Ayer the Caracterization of Sense-data

I ITHE CHARACTERIZATION OF SENSE-DATA

6. Acrs eNo Onyncrs rN SrNsATroN

I nevB not so far attempted to give any explicitdefinition of the word " sense-datum ". I havechosen rather to indicate its usage by giving examplesin which sentences referring to sense-data are intro-duced as translations of sentences the meaning ofwhich is already known. The general rule whichone may derive from these examples is that thepropositions we ordinarily express by saying thata person I is perceiving a material thing M, whichappears to him to have the quality tc,may be expressedin thO sense-datum terminology by saying that I issensing a sense-datum s, which really has the qualityx, and which belongs to M. In this case it is assumedthat the word " perceive ", or whatever word maybe employed to designate the kind of perceiving thatis in question, is being used in such a way that tosay that a material thfrrg M is perceived entails sayingthat it exists. If we do not make this assumption,then we must say not that s belongs to M but onlythat A takes it to belong to M, and so allow forthe possibility that M does not exist; but in other

5E

II THE CHARACTERIZATION OF SENSE-DATA se

respects the translation is the same. And from this

it follows that to assert that people actually do

experience sense-data need be to assert no more thanthat such propositions as that I am now perceiving a

clock or a pen or a table, in a sense of " perceiving "that does not necessarily entail that these objectsexist, are sometimes true. And I do not see how it

can be denied, not merely that many propositions of

this kind are in fact true, but also that we often havegood reason to suppose that they are. f can thereforeclaim to be using the word " sense-datum " in such a

way that there can be no serious doubt that sense-

data actually are experienced.In following this procedure, I shall, I think, be

giving to the word " sense-datum " the meaning thatthe philosophers who have adopted the " theory ofsense-data " have, in general, intended it to have,though I shall show that some of them have also beeninclined to make statements about sense-data that areinconsistent with this usage. And the definition ofsense-data that these philosophers commonly give isthat they are the objects of which, in sense-percep-tion, one is directly aware. By this they must beunderstood to mean, not that only those objects ofwhich someone is in fact directly aware are to becalled sense-data, but rather that the word " sense-datum " is to stand for any object of which it isconceivable that someone should be directly aware.They do not always say this; but inasmuch as theyconsider themselves entitled to refer to possible

as well as to actual sense-data, it may fairly be

6o FOUNDATIONS OF EMPIRICAL KNOWLEDGE n

assumed that this is what they mean.The first criticism that I have to make of this

definition is that it is not illuminating. For thereis no accepted meaning of the expression " directawareness " by reference to which it can be madeclear without further explanation what is to be meantby the word " sense-datum ". I think that we shouldordinarily say that we were directly aware of anyobject, our belief in the existence of which was basedupon sense-experience and did not involve anyconscious process of inference. But if the expression" direct awareness " is used in this sense, then it willbe true to say that we are directly aware of materialthings, such as chairs and tables and pens. Foralthough our beliefs in the existence of such thingsare inductive, in the sense that they imply more thanwould be implied by a mere description of theexperiences on which they are based, they often donot involve any conscious process of inference. Theprdposition that I am holding a pen in my hand isnot equivalent to any proposition or set of pro-positions that describe my present sense-experiences,though they provide me with the only grounds I havefor asserting it. But this does not mean that I haveconsciously gone through any process of inferring itfrom them. In tirtue of my sense-experiences Isimply take it for granted that this is a pen. Andthis is the way in which we actually arrive at a greatmany of our beliefs in the existence of material things.But, as a rule, those who define sense-data as theobjects of which one can be directly aware maintain

tr THE CHARACTERIZATION OF SENSE-DATA 6r

that one cannot be directly aware of such objects astables or pens. And the reason why they maintainthis is that our perception of such things may alwaysbe delusive. In other words, they define " directawareness " in such a way that if someone is directlyaware of an object *, it follows that x exists and thatit really has whatever properties it is appearing tohave. But having given this meaning to the pro-

positional function " A is directly aware of cc ",where, it may be asked, are they to find values thatwill satisfy it I The answer is that they providethese values by introducing the word " sense-datum ", or some synonymous term, and using it inthe way that I have indicated. That is to say, theexpressions " direct awareness " and " sense-datum "are to be regarded as correlative; and since each ofthem is being used in a special, technical sense, it isnot satisfactory merely to define one in terms of theother. It is necessary first to employ some othermethod, such as the method of giving examples, inorder to show how one or other of them is to beunderstood.

This definition of sense-data as the objects ofdirect awareness is often associated with a par-

ticular view about the analysis of sensations whichit does not logically entail. This view is that it ispossible to discriminate in any sensation at least two

distinct factors, one of them being the act of sensingand the other the object sensed. I say that theacceptance of the definition does not entail theacceptance of this analysis of sensation, because it is

c2

6z FOUNDATIONS OF EMPIRICAL KNOWLEDGE rI

clearly possible to assert that we are directly aware

of sense-data without implying that the expression

" direct awareness " is a name for any sort of intro-

spectible act. And indeed it seems to me very

doubtful whether there are such acts. It is true, as

Professor G. E. Moore points out in his " Refutation

of Idealism "rt that the expressions " blue " and

" consciousness of blue " are not synonymous, and

that my consciousness of blue and my consciousness

of green have something more in common than

what is common to blue and green. But it does not

follow that this common element of consciousness

is a distinct, individual factor itt *y sensation. It

may well be that the characteristic in virtue of which

it may be said that the blue and green sense-data are

both experienced by me is a relational characteristic,

which does not involve either myself, conceived as a

substance, or any such thing as that for which the

expression " act of sensing " is supposed to be a

name, but only certain other sensible, or intro-

spectfole, objects. And whether such an analysis is

correct or not, there is nothing in Moore's argument

to refute it.There are, however, some philosophers who base

their belief in the existence of these acts of sensing,

not on any a priori argument, but on the evidence of

their own introspectidn ; and I do not wish to assert

dogmatically that they are wrong. I cannot myself

discover these acts by introspection; but this does

not prove that no one else can. At the same time, I

I Philoso,hical Studies, pp. Z-3o.

rr THE CHARACTERIZATION OF SENSE-DATA 63

think that those who do claim to be able to discoverthem in this way may, perhaps, be making an un-warranted inference from a different empirical fact.It is characteristic of some sense-data that theyappear to be sensibly outside ourselves ; by which Imean only that they occur in sense-fields that havethe property of sensible depth. This is true of visualand tactual data, and also of auditory and olfactorydata when they are ascribed to an objective source.Now, because of our knowledge of their immediatecausal conditions, we tend to think of our sensationsas occurring somehow inside ourselves ; and there-fore it is assumed that sense-data that are at a distancefrom the somatic centres of their sense-fields, andthus may be said to be sensibly outside ourselves,cannot make up tJre whole content of the corre-sponding sensations; and so acts of sensing arebrought in to fill the gap. If this explanation werecorrect, we should expect to find that those who madethis distinction between act and object in theiranalysis of sensation were more confident of itsvalidity in respect of sight and touch than in respectof organic and kinasthetic sensations; and this is,in fact, the case.r But it is clear that if those whobelieve in the existence of acts of sensing are tacitlyrelying on this argument, their conclusion is notestablished. The most that they can prove is thatsome sensations are not identical with the relevantsense-data, or in other words that the expression" sensation of. x" is not synonymous with " cc".

I Cf. C. D. Broad, Scimtific Thought, pp. zS4-7.

6q FOUNDATIONS OF EMPIRICAL KNOWLEDGE II

But while we might conclude from this that in assert-ing that a sense-datum was experienced, we wereasserting that it was related to something other thanitself, it would not follow that this other term wasan act of sensing. For, as I have already remarked,there is no logical inconsistency in holding boththat " the sensation of r " is not synonymous with

" tc ", and also that the proposition that the act ofsensing is a distinct, individual factor in any sensationis empirically false.

However, the philosophers who maintain thisdistinction between act and object in their analysisof sensation do not, for the most part, desire merelyto call attention to an empirical matter of fact.They consider the distinction to be philosophicallyimportant because they think that they can use itto refute Berkeleyan idealism. For, according toBerkeley, colours and shapes and sounds and allother " sensible qualities " are mind-dependent, inas-much as thpir existence consists in their being per-ceived ; and since material things are, in his view,nothing but collections of sensible qualities, he con-cludes that they too cannot exist apart from aperceiving mind.' But here, it is argued, he fallsinto error through failing to distinguish between theobject of a sensation and tfe act. Acts of sensation,it is said, are indeed mind-dependent I but it doesnot follow that their objects are ; for there is nogood ground for supposing that the object of a

I Vide A Treatise concerning the Principles of Human Knoailedgeand Three Dialogues between Hylas and Philonous, passim.

It THE CHARACTERIZATION OF SENSE-DATA 65

sensation cannot exist apart from the act. And ifwe assume, as some philosophers do, that these actsof sensation are acts of knowing, then it is held thatwe can go so far as to maintain that the objects ofour sensations must exist independently of the acts ;for it is supposed to be an essential characteristic ofknowledge that what is known exists independentlyof the knowing of it. But this argument is two-edged, as Professor Prichard has recently shown.'For he maintains that it is self-evident that the exist-ence of what he calls secondary qualities and Ishould call sense-data, does depend upon theirbeing perceived, in Berkeley's sense of the word ;and therefore that the proper conclusion of the fore-going argument is not that the objects of our sensa-tions exist independently of the acts, but that acts ofsensation are not acts of knowing. And'since hesupposes, with some historical justification, thatsense-data are defined as the objects of perceptualacts which are taken to be acts'of knowing, he con-cludes that there are no such. things as sense-data.For he argues that to say that there are sense-data issimply a misleading way of saying that " perceivingis a kind of knowing ", and that this proposition isfalse.

7. " Essn rsr PERctpr "

But, setting aside for the moment the questionwhether perceiving, in the sense here in point, is or

I In a paper called " The Sense-datum Fallacy ", AistotelianSociety Supplementary Proceedings, rgg8.

66 FOUNDATIONS OF EMPIRICAL KNOWLEDGE rr

is not a kind of knowing, I wish first of all to examinethe Berkeleyan principle that to exist is to be per-ceived. I shall consider it first in relation to materialthings, and then in relation to sense-data. And hereit must be understood that in using the term " sense-datum " I am not presupposing that " sensiblequalities " are objects of knowledge, in Prichard'ssense ; nor am I making any other covert assumption.My use of the term is so far to be understood onlyby reference to the examples that I have given; andwhatever further implications I intend it to carry, Ishall explicitly introduce.

