me somefaculty, though hitherto

rrnknown o me, which produced them'" t

It is true that Descartes himself does not abide

by these conclusions. But that is only becausehe

rhinks that he is able logically to demonstrate he

cxistenceof God, and that the assumption that the

,Medi tat iotuontheFirstPhi losophy,Medi tat ionVl 'p'r3rof

thc Everyman edition.

38 FOUNDATTONS OF EMPIRICAL KNOWLEDGE rI THE ARGUMENT FROM ILLUSION 39



teachings of the sensesought in general to be dis-trusted argues a deceitfulness in God which wouldbe inconsistentwith his goodness. And it is on this

varretrescorresponding o them, although, perhaps,not in reality like them',.2 His reasoni"g ut tiri.

point is not, in my opinion, valid ; but we are not

t Op . cit. p. r35. t Op,

it as round. When he came close to it and found

that it then looked square, he would discover that

" his senseshad deceived him ". But the reason

why he was deceivedwould be that he had assumed

that because n object presenteda certain appearance

in one set of conditions, it would still present the

same appearance n another. In this particular

instancewe should say that the man's assumption

was rrational as well as mistaken, nasmuch as t ran

counter o our experienceof the way in which things

tlo alter their appearances.But even f the assump-

tion had been rational, in this sense, t might still

havebeen false. And the samewould apply to any

otherudgement

that involved an inference,whethert:onsciousor unconscious, rom past to future ex-

pcrience. That is to say, the fallibility of the

$cnses s simply an instance of the fallibility of all

inductive reasoning. It consists n our being unable

lirrmally to deduce the conclusion of an inductive

:rrgument rom its premises I and this is a necessary

lirct, n the sense hat the propositionwhich expresses

it is analytic. For an argument the premises ofwhich did formally entail the conclusion would, by

rlcfinition, not be inductive. What is contingent is

thc fact that some of the inductions that we make are

t In speaking of " the premises of an inductive argument " Irrrcrn here to refer only to a set of singular propositions. I admittlurt the particular conclusion of such an argument may followlirrmally from its premises if these are taken to include some generalproposition. But the point is that this general proposition will not

rtsclf be formally deducible from any set of singular propositionsrlurt refer to the particular empirical facts in which alone the




actually unsuccessful; but this does not affect thelogical status of inductive reasoning. We may sayindeed that the probability of the conclusionswhicirwe reach by inductive methods would, in general,be higher if these methods had never yet actuallyfailed us. But the difference that this would maklto the reliability of our judgements would be adifference of degreeand not a difference of kind.

Exactly the same remarks apply to the errors injudgements " that rest on the internal senses,.The mistake that is made by the person who feelspain in an amputated limb is that of assuming thatbecause such sensations of pain have previously

been correlated with further tactual and visualexperiencesof a certain kind, the correlation musthold good in this casealso. When he subsequentlylearns that these further experienceswere not, infact, obtainable, that in what would be described asnormal conditions his leg would not have beenvilibJg or tangible at the time that he was feeling thepain, he realizes hat his internal senses avedeceived

him. But hereagainhis error is an error of inference,and one to which he would be theoretically subjecteven if he were never actually deceived. perhaps itis the actual occurrenceof such illusions that firstcalls our attention ,to the fallibility of perceptualjudgements ; but it is not on this fact that th.i,fallibility logically depends.

The caseof dreams s not quite so simple. The

argument here is that inasmuch as we often believe

I THE ARGUMENIT FROM ILLUSION 4I

:rrc n fact dreaming, we can never be sure that we

arc not dreaming. To this many peoplewill be con-

tcnt to reply, with Locke, that " if anyone will be

so sceptical as to distrust his senses,and to affirm

that all that we seeand hear, eel and taste, hink and

do, during our whole being, is but the series and

tlcluding appearance f a long dream whereof there

is no reality ; and thereforewill question he existence

of all things or our knowledge of anything; I must

clcsirehim to consider hat if all be a dream, then he

rloth but dream that he makes the question and

so it is not much matter that a waking man should

answerhim ".t But this rejoinder is more emotion-

ally than intellectually convincing. It is true that

if the man who says that all is a dream intends

thereby to assert hat all the propositions that any-

body ever expressesare false, he cannot without

contradiction nclude in this totality of falseproposi-

tions the proposition which he is himself expressing.