If the principle that esse est percipi, as applied tomaterial things, is to be interpreted to mean that alogical contradiction is involved in asserting theexistence of any material thing that is not actuallybeing perceived, then it is plainly false. For even ifit is the case that no propositions asserting theexistence of unperceived material things are, in fact,eveg true, they are still not self-contradictory. Butis it the case that no such propositions are true ?It has indeed been argued that we can never haveany reason to believe in their truth, on the groundthat the only evidence \ile can have for the existenceof a material thing is that we actually perceive it,and that it is impospible to perceive something exist-ing unperceived.. But this argument is invalid. Itis true that if we were to regard the existence ofmaterial things as being logically independent, notmerely of actual perceptions, but also of all possible

t Cf. W. T. Stace, " The Refutation of Realism ", Mind, rg34.

rr THE CHARACTERIZATION OF SENSE-DATA 6l

perceptions, then we should have no good reason to

believe that they ever existed unperceived. But thisis not an acceptable analysis. The criterion by which

we determine that a material thing exists is the truth

of various hypothetical propositions asserting that if

certain conditions were fulfilled we should perceive

it.' These propositions cannot indeed be formally

deduced from the propositions which describe the

contents of our actual perceptions. But they can be

derived from them by an ordinary inductive argu-

ment. We may say, therefore, that we are justified

in believing that material things exist when no oneis actually perceiving them, because the fact that we

do constantly perceive them in certain conditionsgives us a good inductive ground for believing, attimes when we are not perceiving them, that we

should be perceiving them if these conditions werefulfilled. If the conditions are not, in fact, fulfilled,

then we cannot put our belief to a direct empirical

test. But this does not,mean that it is baseless. For,as I have already remarked, it is not required for thevalidity of the hypothetical propositions about sense-data, in terms of which one can analyse propositions

about material things,2 that their protases should beempirically realized. And it is this that constitutesthe independence of material things. Accordingly, Iconclude that it is not a necessary condition of theexistence of a material thing, or for our belief in itsexistence to be justifiable, that it should actually be

r For an explanation of.,H".J"f,."." V of this book.


perceived. Nor is it an altogether sufficient condi-tion. For I am using the word " perception " insuch a way that it is possible for a perception to bedelusive.

In reaching this conclusion that the existence ofmaterial things is to be determined with referenceto possible rather than actual perceptions, I havebeen guided by the way in which the relevant wordsare currently used. But this is not a method thatone can extend to the case of sense-data. For herethe problem is not to elucidate the conventions thatgovern the use of an existing language, but to formu-late conventions for an artificial language. In thecase of material things, one is able to make use of aprevious understanding of the meaning of existentialpropositions in order to criticize the principle thatto exist is to be perceived. But in the case of sense-data one has no such previous understanding. Theposition in this case is that one must decide whetherto atcept or reject this principle in order to determinewhat one is going to mean by saying that a sense-datum exists.

Now one of the purposes which the introduc-tion of the sense-datum terminology is intendedto serve is that it should enable us to deal withthe problems which arise from the fact that materialthings can appear to have qualities that they do notreally have, and can appear to exist when they donot. It is this that is effected by the translation ofsuch sentences as " f am perceiving a brown carpet,which looks yellow to me " or " The drunkard sees

rr THE CHARACTERIZATION OF SENSE-DATA 6q

animals which are not really there " into " I amsensing a yellow sense-datum which belongs to abrown carpet " or " The drunkard sees sense-datawhich he takes to belong to animals, but which donot really belong to anyhing ". The advantage ofthis procedure is that it makes it possible for us to saythat something real is being experienced even in caseswhere our perceptions are delusive. But this ad-vantage is sacrificed if we extend the distinctionbetween appearance and reality to sense-data them-selves. And for this reason we should not allow anymeaning to such sentences as " I am sensing a yellowsense-datum, but it is really brown " or " Perhapsthe sense-data that I am now sensing do not reallyexist ". It is true that some philosophers have beeninclined to attach meaning to sentences of this kind.They have seriously considered the possibility thatsense-data as well as material things might appear tohave properties they did not really have. And itmay be argued that this'is simply a question of analternative specification of the rules of the sense-datum language, and that they are free to adopt sucha convention if they wish. But the answer is that byadopting this convention they come to treat sense-data as if they were themselves material things orcharacteristics of material things ; and in that casethe terminology of sense-data becomes superfluous.The point of introducing it was to clarify the meaningof the sentences in which we ordinarily refer tomaterial things by using a language of a differentstructure from that of our ordinary language. But

7o FOUNDATIONS OF EMPIRICAL KNOWLEDGE rr

if the reference to sense-data is to serve this purpose,it is necessary that the criteria that determine theproper use ofsuch sentences as " this sense-datum isbrown " should be different from those that deter-mine the proper use of such sentences as " this carpetis brown ". If we make them the same, our newterminology becomes a mere reproduction of the old.It eliminates none of the familiar problems, and sofar from being a source of clarification, it creates addi-tional confusion by suggesting that the introductionof sense-data is not just a linguistic expedient, butmarks the discovery of a new kind of material thing.

We must decide therefore not to admit the dis-tinction between veridical and delusive perceptionwith regard to sense-data. And this means that wemust make it a sufficient condition of the existenceof a sense-datum that it should actually be sensed.But are we also to make it a necessary condition ?A conceivable objection to this is that we wish tospeak not merely of actual but also of possible sense-data ; for in analysing propositions about materialthings we have to refer, not so much to the sense-experiences we are actually having, as to those thatwe should be having if certain hypothetical condi-tions were fulfilled. And it may be argued that, inso far as these hypqthetical propositions are true, wemust allow that possible sense-data, or sensibilia, asBertrand Russell called them,r also exist withoutnecessarily being experienced. But there is really noreason why we should draw this conclusion. We do

I Vide Mysticism and Logic, pp. r48 ff.

tI THE CHARACTERIZATION OF SBNSE-DATA 7t

indeed make the truth of these hypothetical proposi-tions a criterion for the existence of material things,but it is not necessary that we should extend thisprocedure to the case of sense-data. The convenienceof our use of the word " existence " as applied tomaterial things consists in the indefiniteness and thegenerality of its reference. It enables us to assert anumber of hypothetical propositions about the con-tents of our sense-experiences without having tospecify them individually. But no such advantageis to be obtained from the proposed usage of" existence " in connexion with possible sense-data.For if we are to avoid identifying these sensibilia withmaterial things we shall have to take as a criterion forthe existence of a sensibile the truth of a singlehypothetical proposition; and in that case we shallbe able to express no more by asserting that thesensibile exists than we are already able to expressby asserting the hypothetical proposition in question.And not only is there no positive advantage inextending the use of the word " existence " in thisway to possible sense-data ; there is also the dis-advantage that the proposed usage, though not indeedidentical with that which is adopted in connexionwith material things, is sufficiently like it to be a rcadysource of confusion. Accordingly, I find it advisableto make it a necessary as well as a sufficient conditionof the existence of sense-data that they should infact be sensed. I shall continue to speak of pos-sible sense-data as an alternative way of assertingthe relevant hypothetical propositions. But only


sense-data that are actually experienced will be saidto exist.

It may be noted that this decision to accept theprinciple that the existence of sense-data consists intheir being experienced involves neither an admissionnor a denial of the distinction between acts of sensa-tion and their objects ; and so a refutation of theprinciple that consists merely in an attempt toestablish this distinction cannot have any validity forus. Moreover, it seems that those who seek to" refute idealism " in this way are making the mistakeof treating the question whether the existence otsense-data consists in their being experienced, notas a question concerning the rules of the sense-datum language, but as a question of fact. ProfessorG. E. Moore, for example, in his " Refutation ofIdealism ",' expressly asserts that he regards esse estpucipi as a synthetic proposition. From this it maybe inferred that he is using the word " existence " ina different sense from that in which I have decidedto use it in connexion with sense-data, but he doesnot tell us what this sense is. He asserts dog-matically that the sentences " blue exists " and " theconsciousness of blue exists " do not express equiva-lent propositions, and suggests that the failure to seethis is mainly reeponsible for the groundless beliefthat esse est percipi. But while it is fairly clear fromthe context that he is using the word " blue " tomean what I should call a blue sense-datum, it isnot at all clear how he is using the word " exists ".

. Philosophical Studies, pp. r-3o.


If the sense intended is that in which the word is

applied to material things, then, as I have shown, he

is right in maintaining that not everything that exists

must also be in fact perceived. But I have shown

also that the effect of extending this usage of

" existence " to sense-data is to make the sense-

datum terminology superfluous. There is no point

in talking about sense-data at all unless we adopt

rules for the use of this technical term that are

different from the rules we already follow in speak-

ing about material things. And for this reason it is

not legitimate to assume, as Moore apparently does,

that a question which has a factual meaning with

reference to material things must also have a factual

meaning when it is made to refer to sense-data.

This is not, of course, to say that no true factual

statements can be made about sense-data, but only

that it is advisable, if one is going to exPress empirical

propositions with the help of this technical term, to

begin by having a clear'understanding of what is

involved in its use.To show how the neglect of this point can lead

to confusion, I may refer to another article by Moore

which he entitles " Some Judgements of Percep-

tion "., In this he maintains that it is quite certain

that when one makes such a judgement as that " this

is an inkstand ", there is one and only one object

about which one is making this judgement, though

the judgement one is making about it is certainly not

that it is itself a whole inkstand, and that this objectI lbid.pp. zzo-s2.

7+ FOUNDATIONS OF EMPIRICAL KNOWLEDGE rr

is a sense-datum. He then raises the question howthis sense-datum is related to the inkstand, and saysthat he is inclined to adopt the view that it is literallypart of it. He admits that in order to maintain this hemust also hold that sense-data can sometimes appearto have qualities that they do not really have, andthat this supposition is one that is often thought to benonsensical. He says, however, that he is not him-self convinced that it is nonsensical. But now wemust ask why it should be thought to be certain thata judgement such as " this is an inkstand " is reallyabout a sense-datum. Why should it not be aboutan inkstand ? The answer is that the judgement isheld not to be about the inkstand because it isassumed that the inkstand itself is not sensibly" given ". There is indeed a familiar use of the word" about " according to which it would be proper tosay that the judgement " this is an inkstand ', reallywas about an inkstand. But this is not the way inwhich Moore is using the word. He is using it insuch a way that to say that a judgement is about anobject * entails that * is given, in a sense in whichsense-data can be said to be given but inkstands can-not. But what exactly is this sense ? Why may wenot say that inkstands are given, rather than sense.data ? ft is becauge the perception which gives riseto our judgement'ihat " thi. i.

"r, inkstand ,, mav

always be qualitatively or existentially delusive.There may not really be an inkstand there, or it maynot really have the qualities that it appears to haveon this occasion. But if this is the reason whv Moore


holds that the inkstand is not sensibly given, then itis clear that he cannot consistently allow that thesense-datum, which is supposed by him to be thatwhich the judgement is really about, may itself appearto have qualities that it does not really have. For in

saying that the judgement that " this is an inkstand "is really not about an inkstand but about a sense-datum, which is a way of saying that it is not theinkstand but only a sense-datum that is sensiblygiven, he is implying that the perception of sense-data cannot be delusive in the sense in which theperception of an inkstand can. If, therefore, he

maintains that the sense-datum as well as the ink-stand can appear to have qualities that it does notreally have, he falls into self-contradiction. He isnot entitled to assume that because the distinctionbetween veridical and delusive perception applies tomaterial things, it applies also to sense-data. Forthis assumption is inconsistent with the way inwhich he intends that the word " sense-datum "should be used.

We have seen that the reason why some philos-ophers have been anxious to deny that the existenceeven of sense-data consists in their being perceived,is that the acceptance of this principle is thought tolead to idealism. It is believed that if we admit thatthe objects of which we are directly aware can existonly when we are experiencing them, we shall be

committed to the view that everything that we

experience " exists only in the mind ", and that this

is to put in question the reality of the external world.