'I'o suppose hat he did so,include it would be to

cxtend o h is case he familiar paradox of Epimenides

the Cretan, who said that all Cretans were liars. It

is, however, possible o deny any given proposition

that is put forward without expressingone's denials

in such a way that a proposition is ever used to

ncgate itself. And in this fashion, assuming that

thc classof propositions denied is not itself incon-

sistent, a universal scepticism can be maintained

without self-contradiction. Moreover, the view I am

I An Essay concerning Human Understanding, Book IV, ch. z,

42 FOUNDATIONS OF EMPIRICAL KNOWLEDGE r I THE ARGUMENT FROM ILLUSION



discussing s not that we are n fact living in a dream,but only that we cannot be sure that we are not;and to this Locke's objection does not apply. Noris it satisfactory to say, as Locke also does, that" there is a very manifest difference between dream-ing of being in the fire and being actually in it ".For it is not because hey are intrinsically differentfrom the sensations hat we have when we are awakethat we pronounce our dream sensations to bedelusive,but rather because hey do not fit into thegeneral order of our experience. One remembersmaking certain perceptual udgements which were inaccord with the experiencesone was having at thetime, but are not corroboratedby one'spresentper-ceptions and therefore one says that one musthave been dreaming. But can I use a test of thiskind to show that I am not still dreaming now ?Does it prove anything more than that, if I amdreaming, the character of my dream is such thatit is possible to discover among the experiencesw6ich constitute it a distinction analogous tothat which is ordinarily drawn between veridicaland delusiveexperiences It does ndeedsometimeshappen that, in the course of a dream, one actuallyraises the question whether the perceptions one ishaving are veridical, and decides, reasonably butfalsely, that they are. Can I then be sure that theexperiences hat f am now having are not part of adream of this kind ? And is there any reason whysuch a dream should not be indefinitely prolonged

?The answer is that if what is here meant by

43

" being sure " is " being able to give a conclusive

tlcmonstration , then it is true that I cannot at any

tirne be sure that I am not dreaming. I am able in

fact to convince myself that I am not, by putting my

pcrceptual udgements to the test of further experi-

cnce and finding that they are substantiated. But

since there is no theoretical limit to this processof

tcstingr t is ahvays ogically possible hat I am mis-

taken. However many favourable ests I may make,

the possibility still remains that my subsequent

cxperienceswill consistentlybe such as to make me

conclude hat the perceptions hat I had to my own

satisfactionproved to be veridical were not so really,

:rndthat I was

dreaming after all. Thisquestion oftlreams s, indeed, only a special caseof the general

problem that Descartes aises. Can we, in virtue of

our sense-experiences,ver be sure of the truth of

any proposition that implies the real existenceof a

rnaterial thing ? Again the answer s that if what we

rcquire to make us sure is a logical demonstration,

then we cannotever be sure. Descarteshimself con-

lirses he question by putting it in terms of a causalthcory of perception. He assumes hat what we are

:rssertingwhen we asserta proposition implying the

t:xistenceof a material thing is that our perceptions

rrrcproducedby material hings, which arethemselves

not definable n phenomenal erms. But in fact, as I

shall argue later on, this is not what we ordinarily

rncanby such propositions ; and, if it were, not only

should we not be able to demonstrate heir truth;wc should not even have any good inductive grounds



44 F'OUNDATIONS OF EMPIRICAL KNOWLEDGE r

for believing them.l However, even if we correctthis mistakeof Descartes's nd interpret propositionsimplying the existence of material things in a

phenomenalistic ashion, the result, as regards the

possibility of their demonstration, s the same. We

do indeedverify many such propositions o an extentthat makes it highly probable that they are true ;but since the series of relevant tests, being infinite,

can never be exhausted, his probability can never

amount to logical certainty. In this case also, the

empirical fact that we sometimeshave good reason

to suppose that we actually are deceived is logically

irrelevant to the argument. For even if all the pro-

positions in question had hitherto been uniformlyverified, this would not remove the possibility thatfuture testswould give a consistentlynegative esult.It would still further diminish the probability of this

hypothesis,but with that I arn not now concerned.