76 FOUNDATIONS OF EMPIRICAL KNOWLEDGE n

But what is meant here by saying that an objectexists in the mind ? Presumably that it is what weshould ordinarily call a state of mind. But in thatcase the proposition that sense-data exist only in themind does not follow from the proposition that theycannot exist unsensed. To say that a colour or asound or any other sense-datum exists only when itis experienced does not by any means entail sayingthat it is a state of mind, in the ordinary sense inwhich a feeling or an emotion is held to be a stateof mind. And indeed there appears to be no goodreason for holding that any such sense-data are statesof mind, in this sense.

But if the assertion that sense-data exist only inthe mind is not to be understood in this way, howare we to interpret it ? I do not think that it ispossible to find any satisfactory meaning for it, unlesswe assume that it is merely a misleading way ofre-stating the convention that the existence of sense-data ip to consist in their being experienced ; butat the same time we can, I think, account for itsbeing made. For what we have here is yet anotherinstance of the misuse of the argument from illusion.In this case the argument is taken to prove that theobjects of which we are sensibly aware do not formpart of an external, r4aterial world. Nevertheless itis assumed that since these objegts are undeniablyexperienced they must in some sense be real ; andas they have been extruded from the material world,it is thought that another receptacle must be foundfor them ; and the only one that appears to be


available is the mind. How indeed the mind issupposed to contain them, it is not easy to under-stand. I do not think that even Berkeley can reallyhave wished to maintain that " sensible qualities "were literally inherent in the mind as in a region ofspace. Nevertheless he was sufficiently misled bythe spatial metaphor to be capable of arguing, insupport of his view that we have no direct visualperception of distance, that the extension and figurec,f an object seem to be in the same place as itscolour, and that the colours that we see cannot beat a distance from us because it is " agreed on allhands, by those who have any thoughts of thatmatter, that colours, which are the proper andimmediate object of sight, are not without themind ".' But, apart from such absurdities- as this,the whole process of reasoning that leads to theattempt to house sense-data in the mind is thoroughlyconfused. For, in the first place, I have shown thatthe argument from illusion does not prove that theobjects of which we are sensibly aware are notconstituents of the material world, if this is regardedas a question of fact. It does indeed provide us

with a motive for altering our terminology in a waythat results in our saying that we are directly aware,not of material things, but only of sense-data. Butthis means that we are adopting a new method ofdescribing our perceptual experience, which is asubstitute fcir our ordinary method of descriptionand cannot simply be grafted upon it. To ask

I A New Theoty of Vision, section xliii.


whether sense-data inhere in the material world orin the mind is thus entirely to misconceive theircharacter. If we agree to say that the objects ofwhich we are directly aware are always sense-data,then we are deciding to treat them and not mindsor material things as the units in terms of which weare to describe our perceptual experience. Thequestion, therefore, tlat we must ask is not howsense-data are to be incorporated in the categoriesof mind or matter, or whereabouts they are to belocated in physical space, but rather how our con-ceptions of " mind " and " material things " and" physical space " are to be analysed in terms ofthem. Accordingly, if we conclude that sense-dataare not states of mind, we must not then look roundfor some other container for them, such as the brainor " the psycho-cerebral compound ". For it isalready a mistake to suppose that they can be phasesof any substance, or anywhere in physical space, at all.

8. SnNsruc eNo KnowrNc

It appears, then, that in order to " get outsidethe circle of our minds " it is not necessary for usto maintain that our awareness of sense-data is a kindof knowing. And indeed, if it is essential to know-ledge that the obj6ct known should exist inde-pendently of the knowing of i{, I have implicitlydenied that our awareness of sense-data is a kind ofknowing, for I have made it a necessary and sufficientcondition of the existence of sense-data that they

rr THE CHARACTERIZATION OF SENSE-DATA 7s

should in fact be experienced. But I am not surethat this is essential to knowledge, as it is ordinarilyunderstood. The difficulty here is that the ex-pression " object of knowledge " is ambiguous. Itis used with reference both to our knowledge of thetruth of propositions, and to our acquaintance withparticular things. If we are concerned with pro-positional knowledge, then there is justification forsaying that knowledge presupposes the independenceof its object ; for it is essential to knowledge of thiskind that the truth of the proposition known shouldnot depend upon our knowing it. But it does notfollow from this that the existence of a thing whichwe know, in the sense of being acquainted with it,must also be independent of our knowing it. Further-more, this second sense of " knowledge " is itselfambiguous. For we tend to speak indifferently ofbeing acquainted both with sense-data and withmaterial things ; but if the term " acquaintance " istaken, as it usually is, to be 'equivalent to " directawareness ", then in the sense in which it is applic-able to sense-data, it is not applicable to materialthings. Accordingly, whether or not there ishistorical justification for speaking of knowing sense-data, it seems advisable for the sake of clarity not touse the word " knowledge " in this sense. To avoidambiguity, I shall in future use the word " aware-ness " only in connexion with sense-data, and theword " perception " only in connexion with materialthings, and I shall restrict the use of the word" knowledge " to its propositional sense.

8o FOUNDATIONS OF EMPIRICAL KNOWLEDGE u

If one uses the word " knowledge " in this way,one cannot say that the awareness of sense-data isitself a kind of knowing ; but we will still have to saythat it involves knowledge. For the meaning of theexpression " direct awareness " is such that, when-ever we are directly aware of a sense-datum, itfollows that we know some proposition whichdescribes the sense-datum to be true. But what isthe character of this knowledge ? In the ordinaryway, I think that all that is required for an empiricalproposition to be known is that it should in fact betrue, that no doubt should be felt about its truth,and that the belief in it should not have been reachedby way of a belief in any false proposition, and shouldhave good inductive grounds ; and in this sense wecan claim to know propositions about the existenceand properties of material things, and even generalpropositions, although our perceptions can neverafford us a logical guarantee of their truth. But it iswidety held that when we speak of knowing a pro-position which describes a presented sense-datum, weimply more than this. It is held that we imply, notmerely that our belief in the truth of such propositionis not, in fact, mistaken, but that it could not con-ceivably be mistaken. And it is this view that issometimes expressed by the assertion that such pro-positions are " indubitable " or f incorrigible ".

But now it may be asked: How can any empiricalproposition be indubitable in this sense ? We saythat an a priori proposition is indubitable because itscontradictory is self-contradictory, but this cannot be

rr TIIE CHARACTERIZATION OF SENSE-DATA 8r

the ground for asserting that propositions describingpresented sense-data are indubitable; for it is notmaintained that these propositions are analpic. Butwhat other ground can there be for such an asser-tion ? It is not as if it meant only that we werepsychologically unable to doubt these propositions;for that might be true also of propositions about theexistence of material things, which are supposed to bedistinguished from propositions about the existenceof sense-data partly by the fact that they are notindubitable in the sense that is here in question. Itmeans rather that to doubt the truth of a propositionwhich describes a presented sense-datum is logicallyincorrect. But surely this is a condition that noempirical proposition can possibly satisfy I

To see how this objection may be met, let me tryto give an instance of the expression of an " incor-rigible " proposition. Suppose that I experience avisual sense-datum which I describe by saying " thisis green ", and suppose that I am using the sentencemerely to designate the sense-datum in question, andam not implying that it is in any way related to any-thing that I am not simultaneously experiencing. Isthere then any possibility of my being mistaken ?The answer is that I can at least be making a verbalmistake. It may, for instance, be a rule of thelanguage I am supposed to be using that the correctname for the colour of such a sense-datum is " red "and not " green ". And if this is so I am, in asense, misdescribing what I experience. Butmistake is not a mistake of the same kind as

mythat

8z FOUNDATIONS OF EMPIRICAL KNOWLEDGE rr

which I should be making if I took the sense-datumto belong to the wrong material thing, or made anyother judgement that involved relating the sense-datum falsely to further actual or possible experi-ences. Suppose, for example, that I had said " thisis a green book " and that there was really no bookthere, or that the colour of the book was not reallygreen. In that case, I should have made not merelya verbal error but an error of fact; for I shouldimplicitly have asserted a number of hypotheticalpropositions which, as I might subsequently dis-cover, were not true ; I might find, for instance, thatwhen I went to open the supposed book it appearedhollow inside, or that when I observed it in a clearlight it never again looked green. But whether ornot I actually discover that I am mistaken is not, forthe purpose of this illustration, of any importance.What is important is that the possibility of my beingmistaken, in what is not merely a verbal sense,deppnds upon the fact that my judgement goesbeyond the evidence upon which it is immediatelybased. It connects an experienced sense-datum withother possible sense-data which are not simul-taneously given, and in doing so it allows room fordoubt and error. For it cannot be formally deducedfrom a mere description of the immediate evidencethat these further- sense-data would, in fact, beobtainable even if the relevant conditions were ful-filled. In other words, the reason why I can bemaking an error of fact in asserting a proposition ofthis kind is that the proposition is not completely

tt THE CHARACTERIZATION OF SENSE-DATA 8s

verified by the existence of the sense-datum upon theobservation of which it is based ; nor can it ever becompletely verified. Consequently, although it maybe irrational on the evidence to express doubt in itstruth, it is always significant. But it is held to becharacteristic of an " incorrigible " proposition thatit is completely verified by the existence of the sense-datum which it describes ; and so it is inferred that todoubt the truth of such a proposition is not merelyirrational but meaningless; for it is only significantto doubt where there is a logical possibility of error.The argument is, in short, that if one uses a sentencesuch as " this is green " merely to designate a presentsense-datum, then no proposition is being assertedto the truth of which any further evidence would berelevant. And from this it is concluded that all thatone can properly mean in such a case by saying thatone doubts whether this is green is that one isdoubting whether " green " is the correct word touse. And the same would aplly to any other sen-tence that was used only to designate some feature ofwhat was actually being experienced, without beingintended to carry any further implication. Thepropositions which such sentences were supposedto express might be said to be indubitable on theground that it w.as not significant to say that onedoubted them in any other but a purely verbal sense.

From this it may be concluded that all that isinvolved in the claim that there are indubitable orincorrigible empirical propositions is that people dosometimes use sentences in the way that the sentence

8+ FOUNDATIONS OF EMPIRICAL KNOWLEDGE II

" this is green " was used in my example, or at anyrate that they could use certain sentences in this way,if they chose. And I do not see any reason to denythat this is so. It is indeed to be remarked that whatI have been calling verbal doubt is also, in a sense,factual; for it is, after all, a question of fact whetherone's use of a word on a given occasion is or is notin accordance with one's own habitual usage, or withthe usage of other people. But the point is thatwhereas, in the case of most empirical propositions,it would still be possible to doubt them even if therewere no doubt that the relevant words were beingcorrectly used, this possibility does not extend to theclass of propositions about sense-data that I havebeen considering. And the reason for this is simplythe linguistic fact that we do not attach any meaningto the statement that we doubt these propositions, inthis further non-verbal sense of " doubting " inwhich we do attach meaning to the statement thatwe j.oubt the others.