It must be admitted then that there is a sense n

whic.|r it is true to say that we can never be sure,

with regard to any proposition implying the exist-

enceof a material thing, that we are not somehowbeing deceived ; but at the same time one may

object to this statement on the ground that it is mis-

leading., It is misleading because t suggests hat

the state of " being.,sure is one the attainment of

which is conceivable, but unfortunately not within

our power. But, in fact, the conception of such a

state s self-contradictory. For in order to be sure,I Part IV.

. Cf. John Wisdom, " Philosophical Perplexity ", ProceedingsoJthe Aristotelian Society, 1936-7,


in this sense, hat we wese not being deceived,we

should have to have completed an infinite series of

vcrifications ; and it is an analytic proposition that

one cannot run through all the members of an

infinite series. Nor would the situation be different

if our beliefsconcerningmaterialhingswere ounded,llot upon sense-perception, ut rather, as Descartes

suggested,upon some " intuition of the mind ".

Iior we could not conceivably have the means of

tlcmonstrating that any such intuition was infaltble.

Accordingly,what we should say, f we wish to avoid

rnisunderstanding,s not that we can never be certain

that any of the propositions n which we expressour

pcrceptual udgements are true, but rather that the

notion of certainty doesnot apply to propositionsof

this kind. It applies to the a priori propositionsof

Iogicand mathematics,and the fact that it doesapply

to them is an essential mark of distinction between

them and empirical propositions. But once this is

rccognized,he motive for scepticismhas gone. The

' rnistake of the sceptic is not that he maintains any

ftrlsehood,or doubts some self-evident truth, but

that he insists on applying to one classof proposi-

tions a criterion that is appropriate only to another.

'l'he positive result of his argument is to call our

irttention to the fact that the relationship of proposi-

tions describing the contents of our sense-experi-

(:nces to propositions implying the existence of

rnatcrial hings is not that of premises o conclusion

in a deductive argument. Our procedure in such

.'ascs s alwavs inductive, and it remains inductive



46 FOUND.ATIONS OF EMPIRICAL KNOWLEDGE r

however much sensible evidence we may accumulate.

This may indeed be regarded as a truism; but it is

all that the argument I have been examining shows.

5.

THEoRTES F PrncnprroN As ALTERNATTvE

Lencuecrs

The final use of the argument from illusion which

I have to consider is that which is supposed o refute

the perceptual theory of Naive Realism. And for

this I cannot do better than refer to Professor Price's

book on Perception.' According to him, what is to be

understood by Naive Realism is the thesis that visual

and tactual sense-data are parts of the surfaces ofmaterial things ; which, since he defines a sense-

datum as that which is immediately given in per-

ception, is equivalent to the thesis that it is always a

material thing that we see or touch. And he main-

tains that if the argument from illusion proves

nothing else, it does at least refute this thesis. But

this is a question with which I have already dealt ;

and I have shown that the answer depends upon theway in which the thesis s interpreted. If one assumes

that the naive realist is using the words " see" and

" touch " in such a way that to say of an object that it

is seenor touched doesnot entail sayingthat it exists

or that it really has'the properties that it appearsto

have, then it is clear that the argument from illusion

cannot be used to refute him. If, on the other hand,

he is using these words in a way that does carry these

r THE ARGUMENf FROM ILLUSION +7

implications, then the argument does refute him, so

long as he adheres n other respects o our ordinary

usage of words. But if he is prepared to modify

this usage and introduce suitable conventions with

regard to the variability of material things, then, as

we have seen, it is possible for him consistently tomaintain his position without running counter to the

empirical facts.

A third possibility, which is suggestedby Price's

formulation of the thesis, is that it should be in-

terpreted as a theory about the analysis of material

things in terms of sense-data. According to this

interpretation, what the naive realist is asserting is

that " ,4 is perceiving ff ", where r stands for amaterial thing, can be analysed nto " I is sensing a

sense-datumand knowing it to be part of the surface

of tc ", And this can indeed be refuted by means of

the argument from illusion. For we can say to the

naive realist : " You say that perceiving a material

thing can be analysed nto sgnsing a sense-datumand

knowing that it is part of the surface of a material

thing. But is it not significant to say that the sense-

datum by means of which I is perceiving a coin is

round, and that the sense-datum by means of which

B is perceiving the same coin is ell iptical ? And

would it not be self-contradictory to say that a part

of the surface of a coin was both elliptical and

round ? "

As an argument ad hominemthis may be decisive.