9. Tnn Ennons oF FoRMALISM

A curious fact about this question of incorrigiblepropositions is that the answers which philosophershave given to it h4ve usually been made to dependupon the view they have taken of what, at firstsight, appears to be an entirely different problem.For, on the one hand, it has been assumed by thosewho maintain that some empirical propositions mustbe incorrigible that their case would be proved if


they could show that the truth or falsity of empiricalpropositions depended in the last resort, not uponrelations that they bore to one another, but upontheir accordance or discordance with observablefacts ; so that the main arguments which these philo-sophers have produced in favour of there beingincorrigible propositions have been arguments againstthe coherence theory of truth. And, conversely,what those who have rejected the view that there canbe incorrigible propositions appear, for the mostpart, to have been mainly anxious to deny is that anypropositions do, or can, refer to facts, unless this isinterpreted as being merely a misleading way ofsaying that they have certain relations to otherpropositions. In dealing with this point, the coursethat I shall follow is to consider first theprollem ofthe relationship of propositions to facts, and thelr. tosee what bearing, if any, the solution of this problemhas on the question whether any empirical'proposi-tions are incorrigible.

The position of those who deny the possibility ofexpressing propositions that refer to empirical facts,in the sense in which " reference to a fact " isordinarily understood, mry be summarized as follows.They hold that all that is necessary for the specifica-tion of a language is an account of what ProfessorCarnap has called its formation and transformationrules. The formation rules determine what com-binations of signs are to constitute proper sentencesof the language; the transformation rules prescribethe ways in which these sentences may legitimately


be derived from one another. Both these sets ofrules are held to be purely formal in character; andthis means that they contain " no reference to themeaning of the symbols (for example, the words) orto the sense of the expressions (e.g. the sentences),but simply and solely to the kinds and order ofthe s5rmbols from which the expressions are con-structed ".r Consequently, if a distinction is madebetween the sentences that express a priori and thosethat express empirical propositions, it can be onlyin virtue of a difference in the form of the symbolswhich they contain, or in the nature of the formalrelations which they have to other sentences. Inthe case of languages which allow the expression ofempirical propositions, it is thought possible to markout a special class of sentences, which are referredto as t' observation tt or " protocol " or " basic t'

sentences; and it is held to be a necessary andsufficient condition of the admissibility of anysentegce that is intended to express an empiricalproposition that some observation-sentence shouldbe derivable from it in accordance withthe establishedrules of the language. One must not, however, bemisled by the name of these observation-sentencesinto supposing that they refer to observable facts;or that the test whicfi they provide for the validity ofempirical propositions is one of correspondence withfact, as this is ordinarily understood. For theproperties by which these sentences are essentiallydistinguished are intended to be purely formal ; and

I Rudolf Carnap, The Loghal Syntax of Language,p. r.

rr THE CHARACTERIZATION OF SENSE-DATA 8l

the only criterion that is allowed for determiningthe truth or falsity of any " observation-statement "is the formal possibility of incorporating it in agiven system. Nor does this criterion give them anyadvantage over statements of other kinds. For it isheld that even when a hypothesis " proves to belogically incompatible with certain protocol-sentences,there always exists the possibility of maintaining thehypothesis and renouncing acknowledgement of theprotocol-sentences ".' All that is required is thatthe system of propositions which is accepted as trueshould be formally self-consistent. And this is sup-posed to be the sole criterion of their truth.

I do not think that any very elaborate argumentis needed to show that this theory is altogetheruntenable. In the first place, it is not true that weare able to use or understand a language when weare acquainted only with its formation 4nd trans-formation rules. These rules are indeed sufficientfor the characterization of a purely abstract systeryof logic or mathematics, so long as norattempt. ismade to give the system a material interpretation;but they are not sufficient for the charactefization ofany language that serves to communicate proposi-tions about matters of fact. From the transforma-tion rules we can learn that in any situation in whichwe are entitled to use a given sentence of the language,we are also entitled to use certain others ; butneither they nor the formation rules afford us anymeans of knowing what are the situations in which

r lbi.d.p. y8.


any single sentence can legitimately be used. Butuntil this is determined the " language " in questionis not, in the ordinary sense, a language at all. Forit does not serve to communicate anything. It ismerely a formal calculus which we do not knowhow to apply. For it to become a language it isnecessary that some at least of the expressions thatit contains should be given a meaning. And this iseffected by the method of ostensive definition, thatis, by correlating these expressions, not with otherexpressions, but with what is actually observed.Professor Carnap has indeed asserted that these" so-called ostensive definitions " are themselves

" translations of words ". According to him, todefine, for example, an elephant, ostensively, ismerely to lay down the transformation rule "'ele-phant':animal of the same kind as the animal inthis or that position in sprce-time ".r But this isclearly a mistake. It is true that if I teach someonethe rneaning of the English word " elephant " bypointing to a particular animal, the information hereceives is that an elephant is an animal of the samekind as that which he is observing at a particularplace and time. But this is not to say that the word" elephant " is synonymous in English with any suchexpression as " anima,l of the same kind as that which,on July znd, rg3g, was standing 30 yards south-westof the bandstand at the London Zoo". For even ifit is a fact that an elephant was actually to beobserved at that particular place and time, it is not

t The Unity of Science,p.39.

rr THE CHARACTERIZATION OF SENSE-DATA 8s

a necessary fact. I may be expressing a false pro-

position if I say that there was no animal standing

3o yards south-west of the bandstand at the London

Zoo on July znd, t939, or that there was such an

animal but it was not an elephant; but I shall

certainly not be contradicting myself. The spatio-

temporal description may serve to indicate the

context in which some word has been, or is being,

ostensively defined, but it is not in itself a si&stitute

for that context. And it is only by reference to ar

empirical context that any ostensive definition is to

be understood.Having allowed the possibility of ostensive defini-

tions, we might then consider that we required a

special name for the class of sentences whose con-

stituents were defined in this way; and it would

be reasonable for us to call them observation-

sentences, on the ground that they were intended to

refer to what could be directly observed. But this

cannot be what those who'uphold the theory I am

now considering mean by ao observation-sentence.

For, in effect, they deny the possibility of using any

sentence to refer to what can be directly observed.

As they use it, the term " obseryation-sentence " is

purely syntactical. The sentences to which it is

applied ^re

distinguished from other sentences

merely by the fact that they contain different words

and obey different transformation rules. But what

is the point of this conventional distinction ? It

seems to be intended to furnish a principle of

empiricism. The theory is that one excludes meta-


physics by asserting that every proposition that isnot analytic must be " empirically testable " and bydefining this " testability " in terms of the derivationof observation-sentences. But this use of the term" empirically testable " is again very misleading.For what sort of empirical test is provided by thefact that the conventional rules of a language allowone sentence to be formally derived from another,when neither sentence is to be understood as record-ing what can be actually experienced ? One doesnot become an empiricist merely by a free use of theword " empirical " or the word " obseryation ".

If we were to take this view of the nature ofobservation-sentences, it clearly would not matterwhether we agreed with Carnap that " every concretesentence belonging to the physicalistic system-language can in suitable circumstances serye as anobservation-sentence ",' or whether we required thatthese sentences should have a special form. For itwduld only be a question of how we chose to applya syntactical designation ; and in either case itwould be an arbitrary choice. It has, indeed, beenargued 2 that the form in which we choose to castthese sentences may make a difference to the" stability " of the propositions they express, in thesense that it may $ive them a greater or lesser chanceof being retained in the accepted system of proposi-

r " tjber Protokollsitz e " , Erhmntnib, Band g, p. 224.I Cf. Otto Neurath, " Protokollsitze", Erhenntars, Band 3 ; and

for criticism, my article on " Verification and Experience,t, pro-ceedings of the Aistotelian Society, 1936-1, where I put for.ward amore detailed refutation of the whole of this version of the coherencetheory of truth.

rr THE CHARACTERIZATION OF SENSE-DATA 9t

tions, and so of being " true " in the only sense in

which the exponents of this theory recognize the

notion of truth. But this must be a mistake. For

if the only possible criterion of the truth of a pro-

position is its logical compatibility with other pro-

positions, then it cannot be necessary to devise a

special form for its expression in order to secure its

retention in the accepted system. It is sufficient for

us, if we wish to retain it, simply to decide to exclude

all propositions that are inconsistent with it. This

is admittedly an arbitrary procedure; but so would

be any other procedure that we could adopt. Pro-

vided that it is internally self-consistent we may' on

this view, regard as " true " atty system of proposi-

tions that we choose. And so long as the suggestion

of an appeal to the observable facts is ruled out as

meaningless, no question of justification can arise.

But suppose now, what is admitted to be possible,

that we are confronted with two mutually exclusive

sets of propositions, each of which is internally self-

consistent. Are we to say that both are independently

true ? If we do, we contradict ourselves, according

to the meaning that we ordinarily give to " truth ".

We must look therefore for some method of deciding

between these incompatible systerns. But what

method can there be if the only criterion of the

truth of any system is its internal self-consistency ?

I believe that this objection is unanswerable. An

attempt has indeed been made to answer it by saying

that theaccepted

system is that which happens to be

accredited observers, such as thetrueby

s2 FOUNDATIONS OF EMPIRICAL KNOWLEDGB rr

scientists of our era.r But now we may ask whetherthe true system is merely that which contains initself the proposition that it alone is accepted by thesescientists, or whether it is that which is so accepted,as a matter of fact. In the former case the proposedcriterion does not effect what is required of it. Foreach of several incompatible systems might containthe proposition that it alone was the accepted one,without being internally inconsistent. But to adoptthe other alternative is to abandon the coherencetheory of truth. For it involves, what the advocatesof the theory hold to be impossible, a comparison ofa proposition with the empirical facts. And if thiscomparison is allowed in the case of a propositionabout the behaviour of contemporary scientists, whynot in other cases also ? Is it not conceivable eventhat actual observation might show that con-temporary scientists sometimes made mistakes ?

19. SnwrrNcES, PRoposrrroNs, AND FAcrs

I hope that by now enough has been said to showthat the consequences of denying the possibility ofusing words to refer to empirical facts are altogetherunacceptable. But I have still to explain how thisdenial ever came to be made. One may infer thatit is due to some confusion of thought, if the fore-going argument is correct. But what is the sourceof this confusion ? I suggest that it may be found

t Cf. Carnap, " Erwiderung auf die Aufsfltze von E. Zilsel undK. Duncker ", Erhenntnis, Band 3, pp. r79-8o; and Carl Hempel," Some Remarks on Empiricism ", Analytis, vol. iii, No. 3.

n THE CHARACTERIZATION OF SENSE-DATA 93

in the formulation of the two illegitimate questions:

How is it possible for symbols to have meaning ?

and, What is it that they mean ? And I hope that

in discussing these questions I shall be able to

elucidate the meaning of such puzzling but indis-

pensable expressions as " the reference of symbols

to objects " or " the comparison of propositions with

facts ".I shall begin with the question, What does a

symbol, for example a word, or a sentence, mean ?