But all that it proves is that the naive realist's thesis

48 FOUNDATIONS OF EMPIRICAL KNOWLEDGE r I THE ARGUMENT FROM ILLUSION 49



datum language so that if we refute him by in-

terpreting his thesis in terms of sense-datawe are

begging he questionagainsthim. It is indeed mis-

leading to speak of refutation at all in this context,

unless we assume that the naive realist is merely

denying that there are illusions; and I do not think

that any philosopher has really intended to denythis empirical fact. For my argument has shown

that if we do not make this unwarranted assumption

we have to regard the thesis of naive realism, not asan assertion of fact, but asa proposal to usewords ina certain fashion; and the same is true, as we have

seen,of the thesis hat we experience ense-data. Ifwe accept the sense-datum erminology, then wemust reject the terminology of naive realism;

forthe two are mutually incompatible. But this is notto say that we regard the theory of sense-dataastrue and the theory of naive realism as false. For Ihave shown that these so-called theories of percep-

tion q,renot theories at all in the ordinary sense,andtherefore that the notions of truth and falsehood,

which we apply to scientific theories, are not appli-cable o them.

To illustrate further this important point thatwhat philosophers call theories of perception arenot theories at all, in.,thescientific sense, may referonce again to Price's book. In a chapter which heentitles " Some Modifications of Naive Realism ",he examines " three ways in which philosophers

have attempted to modify the Naive Realist thesis,so as to make it defensible ". These he calls " The

'fhcory of Multiple Location ", which he attributes

to ProfessorWhitehead," The Theory of Compound

Things ", which he attributes to Professor Alex-

ander, and " The Theory of Appearing ", which he

attributes to Professor Prichard and Professor G. E.

Moore. The hypothesiswhich is said to be contained

in the Theory of Multiple Location is that, besidesthe properties " which characterize it simpliciter ",

a material thing also has properties " which char-

acterize t only from a place " ; and these properties

belong to it just as ruly and objectively as he others.

" The penny just rs elliptical from this and that place

and kinked from such and such others; exactly as

in itself (and from certain specially favoured places)

it is circular and smooth in outline."' To this

Price raises the objection that the theory does not

meet the casesof double vision and complete hallu-

cination. He maintains that it will not do to say

that a thing is doubled from a certain place, " for

'doubleness' is not a quality at all"., Nor is

" hallucinatoriness " a quality. When we say that

a perception of a material thing is hallucinatory, we

are not saying that the thing either has or appears o

have a peculiar property ; what we are saying is thatthere really is no such thing. The same objections

are held to apply to the Theory of Compound

Things, according to which " illusory visual and

tactual sense-datado really form parts of the surface

of an object, but of a compound object. The stick is

not bent, but the compound stick-plus-water really is

, Perceftion, p.55-6. ' P.52.

EMPIRICAL KNOWLEDGE r



so FOUNDATIONS OF

bent, and the crooked sense-datum is part of its

surface. These compound objects really do exist in

external Nature and do have their qualities just as

' simple ' objects ike sticks have theirs." ' Professor

Price remarks that, to be consistent, the exponents

of this theory ought to include among the con-

stituents of their compound objects " the sense-

organsand brain of the observer ;" for they also

affect the nature of what is experienced and in that

case every object that we see or touch will be com-

pound, in the sensen which this word is here being

used. But is this a proper sense? asks Professor

Price. Surely all that there really is, is a group of

causally related objects. If they formed a compound

" it ought to be possible to point to at least oneattribute possessedby all of them in common and

not by any of them in isolation - for instance, a

characteristic shape or mass or inertia, or some such

causal property as serves to distinguish a chemical

compound from other compounds and from its own

constituents. But no such attribute can be found." r

And finally, he objects that even if there were such

compound objects, thetheory would not save Naive

Realism; for they would be very different from the

material things that we ordinarily suppose that we

observe. i:

With regard to the Theory of Appearing, we are

told that it consists n taking what we seeor touch to

be always " a part of an object's surface appearing

to someoneto have certain characteristics. On one

t lb id.58-9. ' P.59. ! P.6r .

I THE ARGUMENT FROM ILLUSION 5t

form of the theory ' appearing' is the name of a

unique and unanalysable hree-term relation between

:r part of an object's surface,a characteristic or set of

characteristics,and a certain mind. On another

tbrm of it,' A appears6 to S ' stands for a unique

and unanalysablekind of fact about /-D-ness and

S's mind ".' ProfessorPrice allows that this theory

can deal with such straightforward phenomena as

distortions of perspectiveand colour ; but he main-

tains that it cannot account for the cases,such as

those of double vision and complete hallucination,

where there is a transition " from qualitative appear-

ing to existential appearing". For in these cases

there exists no material thing to be a term of the

unanalysable elation, or an ingredient in the un-analysableact.

Now the point which here again seems o have

cscaped oth the exponentsof these" theories" and

their critic is that, as the word is ordinarily under-

stood, they are not alternative theories at all. This

rnay become clear if I employ the method of com-

parison. Suppose, or example, hat we are looking

fbr a theory to account or the nature of our dreams.'fhen one hypothesiswhich we may consider s that

they correspond o physical stimuli ; another s that

they are prognostications of future waking experi-

cnces; another, that they are the expressionsof

unconsciouswishes and fears. To decide which is

the most probable we apply them to the empirical

facts. We then find that although there are

I P.62.



sz FOUNDATIONS OF EMPIRICAL KNoWLEDGE t

experiences hat favour the first of these theories, as

when one dreams of being naked and wakes to find

that the blankets have slipped, or when a dream

experienceof the sound of church bells merges nto

a waking experience of the ringing of an alarm-

clock, yet even n these exceptionalcases he theorydoes not cover the details of our dreams; and in

general t provides no explanation for the variety and

diversity of their contents. Similarly, we find that

the correspondences that have so f.ar been dis-

covered between people's dreams and their sub-

sequent waking experiencesare too few and uncertain

to provide a basis for any generalizationon which

we can rely. On the other hand, the empiricalevidence that psycho-analysts have collected in

favour of the third theory is comparatively strong;

and it has been confirmed by the successwith which,

on being applied to the treatment of nervous dis-

orders, it leads to the detection of " unconscious

experierlces . If, therefore, we decide that the

theory that our dreams are the expressions of un-

consciouswishes and fears is the most probable ofthe three, it is becausewe find that we are able to

deduce from it a greater proportion of the known

empirical facts, and becauseby its means we are

able to make more reliafle inferences rom the known

to the unknown.

Pursuing this comparison, let us accordingly

attempt to apply those three so-called theories of

perception, " the theory of multiple location ", " the


appearing , to the empirical facts. Can we discover

any empirical evidence that favours any one of these

theories rather than another, in the sense n which

the evidence put forward by the psycho-analysts

favours their theory of dreams ? The answer is that

the comparison at once breaks down. For we findthat every conceivable experience, in the field to

which these theories refer, can equally well be sub-

sumed under any of them. Each of them will cover

any known fact; but none of them, on the other

hand, enablesus to make any inference at all from

the known to the unknown. No matter which of

them we adopted, we should be able to describeour

perceptions, whatever their nature ; what we shouldnot be able to do would be to make any predictions.

But if the relation of these three theories to the

relevant phenomena s precisely the same, then, as

theories, they are .not distinguishable from one

another. And if they allow no possibility of extra-

polation, if the actual course,of our experiencecan

havb no bearing upon their truth or falsehood, t is

misleading to call them theories at all.At this point it may be objected that our experi-

ence may very well be relevant to the validity of

these heories. For have we not seen hat the ground

trpon which Professor Price rejects them is that they

fail to account for such phenomenaas hose of double

vision and complete hallucination ? And does not

this show that they are capable of being empirically

tested, and that they do not all describeall the known

s4 FOUNDATIONS OF EMPIRICAL KNOWLEDGE r I THE ARGUMENT FROM ILLUSION 55



indeed is the opinion one would derive from a

casual reading of Price's arguments. But when one

comes to examine them more closely, one finds that

his objections are purely linguistic. We have seen,

for example, that he rejects the " theory of multiple

location " on the ground that one cannot say that a

material thing has the quality of being double orbeing hallucinatory from a certain place, since being

double and being hallucinatory are not qualities.