This is a question that we have no difficulty in

answering, so long as it is taken to refer only to

particular cases. If anyone does not understand the

meaning of a particular sentence, or word, there are

various ways of enlightening him. One method is

that of translation into another language, or into an

equivalent expression of the same language' as when

it is said that " I am thirsty " means " J'ai soif ", or

that " nephew " means tt brother's or sister's son t"

In such cases it is assumgd that the explanatory

symbols are already understood by the person for

whose benefit the translation is made. Another

behave. I may, for example, refer him to Shake-

speare's tragedy of. Othello or to Proust's Dz C6tC de

chez Swann. To this it may be objected that it isDz


possible to be jealous without behaving exactly likeeither Othello or Swann, and that what is wanted is adescription of the necessary and sufficient conditionsof jealousy, which would be tantamount to giving anexplicit definition of the word. But the fact is thatthe word " jealousy ", Iike many others, is usedwith too little precision, for it to be possible toindicate any single set of circumstances the occur-rence of which is both necessary and sufficient forits proper application. It is applied to a relativelyindefinite range of situations ; and it is for thisreason that the method of giving examples is appro-priate for making its meaning understood. Finally,there is the method of ostensive definition, which weemploy when we indicate the meaning of a symbolby correlating it with some perceived object orevent ; and this differs from the other methods inthat it does not, to be effective, require any previousunderstanding of the meaning of other symbols. Itis tempting to think that this method is logicallyindispensable, which I take to have been the viewof those empiricists who denied the possibility of" innate ideas ". But this is not so. For it islogically conceivable that people should use wordscorrectly and understand their meaning without anyprocess of learning at all. What is necessary, as wehave seen, is that a language should contain non-formal mles of meaning; but we must not confusethe employment of these rules with the process oflearning them. We have, however, good reason tosuppose that the correct use of symbols has, in fact,

rt THE CHARACTERIZATION OF SENSE-DATA es

to be learned. And so we may say that it is at leastcausally requisite that the meaning of some ex-pressions should be explained ostensively, if any areto be understood.

I have now given a description of the ways inwhich one may explain the meaning of any givensymbol. But, paradoxically, this does not enableone to answer the general question : What do symbolsmean ? For the fact is that those who ask this

assumption that " meaning " is a relation which asymbol bears to something. And the puzzle is todiscover the nature of this other term.

If we take this p:uzzle seriously, our first inclina-tion may be to try to solve it without postulatinganything beyond what we experience. The meaningof a sentence, we may say, is an empirical fact. Andwe may explain that we give the name of " empiricalfact" to whatever can be actually observed. If wesay this, we shall have, presumably, to draw a distinc-tion between those sentences that signify directlyand those that signify indirectly, and to define themeaning of the sentences of this second class interms of the meaning of the observation-sentencesto which, as we must argue, they are capable ofbeing reduced; and here already there are seriousdifficulties. But even if we suppose these difficultiesto be overcome, and at the same time set aside the


problem of the sentences that are ordinarily held toexpress a priori propositions, by saying either thatthey too are empirical, or else, more plausibly, thatthey are not meaningful in the same sense as theothers, our theory will still be untenable. And tosee that this is so, we have only to reflect that somesentences are used to express what is empiricallyfalse. Now there is no question but that thesesentences are meaningful. But what they mean can-not be empirical facts ; for, in this case, there areno such facts. Moreover, it is clear that the meaningof a sentence is independent of the truth or falsity ofwhat it is used to express. If, for example, I saythat there is a stove itt *y room and there is actuallyno stove in my room, I am expressing a falsehood;if there is a stove, f am expressing a truth ; but themeaning of my sentence remains the same, whetheror not there is actually a stove iq my room. But ifthe meaning of a sentence is the same, whether whatit expresses is true or false, and if in the case whereit expresdes a falsehood it cannot mean an empiricalfact, then it does not mean an empirical fact evenwhen it happens to express what is true. And sowe must look for some other answer to the question,What do sentences mean ?

The course that is favoured by most philosopherswho have paid attention to this question is to inventa class of would-be facts, which they call " proposi-tions ". By doing this they are able to provide averbal solution for the problem that I am discussing ;but the solution is no more than verbal. We are

rr THE CHARACTERIZATION OF SENSE-DATA e7

told that what a sentence means is a proposition;

but if we then ask what a proposition is, the only

definition available is that it is what a sentence

means. These propositions are, admittedly' not

capable of being observed; nevertheless it is some-

times held that they are objectively real, even though

they do not exist in the way that natural objects

exist. But no criterion is given by means of which

one could decide when a thing was to be accounted

" real ", in this non-empirical sense. It is not

surprising, therefore, that philosophers who regard

themselves as empiricists should find themselves

unable to attach any significance to this notion of

real propositions; and so they come to see no

alternative but to try to define the meaning of

symbols exclusively in terms of their relations to

other symbols. And thus they fall into the errors

of formalism, which I have abeady exposed.

I shall not cast about for further answers to the

question, What do symbols rnean ? For the view

that I am taking is that the reason why this problem

appears to defy solution is that there is really no

problem to solve. We cannot find " the other term

of tn" relation of meaning ", because the assumption

that meaning is a relation which somehow unites a

symbol with some other unspecified object is itself

"iron.orrr. What one is asking for when one seeks

to know the meaning of a symbol is an explanation

of the way in which the symbol is used. What

form such explanations may take in particular cases

I have already shown. But I cannot deal in the same


way with the general case, for the simple reason thatthere is no general usage to explain. There is noone thing that all symbols mean. Accordingly, ifone is asked what is the meaning of a sentence or aword, one must counter with the questions: Whatsentence ? What word ? Then, if instances aregiven, one may be able to explain what they mean.But until the question is made definite in this wavit cannot possibly be answered. It is ind.eed con-ceivable that there should be a language of suchpoverty that every one of its expressions was equi-valent to every other; and then if one were askedwhat the sentences of this language meant, onewould be able to answer, either demonstratively, orelse by setting forth a sentence of some otherlanguage, which was a translation of them all. Butthis condition is not satisfied by any language thatanybody has ever used. Thus, f can explain themeaning of any given sentence in English by oneof the methods that I have described. But I cannotexplain'what English sentences mean, in general.For their meanings do not happen all to be the same.

Analogous to the question, What do symbolsmean ? is the question, What are the objects of thevarious activities of our understandings ? What isit that we know, or doubt, or suppose, or imagine,or wonder, or believe ? It may be thought that onecan dispose of the case of knowledge by saying thatit is always facts that are known ; but this answeris not available for the other cases. For it is admittedthat one can believe , and a fortzbri suppose or imagine

rI THE CHARACTERIZATION OF SENSE-DATA e9

or doubt, what is actually false. Here again, recourseis had to the notion of a proposition ; and here againthe solution is purely verbal. For a proposition, inthis usage, is itself defined as the object of theseintellective acts. But the fact is that " believing "," imagining " and the rest, resemble " meaning " inthat they are not relations, like " loving " or " kill-irg ", that require any real object at all. The onlyway in which I can answer the question, What is itthat you doubt, or wonder, or believe ? is by givingactual examples. " I doubt whether all physicalphenomena can be explained in terms of the fieldtheory " I " I wonder whether Sir Philip Francisreally was the author of the Junius letters " ; " Ibelieve that the introduction of the mandatorysystem has been beneficial to the natives in Africa ".But so long as the question is left indefinite, it cannotbe answered. There is no one thing that peoplebelieve, any more than ther'e is one thing that allsymbols mean. And the same, it may be added, istrue of knowledge. To say that the objects of know-ledge are facts is no more illuminating than to saythat the objects of belief are propositions, except inso far as it indicates that we do not speak of knowingwhat is not actually true. For the sense of " know-ledge " which is here in question is not that of

" acquaintance " or " awareness ", but that in whichwe speak of knowing that something is the case.Consequently, the word " fact " cannot, in this usage,be regarded as a name for what can actually beobserved, but only as the equivalent of " true

roo FOUNDATIONS OF EMPIRICAL KNOWLEDGE rt

proposition ". And it would therefore be less mis-leading if we said outright that it was always a pro-position that was known. But the truth is that" knowing ", ir this sense, is not, any more than" believing ", a relation that requires any real object.If I am asked what do I know, I again can do no morethan give instances. " f know that light travels witha velocity of approximately 186,ooo miles a second."" f know that 7 is a prime number." " I know thatvultures are carnivorous." Having given suchinstances, I can then go on to classify them. I may,for example, distinguish what belongs to mathe-matics from what belongs to physics, or what isgeneral from what is particular, or what is necessaryfrom what is contingent. And the fact that suchclassifications are possible may be taken to show thatsome instances of knowledge have something more incommon than what is needed merely to constitutethem instances of knowledge. But this does notjustify the view that all instances of knowledge have acornmori feature of this kind. The fact that we areable to give a name to the " common object " of allacts or states of knowledge is not to the purpose.For to say that people always know, or believe,propositions is no more informative than to say thatthey love their beloveds, or hunt quarry, or eat food.It tells us no more thin that we believe what webelieve, and know what we know.

It must not be inferred from this that I wish tocondemn the use of the word " proposition " alto-gether. Indeed, it will have been observed that I

rr THE CHARACTERIZATION OF SENSE-DATA ror

have constantly used it myself. And it is necessaryto have such a word in order that one may be able torefer to the meaning of sentences without having tospecify them particularly. Its use may be illustratedby such an example as " In the course of his speechhe asserted a number of propositions of which I cannow remember nothing except that at least three ofthem were false ". This cannot be translated by

"texpression which contains a clause of the form " hesaid'a' and'b'and'c', and not aandnot 6 and not c",where "a." ," b" and"c" are designations of particularindicative sentences. For it is possible that he did notsay " a" ,"b" and" c" ,butmadesomeotherstatements," d r" " e" ar'd " f",which were false ; and even if he didin fact say"a" ,"b" and "c", and it is false thataorb orc, it is still consistent with the truth of " he asserted atleast three false propositions ", that these statementsshould not have been the ones in question. Thiscase is analogous to that of expressions like " some-one has left his gloves in my room ", which is notequivalent to " Smith has left his gloves in my room",even though it was in fact Smith who did leave them.For it is equally consistent with the truth of " some-one has left his gloves itt *y room " that it shouldnot have been Smith, but Jones or Robinson orsomeone else. Nor can the word " someone " bereplaced by a disjunction of names, or descriptions,of particular persons. For the number of possiblecandidates is infinite. And, for the same reason, itis not possible, in cases where the word " proposi-tion " is used in this indefinite way, to replace it by

ro2 FOUNDATIONS OF EMPIRICAL KNOWLEDGE rr

means of a disjunction of names, or descriptions, ofparticular sentences. But is not the indefinite word" sentence " itself sufficient for this usage ? Why dowe require to speak of propositions as well ? Theanswer is that we often wish to make statements whichapply not merely to a given indicative sentence, butalso to any other sentence, whether of the same or ofa different language, that has the same meaning;and our use of the word " proposition " enables usto do this concisely. Thus, to take a simple instance,if I say that the proposition " this is red " entails" this is not green ", I am not saying merely that theEnglish sentence " this is not green " can legitimatelybe derived from " this is red " ; for my assertioncould equally well be understood in terms of someother language, such as German or French. ftapplies not only to these particular English sentencesbut to any sentences, in any language, that areequivalent to them. And it is this that the use ofthe dord " proposition " conveys. In general, weuse the word " proposition " rather than " sentence "whenever we are concerned, not with the preciseform of an expression, or the fact that it belongs to aparticular language, but with its meaning. Con-sequently,we speak of propositions, and not sentences,as being true or falsd:

In this way we arrive also at the use of the word" proposition " to stand for the " object " of anintellection. For only that which is true can beknown; and only that which is capable of beingtrue or false can be supposed, or doubted, or

rI THE CHARACTERIZATION OF SENSE-DATA ro3

imagined, or believed. Here again, we require awordfor the purpose of indirect reference. For example, ifI say that my friend believes whatever he reads in hisnewspaper, I am not thereby saying " my friend be-lieves that p, or my friend believes that q ", where" p" and"q" are designationsof particular sentences,even if it happens to be the case that his newspaperdoes in fact contain the sentences "p" and"q"; for itis consistent with the truth of his believing what-ever he reads in his newspaper that " p" and "q"should not be among the sentences that it contains.Or again, if I say " he has forgotten more than Ishall ever know ", I am not specifying precisely whatit is that he has forgotten and I shall never know.No doubt if one were determined not to use anyword in the way in which the word " proposition "is used, one could always find some other form ofexpression. In the examples that I have just given,I have avoided it by the use of indefinite relativeclauses. But this course is'not always feasible. It '

will not serve us in the case of such an example as

" he asserted two propositions, which you probablybelieve but I doubt ". This might perhaps be trans-lated into " two of his assertions are probable beliefsof yours, but doubts of mine ", but I think it wouldbe generally admitted that this was not a felicitousexpression, and that the use of the word " proposi-tion " was more convenient. Nor need it give riseto any philosophical perplexlty so long as we remem-ber in using it that " meaning ", " knowing ",

" believing " and the rest are not relations, like

ro4 FOUNDATIONS OF EMPIRICAL KNOWLEDGE n

" loving " or " killing ", that require a real object,and that to say that people know, or believe or doubtpropositions, or that a proposition is what a sentencemeans, is, at best, to explain the use of the word" proposition ", and is not to make a statement offact.

With this I complete my discussion of thequestion, What do symbols mean ? The question,How is it possible for symbols to mean ? may bedealt with more shortly, for it is a product of thesame confusion as the other. Again, it is assumedthat meaning is a relation, analogous to loving orkilling ; but this time the puzzle is not to find theother term of the relation, but to discover the natureof the relation itself. What must be the connexion,it is asked, between the word " cats ", for example,and cats, or between the sentence " this is a pencil "and the fact that this is a pencil, for one to be able tosymbolize the other ? Sometimes the answer givenis th* the relation is causal. " What do we knowwhen we know that our words 'express' somethingwe see ? I see a cat and say'there is a cat'. Some-one also says'Why did you say " there is a cat " ?'and I reply, 'because I saw a cat '. The word'because' must be taken as expressing a relationwhich is, at least partly, that of cause and effect." r

Another answer is that the relation is one of structuralidentity. " The sentence is a picture of reality. Inthe picture and the pictured there must be something

t Bertrand Russell, " The Limits of Empiricism ,,, Proceedingsof thc Aristotelian Society, r935-6.

rr THE CHARACTBRIZATION OF SENSE-DATA ro5

identical in order that the one can be a picture of

the other at all. What the picture must have in

common with redity in order to be able to rePresentit after its manner - rightly or falsely - is the form

of representation. To the configuration of the

simple signs in the sentential sign corresponds the

configuration of the objects in the state of affairs.In the sentence there must be exactly as many things

distinguishable as there are in the state of affairswhich it represents." t Now it is not to be deniedthat the utterance of a symbol is very often causedby a perception of that which it symbolizes; andit is true also that there are some symbols that dohave the same structure as the states of affairs whichthey represent. But it cannot rightly be inferredthat the essence of meaning consists in either ofthese relations, if only for the reason that a symbolmay legitimately be used to express what is false.l

We cannot define the meaning of a symbol in terms

of any relation that is supposed to hold between it

and the fact which it symbolizes, for in cases whereit expresses a falsehood there is no such fact, andconsequently no such relation; and yet the symbol

is still meaningful. Moreover, even when a symbol

does symbolize what is actually observed, it is not

t Ludwig Wittgenstein, Ttactatus Logico-Philosophicus, Pro-positione 4.or, z.t6r, 2.r7, 3.2r, +.o4. I have translated " Satz"by " sentence " instead of " proposition ", which appears in theEnglish version given in the book.

t Cf. R. B. Braithwaite, " The Relation of Psycholory toLogic ", Supplmtentary Proceedings of the Ari.stotelian Society,1938 ; and my own article " On the Scope of Empirical Kntfr-ledge ", Erhenntnis, Band 7, Heft 4.


necessary thdrt its utterance should be caused by aperception of that which is symbolized, or that itshould bear any resemblance to it, whether in struc-ture or content. It does not follow from the factthat I assert that I am seeing a cat at a time when Iactually am seeing a cat that my utterance is theeffect of this perception. I may, for example, havebeen hypnotically induced to make such a statementat that particular time. And while the symbol that Iuse may take the form of a picture or a model, itcertainly need not. Given suitable conventions, Ican express the fact that I am seeing a cat by repre-senting it pictorially; and there may be circum-stances in which the choice of such a method ofrepresentation has practical advantages. For instance,if I wish to express some facts of geography, I mayfind it most convenient to do so by means of a map.But even if this method were universally applicable,and invariably advantageous, which in fact it is not,it still,would not have any logical pre-eminence.There is no reason whatever for supposing that allthe sentences of the English or any other Europeanlanguage are really, in any ordinary sense, picturesor models of what they represent. But this does notin the least prevent them from being meaningful.

The adherents of thp formalist theory of languagehave seen that a symbol need not be connected withwhat it symboiizes, either causally, or by having thesame structure, or indeed by any other naturalrelation. But they have mistakenly gone on to inferthat the only rules of meaning that can be admitted

rI THE CHARACTERIZATION OF SENSE-DATA ro7

in the characterization of a language are the formalrules by which one symbol is connected with another.Against this, I have already shown that if

^" language " is to be capable of being used as alanguage, that is to say, for the purpose of com-munication, it must also be characteized by non-formal rules, which connect some of its symbols, notwith other symbols, but with observable states ofaffairs. What is not necessary, however, is that asymbol, the use of which is determined by such anon-formal rule, should have any further connexionwith what it symbolizes beyond that which is con-stituted by the existence of the rule. The answer tothe question, How is it possible for " red " to meanred ? is simply that this happens to be the symbolthat we have chosen to use to refer to this colour.The spoken word " red " differs from words like

" whistle " or (( hiss " in that it does not in the leastresemble that which it is used to symbolize; butthis fact does not furnish any logical objection to ourusing it in the way that we do. A symbol may ormay not share a common quality with that which itsymbolizes. There is no ground whatsoever forsaying that it must.

I believe that this mistake about the nature ofmeaning is reflected in the ordinary formulation of

" the correspondence theory of truth ". For whenwe are told that a sentence expresses a true proposi-tion if and only if it is used in such a way that itcorresponds to a fact, we are inclined to interpretthe word " correspondence " literally, as implying

ro8 FOUNDATIONS OF EMPIRICAL KNOWLEDGE rr

some sort of resemblance, and then we find ourselvesconfronted with such questions as, What do sen-tences which express false propositions resemble ?and How is it that sentences, which express what wehave every reason for supposing to be true proposi-tions, do not appear in any way to resemble therelevant facts ? - and to these questions we areunable to give any satisfactory answer. But suchproblems do not arise, once it is made clear that theword " correspondence ", if it is to be used at allin this connexion, must not be understood literally.To say that I am using the sentence " there is a match-box on my table " to correspond to the fact that thereis a match-box on my table, or to express a proposi-tion which corresponds to this fact, is to say no morethan that I am using the words " there is a match-box on my table " to mean that there is a match-boxon my table, and there is a match-box on my table.But how do I discover that there is a match-box onmy,table ? How is it to be determined that anyempirical proposition does, in this sense, correspondto a fact ? The answer is that, in the last resort,it is always to be determined by actual observation.I say " in the last resort " because it is necessaryhere to draw a distinction between propositions thetruth of which is determined directly by observation,and those that are verified indirectly. One's groundfor believing a given proposition is often, in the firstinstance, the truth of a second proposition which isevidence for it; and one's ground for believing thesecond proposition may, in its turn, be the truth of

rt THE CHARACTERIZATION OF SENSE-DATA roe

a third; but this series cannot be prolonged in-definitely. In the end it must include at least oneproposition that is believed, not merely on theground that it is supported by other propositions,but in virtue of what is actually observed. For, asI have already shown, we are not entitled to regarda set of propositions as true merely because they sup-port one another. In order that we should havereason to accept any of them, it is necessary that atIeast one of their number should be directly verifiedby observation of an empirical fact.

Is it now possible for us to delimit the class ofpropositions that are capable of being directlyverified ? The only means that I can see of doingthis is to say that a proposition is capable of beingdirectly verified when it is expressed by a sentencethe meaning of which is determined by a non-formalrule. But the consequence of this is that the question,whether a given proposition is or is not capable ofbeing directly verified, does not admit of a straight-forward answer. We must say that it depends uponthe language in which the proposition is expressed.If, for example, we have agreed to use the sense-datum language, we shall have to say that proposi-tions like " this is a match-box " or " this is apencil " are not directly verifiable.t For we musthold that the meaning of sentenceswhich express suchpropositions is to be determined by reference to

t This must not be understood to imply that the validity ofsuch propositions consists in anything other than the occurrenceof the relevant sense-data. Cf. section zz of this book.

rro FOUNDATIONS OF EMPIRICAL KNOWLEDGE rr

sentences which designate sense-data, and that it isonly when a sentence explicitly designates a sense-data that its meaning is determined by reference tofact. On the other hand, when we are teachingEnglish to a child, we imply that propositions aboutmaterial things can be directly verified. For we donot then explain the meaning of sentences like " thisis a match-box " or " this is a pencil " in terms ofsentences which designate sense-data. We indicate itostensively. But whereas the meaning of a sentencewhich refers to a sense-datum is precisely determinedby the rule that correlates it with the sense-datum inquestion, such precision is not attainable in the caseof a sentence which refers to a material thing. For theproposition which such a sentence expresses differsfrom a proposition about a sense-datum in that thereare no observable facts that constitute both a neces-sary and sufficient condition of its truth. We mayindicate the meaning of a sentence of this type bygiving a sample of the kind of evidence that directlysuppbrts the truth of the proposition it expresses, butwe do not intend to suggest that this evidence isexhaustive. Suppose, for example, that I teachsomeone the meaning of the sentence " this is amatch-box " by actually showing him a match-box,opening it, taking out and striking a match, and soforth. If, subsequehtly, he says " this is a match-box ", and there is in fact no 'match-box there, itdoes not follow that he has failed to apprehend therule that I tried to teach him. His error may be, notthat he has mistaken the meaning of the sentence

rr THE CHARACTERIZATION OF SENSE-DATA rrr

he is using, but that, having acquired some evidencein favour of the proposition which it expresses, hehas wrongly assumed the possibility of increasing thesum of this favourable evidence indefinitely. If hethen experiments further and finds the new evidenceunfavourable, he may admit that when he used thesentence " this is a match-box " he was expressing afalse proposition; but this does not mean that hewas not justified in using it, in view of the evidencewhich he had. To be deceived by one's senses isnot necessarily to be mistaken about the meaningof words.

My conclusion is, then, that we can have noreason to believe in the truth of any set of empiricalpropositions unless at least one of them can bedirectly verified; and that, for a proposition to bedirectly verifiable, it is necessary that the meaning ofthe sentence which expresses it should be determinedby correlation with some observable state of affairs,though this correlation need not be univocal. Wherethe correlation is univocal, as in the case of a sentencewhich refers to a sense-datum, it is possible to bemistaken about the truth of the proposition expressed,so long as one is not actually observing the relevantfact. But there is no such possibility of error whenthe sense-datum in question is actually being sensed.For in that case the use of the sentence is prescribedby a rule of the language; so that to make anassertion that does not correspond to the fact iseither to tell a deliberate lie or else to make a verbalmistake. It is for this reason that philosophers have

rt2 FOUNDATIONS OF EMPIRICAL KNOWLEDGE u

held, as we have seen, that sentences of this kindexpress " incorrigible " propositions; and so, sinceany process of verification can be described in termsof the sense-datum language, they have inferredthat unless it is possible for some empirical proposi-tions to be " incorrigible ", there can be no groundfor supposing any empirical proposition to be true.And if they are willing to define an " incorrigible "proposition as one that is expressed by a sentencethe denial of which is a contravention of a non-formalrule of meaning, their conclusion is correct. For itis then a way of saying that if we are to be able toexpress any proposition that is capable of beingverified, it is necessary that, besides the rules whichcorrelate symbols with other symbols, our languageshould also contain rules of meaning, which correlatesymbols with observable facts. It implies that wecannot, as I have shown, .obtain a satisfactorycriterion of truth merely by setting up a formalcalculus for deriving sentences from one another, andthen paying lip-service to empiricism by calling someof them observation-sentences. Accordingly, if oneis to maintain the principle that no sentence can besaid to express an empirical proposition unless someobservation-sentence is derivable from it, it must bemade clear that the gruth of the propositions that areexpressed by these observation-sentences is deter-mined, not by merely formal criteria, but by the factthat they correspond directly, in the sense I haveindicated, to what can be obserryed.

Ir THE CHARACTERIZATION OF SENSE-DATA rr3

rr. Tnp Neruns oF THE " GIvEN "

It follows from what I have said that the questionswhich Professor Carnap classifies as " problems of theso-called given or primitive data "' do not, as hemaintains, fall entirely within the scope of logicalsyntax. His view is that the question " What objectsare the elements of given, direct experience ? " isreally verbal. He thinks that it is equivalent toasking " What kinds of word occur in observation-sentences ? ", and that the answer to this questiondepends wholly upon one's choice of language.Consequently, he holds that it may be decided byconvention whether " the elements that are directlygiven are the simplest sensations and feelings ", orwhether they are " more complex objects, such aspartialgestalts of. single sensory fields " ; for, accord-ing to him, these are not alternative theses about anempirical matter of fact, but'alternative suggestions

about the forms of observation-sentences, and it isopen to us to give our observation-sentences anyform that we please.' But this reasoning is fallacious.

The choice of the language of sense-data to describe

what we observe, rather than the language of appear-

ing, or the language of multiple location, is indeed

conventional; and it is a matter of convention that

in referring to sense-data we should use the particular

signs tJrat we do. But it does not follow from this

that the propositions which are intended to describe

t The Logical Syntax of Language, pp.305-6.

' The Unity of Science, pp. 4S-7.

rr4 FOUNDATIONS OF EMPIRICAL KNOWLEDGE rr

the characteristics of sense-data are true only by con-vention. For sense-data can have properties otherthan those that belong to them by definition; andto describe these properties is not to express a ruleof language, but to make a statement of fact. It istrue that, so long as it is framed in purely generalterms, the question, What do people observe ? doesadmit only of a verbal answer. For there is no onecommon quality that is shared by everything thatis observed. But this does not prove that sense-datacannot be brought under any empirical classifications,or that the question, What do people observe ?cannot be answered factually when it is understoodto refer to particular cases. In this respect it maybe compared with the question, What do peopleeat ? If we answer this by saying that people eatfood, we are treating it merely as a verbal question;for if we are asked what we mean by the word" food ", w€ have to reply that we use it to refer towhatever people eat. But it does not follow from thisthat the subject of food cannot be treated as anempirical science. We should think a philosophervery silly who maintained that all problems aboutnutrition were purely verbal, on the ground that theycould be reformulated as questions about the wordstfrat occurred in nutritiop-sentences. But his argu-ment would be exactly on a level with that whichCarnap uses to dispose of the " problems of theso-called given ". Such questions as whether thegestalt or the atomic theory more adequately describesthe nature of our visual sense-fields must be decided,

II THE CHARACTERIZATION OF SENSE.DATA II5

not by a verbal convention, but by an examination ofthe empirical evidence.

The analogy on which I have just drawn has beenused also by Professor Prichard. " Grant ", he says,

" for the sake of argument that on some occasionI am apprehending in the form of perceiving aparticular sound, a particular colour, and a particularfeeling of roughness. Then, no doubt, any one ofthem is being thus apprehended by me. Neverthe-less it is not a something which is being thusapprehended. If I am eating a number of things,say, some cheese, some bread, and some salt, theytogether form a certain numerical group, viz. thetotality of things which I am eating. But theirmembership of this group does not constitute themthings having a certain cornmon character. . . .The things which I am eating are united simply bymy eating them; and my eating them does notconstitute them things of a certain sort. Indeed tospeak of a something which is,being eaten by me,or of something which is being eaten by someoneis merely verbal, because to be being eaten is nota character of anything. Similarly the colour, thesound and the feeling of roughness which I am thusapprehending are united solely by *y thus appre-hending them ; and though each is one of the thingswhich are being thus apprehended by me, none is asomething which is being thus apprehended by me.There is no such thing as a thing which is beingthus apprehended by me, nor again such a thingas a thing which is being thus apprehended by

rt6 FOUNDATIONS OF EMPIRICAL KNOWLBDGE u

someone."' The argument is not very clearlystated; but I take it that Prichard is maintainingthat what we may call the relational property of" being apprehended " resembles the relationalproperty of " being eaten " in that the possession ofit does not logically entail the possession of anyspecial quality; and with this I am prepared toagree. And if in saying that " to speak of a some-thing which is being eaten by someone is purelyverbal " he means that the proposition " people eatfood " is analpic, I agree with this also; and Iagree that the same applies to the proposition thatpeople apprehend sense-data. But it does not follow,as Prichard seems to think, that " there could notbe such a thing as a sense-datum ", xDY more thanit follows that there could not be such a thing asfood. As I use the word " sense-datum ", to denythat there were sense-data would be to deny thatanything ever was observed; and I cannot believethat Professor Prichard, or any other philosopher,really wishes to deny this. Admittedly, the factswhich one seeks to describe by referring to sense-data could also be described in some other ter-minology. But a proposition does not cease to betrue merely because there is more than one way ofexpressing it. .1

It should by now be clear that in saying that thereare sense-data I am not either assuming or rejectingany special empirical theory about the. nature of

I " The Sense-datum Fallacy ", Supplmuntary Proceedings ofthe Aristotelian Society, 1938, pp. r4-r5.

rr THE CHARACTERIZATION OF SENSE-DATA rr7

what we observe. Nor do I wish to defend all that

the advocates of sense-data have said of them; for

I have already pointed out that their use of the term

has not always been clear or even consistent. One

source of confusion to which I have referred is the

inclination to treat sense-data as if they were a

species of material things. Another with which I

must now deal is the use of a physical rather than a

psychological criterion to determine what it is that

we directly observe. This leads on some occasions

to the view that more is given to us than we are

actually aware of I on other occasioRs, to the view

that only a part of what we apprehend is " reallygiven ". The assumption, in the former case, is that

sense-data can be experienced by us without our

noticing them, and that they can have sensibleproperties that we do not immediately detect; in

the other case, it is that, since the character of what

we observe is affected by our conscious orunconscious

memories of our past experience, it is necessary to

discount these associations in order to discover whatsense-data are really being sensed. If we accept thefirst of these assumptions, we have to allow that

sense-data can have properties that they do not

appear to have. If we accept the second, we have

to allow that they can appear to have properties that

they do not really have. In both cases, therefore, the

tendency is again to assimilate sense-data to material

things. But, as we have seen, the utility of the

sense-datum language depends upon our being able

to make the distinction between sense-data and

rr8 FOUNDATIONS OF BMPIRICAL KNOWLEDGE rr

material things as sharp as possible. We must try,therefore, to avoid these conclusions if we legiti-mately can.

Let us consider first what reason there can be forholding that not all of what we seem to experienceis really given. The cases to which this view isapplied are primarily those in which our actualobservations are at variance with some physiologicaltheory. One might, for example, be inclined tosuppose that the apparent sizes of objects variedwith their distance from the observer, or that thecolours that they appeared to display varied inaccordance with their illumination, or that a changein the position of an object, in relation to theobserver, would involve a change in its apparentshape. But these theories are not completely borneout by the empirical facts. We find that a man looksvery much the same size at a distance of ten yardsas at a distance of five ; and that though, to quoteProfessor Kdhler, " a simple consideration of geo-metricil optics will tell you that during the man'sapproach his [apparent] height must have doubledand his breadth too, so that his total size must havebecome four times the amount it was. at ten ]ards ',.rWe find that a white paper that is seen in the shadowof a screen does not appear the same colour as ablack paper that is sedh in a full light, even thoughthe amount of light that the two pdpers reflect is thesame. The images that they throw upon the retinaof the observer may be equally intense; but the

t W. Ktihler, Gestalt Psychology,p,36.

u THE CHARACTERIZATION OF SENSE-DATA tte

white paper still appears to him to be white, and the

black paper, black.' And this constancy of appear-

ances extends also to shapes. Though philosophers

seem inclined to assume that a round object, when

it is seen obliquely, always looks elliptical, the

empirical fact is that it usually does not. If I look

sideways at a coin, the image that it projects upon

my retina is indeed elliptical, but in spite of that, the

coin still seems to me to be round. And by this I do

not mean that, in spite of seeing it as elliptical,I judge

the coin to be really round, but that it is a round

and not an elliptical shape that I seem actually to see.

Now it is sometimes suggested that the explana-

tion of such phenomena as these is that we doctor our

impressions in the light of our past experience. When

I look at the coin obliquely, the sense-datum that I

am sensing is, it is said, in itself elliptical; at the

same time I know that if I were looking at the coin

vertically, my sense-datum would be circular, and I

take this as evidence that tlre coin really is circular

in shape ; accordingly, I subconsciously correct my

elliptical sense-datum so as to bring it into accord

with the " real " shape of the coin. And the other

examples are dealt with in the same way. In each

case it is assumed that we subconsciously make

allowances for the abnormal conditions, and thereby

transform the character of the sense-data that we

actually sense. In support of this theory it is argued

that we can discover what sense-data are really given

in these cases by so arranging the conditions thatr Cf. Kdtrler, op. cit. p. 58.

r2o FOUNDATIONS OF EMPIRICAL KNOWLEDGE rt

we are robbed of our motives for correcting them.Thus, if I know that the white paper is in shadow,and that the black paper is fully exposed to the light,it is held that I allow this knowledge to affect thecharacter of my sense-data, and that it is for thisreason that the difference in the colour of the twopieces of paper appears to be reproduced in my sense-data, even though the papers are actually reflectingan equal amount of light. But if I isolate the piecesof paper from their visual context by, for example,interposing a perforated piece of grey cardboard,then, if the papers reflect the same amount of light,the colours of the sense-data by which they arepresented will appear approximately the same. Andfrom this it is inferred that even when the pieces ofpaper are not thus isolated, the sense-data that areactually given are really similar in colour; and thatthe difference we appear to see between them is apsychological accretion.