But this is an objection only to the form of certain

sentencesand not to their content. A slight emenda-

tion is all that is necessary o meet it. Instead of

sayrng that a material thing M really is double or

hallucinatory from a place P, the exponents of

the theory will have to say that there are two M'sfrom P, or in the caseof an hallucination, that there

isan M from P but not from other places. Similarly,

the exponents of the " theory of appearing ", instead

of saying that a thing appears double to a given

obsgrver, will have to say that there appear to him

to be two things ; nor need they be disturbed by the

objection that in the caseof a complete hallucination

there really is no material thing to be a term for the

relation of appearing ; for they can answer that their

use of the word " appear " is such that in saying

that a material thi-qg appears to someone they do

not imply that it exists; and indeed this is not

an abnormal usage of the word. In the caseof the

" theory of compound things " the linguistic char-

acter of Price's objection is still more obvious; for

what his arguments amount to is merely that the

word " compound " is being used in an unfamiliar

sense. But to refute a theory something more is

required than criticisms of the manner in which it is

formulated. The essential question is not whether

the theory is expressed n a way that conforms to

our ordinary verbal conventions, but whether it fits

the empirical facts.

It must be said then that if the " theories "againstwhich they are directed really were theories,

in the sense n which this word is commonly under-

stood,ProfessorPrice's objectionswould be without

force. But the truth is not that his objections are

without force, but that he has mistaken the character

of that which he is attempting to refute. For what

he regards as alternative theories are, in fact, what Ishould call alternative languages. As languages, hey

afford us the means of describing what we already

know, but it is not to be expected that we should be

able to deduce anything'from them concerning the

nature of our future experience; for if that were

possible hey would not be languages,but theories

in the ordinary sense. Nor is it to be expected hat

we should be able to discover any factual objections

to them. For what we here have to consider is not

a number of alternative hypotheses concerning the

nature of the empirical facts, but a number of

alternative recommendations concerning the way in

which we are to describe them. And whether or not

we are to accept any of these recommendations is a

question that we have to decide on linguistic grounds.

A feature which the " theories " of multiple

s6 FOUNDATIONS OF EMPIRICAL KNOWLEDGE r I THE ARGUMENT FROM ILLUSION 57



location and compound things have in common withthe " theory " of sense-data, hich has beenadopted

in one form or another by most modern philosophers,

is that they all recommendthat we should makesome

departure from our ordinary mode of speech. Inthis they differ from the " theory of appearing ",

which may indeed be interpreted as being simply theproposal that we should continue to describe per-

ceptual situations in the way that we ordinarily do.Now I have shown that there is no ground for sup-posing that acceptanceof this proposal would makeit impossible to give a truthful description of anyempirical matter of fact ; so that if I reject it, inphilosophizing about perception, it is not becausethe " theory of appearing is

false, but becauseregard the alternative languageof sense-dataas beingmore suitable for my purpose. And this appliesalsoto my rejection of the " theories" of compound thingsand multiple location. The forms of expression

which are advocated in these theories are intended,

like that of the sense-datum heory, to eliminate theproblems which arise out of the ambiguous use ofwords like " touch " and " see in ordinary speech;

but they do not achieve this in a satisfactory way.What makes the " theory of multiple location',inadequate in this respect is chiefly the fact thatvariations in the appearances.ofmaterial things donot depend merely on the position of the observer,but also on many other factors which the theorydoes not specify. They are indeed specified in the" theory of compound things " : but this very virtue

makes its terminology too cumbersome to be con-

venient. There is, moreover, a further objection to

both these alternatives to the sense-datum termino-

logy, in that they lack the advantage, which the

other affords, of enabling us to refer to the contents

of our sense-experiences, ithout referring to material

things.

Accordingly, I may sum up my long discussion

of the argument from illusion by saying that it makes

it seem desirable to use a technical terminology of

some kind in philosophizing about perception; and

that of those that are available the terminology of

the " sense-datumtheory " appears to be the best.

But it must be understood clearly that the acceptance

of this theory involves nothing more than a decisionto use a technical language and I shall attempt

to specify the conventions of this language more

precisely han has been the custom of those who have

used it hitherto. For otherwise it will be found to

defeat its own object !y grving rise to a number of

fictitious problems, which are no less troublesome

than those that it enables us to escape.

Documents

AYER_The Foundations Of Empirical Knowledge_I_The Argument From Ilusion.pdf