Against this, I may begin by remarking that evenas a psychological hypothesis it is open to seriousobjection. Those who advance it assume that byisolating objects from their actual context, and pre-senting them in a homogeneous medium, one is ableto discover how they would appear to someone whosesensations had not been influenced by past experi-ence. But if, as the gestalt psychglogists maintain," the properties of any part of a sense-field dependnormally upon the conditions given in the wholefield, or, at least, in a larger area of it ",t 1hi. assump-

t W. Kdhler, Gestalt Psychology,p, Tz.

rr THE CHARACTERIZATION OF SENSE-DATA r2r

tion is unjustified. And experiments such as the one

that I have just described favour thegestalt hypothesisjust as much as the other. The fact that the white

paper in one case appears to be similar in colour to

the black, and in the other case does not, though in

each case the two reflect the same amount of light'

proves that our sensations of brightness are not

always completely determined by the intensity of

the corresponding retinal images. But there is

nothing in the evidence to show that the additional

factor is to be sought in the influence of past experi-

ence rather than in the character of the present

environment. Admittedly, there is good empirical

evidence for holding that our past experience does

very often affect the character of our sense-data. To

take a simple example, one finds when one learns a

foreign language that its signs do sensibly alter their

look and sound, as one grows more familiar with

their meaning. But it cannot be inferred from this

that whenever our sense-data'appear to have different

properties from those that we should expect them

to have, if their character depended exclusively on

" local stimulation ", the difference can be wholly

accounted for by supposing that we correct them in

the light of our past experience. Not only can we

not simply rule out the gestalt hypothesis ; we are

not entitled to assume even that the influence of the

subjective attitude of the observer, in determining

the character of his sense-data, must be confined to

this particular process of correction. The physio-

logical theory has the advantage of simplicity; but

r2z FOUNDATIONS OF EMPIRICAL KNOWLEDGE rr

I am not myself convinced that it is adequate to thefacts.

In the present context, however, the importantpoint is that even if this theory were adequate to thefacts, it still would not make it necessary, or evenjustifiable, for us to draw a distinction between theproperties that sense-data appear to have and thosethat they really have. Here again the matter at issueis not factual but linguistic. It relates, not to theempirical characteristics of the phenomena in ques-tion, but only to the choice of a method for describingthem. In this case it is suggested that we shouldtake as our criterion of " reality ", with regard tothe properties of sense-data, not what we actuallyobserve, but what it is supposed that we should beobserving if it were not for the influence of our pastexperience. But though this criterion may be work-able, it is surely very inconvenient. If we intendto analyse the " reality " of material things in termsof hypothetical propositions about sense-data, it canonly lead to confusion if we bring in another classof hypothetical propositions for the purpose ofdefining the " reality " of the properties of sense-data themselves. Furthermore, as I have alreadypointed out, the advantage of having the sense-datumterminology is very much diminished if one allowsthe distinction between appearance and reality toapply to sense-data, as well as to material things ;for one of its main recommendations is that it issupposed to enable us to escape the pvzzlements towhich this distinction gives rise. And if, rur seenrs

n THE CHARACTERIZATION OF SENSE-DATA r23

probable, there is in fact no simple method of dis-

counting the influence of past experience upon our

sensations, the choice of this criterion would involve

the further difficulty that we should very seldom be

able to say what the properties of our sense-data

really were. These objections would indeed have

to be disregarded if there were no other wa! of

describing the relevant facts. But this is not the

case. If one makes it a rule that all sense-data really

have the properties that they appear to have, nomatter what may be the causes of their appearing asthey do, one is not in any way debarred from ex-pressing the fact that the character of one's sense-data may be affected by one's past experience. Onemay express it by saying that, given a set of conditionswhich includes the effects of past experience, the

sense-data that a person senses may be different fromthose that he would be sensing if the relevantconditions did not include this factor. But I do notinfer from this that the properties which would beaffected by the alteration in the conditions do notreally charucterize the sense-data which they appearto characterize. For this would be to admit a dis-tinction to which I attach no meaning when it isapplied to sense-data. According to the conventionwhich I am adopting (and I have made it clear thatthere is nothing here at issue except a choice of con-ventions) in the domain of sense-data whateverappears is real.

To show how this convention operates, I shallnow make use of an example which was given to me,

r24 FOUNDATIONS OF EMPIRICAL KNOWLEDGB rr

in the form of a problem, by Mr. Gilbert Ryle. Itis a popular belief that when a man is struck withsufficient force he may have an experience which isdescribed as the experience ofseeing stars. Supposethat this happens to someone we know and that weask him how many stars he saw. He may have toreply that while he was conscious of seeing stars hewas not aware of them as constituting a definitenumber. We may then be tempted to assert that ifhe saw stars at all he must have seen a definitenumber; but that perhaps he was too dizzy tocountthem. But this would be a mistake. We shouldbe assuming unjustifiably that what could be saidabout material things could also be said about sense-data. It is true that if the stars in question werephysical objects, there would be a definite numberof them, whether the observer was aware of it ornot. But if we use the expressioD. " seeing stars " torefer only to the sensing of certain sense-data, then,if the sense-data do not appear to be enumerable,they really are not enumerable. No doubt, if ourobserver had been less dizzy he might have seen, notjust a mass of stars, but a definite number. But thisdoes not imply that his experience was not what hetook it to be ; that he was really seeing a definitenumber of stars, though all that he was consciousof was an indefinite mass. It implies only that indifferent conditions he might have had a qualitativelydifferent experience. And the situation would bethe same if we assumed that he was looking at thestars, in the ordinary sense. So long as we were

Ir THE CHARACTERIZATION OF SENSE-DATA t2s

referring to them as material things, it would bemeaningful for us to say that he was really seeing adefinite number of stars although he was not con-scious of it ; but no such statement could significantlybe made about the sense-databy means of which the

stars were presented to him. For a group of sense-data can be said to be enumerable only if it is"infact enumerated. And to say that it might have

been enumerated, though actually it was not, is

not to say that it had any undetected property,

but only that some other group, which would

have been enumerable, might have occurred in itsplace.

By refusing to draw a distinction between that

which is really given to an observer and that which

he is aware of, I exclude also the possibility of their

being sense-data whose existence is not noticed at

the time that they are sensed. To illustrate the kind

of fact that makes philosophers inclined to admit thispossibility I may take an example from the work of

Professor Broad.t He points out that we sometimeshave the experience of " looking for something,failing to find it, and yet discovering afterwards thatit had been staring one in the face in the very drawerin which we have been looking ", and he goes on toconsider whether in such a case there are goodgrounds for saying that we were unconsciously per-

ceiving the object in question. He states the argu-ments in favour of this view as follows: " If I had

recognized at the time that I was perceiving ther The Mind and its Place in Nature, pp. 4ro-r5.

n2


object, I should certainly have found it. As I didnot find it, it seems reasonable to suppose either thatI was not perceiving it at all or tJrat, if I was, thisperception was not noticed by me. Now, if itexisted, it is hardly likely to have escaped my noticeby mere inadvertence ; for this was the veryexperience which I was wanting and expecting at thetime. . . . [Brt] the object was in such a positionthat light from it must have affected the central partof my retina; and, therefore, it is very unlikely thatit did not produce a perceptual experience at all.Lastly, it might be that in some cases we could addto this presumption something further of the follow-ing kind : Last night I dreamed of the object in acertain place in the drawer; and when I went thismorning and looked again, there it was. Or again:I was hypnotized afterwards and told the hypnotistwhere the object was and he found it there. Weshould then have a pretty strong case, superficiallyat zrry rate, for the view that I had a literally uncon-scious perception of this object when I was lookingfor it in the drawer."

Professor Broad himself denies that these arsr-ments establish the desired conclusion. He admitsthat they " make it highly probable that, if a percep-tion existed at the time, it was literally unconsciousrelatively to the mind which was.then controlling thebody at any rate ". But he does not see that thereis " any reason to believe that a perception of thisobject existed at all ". He says, however, that itseems to him likely that in cases of this kind " there

rr THE CHARACTERIZATION OF SENSE-DATA rz7

really was a sensum in the visual field, which was,in fact, an appearance of the object that we wereseeking; but that for some reason the traces whichwould usually be excited under such conditionswere not excited, or, if they were, failed to producetheir normal effect. This sensum was therefore notselected and discriminated from the rest of the field,and was not recognized as an appearanbe of the objectfor which we were looking." But if this is his viewit is doubtful whether he really disagrees with thosewho maintain that the object in question was un-consciously perceived. He denies that it \Masunconsciously perceived because he uses the word" perception " in such a way that an object cannot besaid to be perceived, even unconsciously, unless somesensum is actually recognized as an appearance of it.But I doubt if those who believe in the existence ofunconscious perceptions do understand the word

" perception " in this way. I think that in their view,to admit that there really was a sensum in the visualfield which was in fact an appearance of the object,although not recognized as such, would be tantamountto admitting that the object was unconsciouslyperceived.

For my part, I do not make this admission.Assuming that the word " sensum " is understoodto mean what I mean by " sense-datum ", I simplydeny that there really was such a sensum in the visualfield. My ground for this denial is admittedly con-ventional. I do not attach any meaning to thestatement that among the sense'data that made up

r28 FOUNDATIONS OF EMPIRICAL KNOWLEDGB rI

an experienced sense-field there existed one that wasnot sensed. But it must be understood that, inadhering to this convention, I am not denying anyempirical fact. The empirical facts are that theobject was in such a physical relation to the obseryer'sbody that in normal circumstances we should expecthim to have perceived it, and that subsequently heseems to remember that he did perceive it. If, inspite of this, he was not, at the time when he wassupposed to be perceiving it, aware of any sense-datum that belonged to the object, it may be con-cluded, first, that one's awareness of visual sense-datais not wholly determined by such physical factors asthe affection of the retina, and secondly, that it ispossible to have the experience of seeming to remem-ber what one has, in fact, not previously sensed. Tomaintain that there must really have been a sense-datum of the object in the obseryer's visual field,althorlgh he was not conscious of it, is not to accountfor the facts in any way, but is merely to adopt analternative criterion of what is to constitute theexistence of a sense-datum. It is to suggest that,instead of making our awareness of a sense-datumthe only criterion of its existence, we should say alsothat a sense-datum exists, or at least that it hasexisted, when certainlhysical conditions are fulfilled,or when we have the experience of seeming to remem-ber it. But I do not see any reason to accept eitherof these proposals. The use of a physiologicalcriterion for determining the existence of sense-data is objectionable because it tempts us to fall into

II THE CI{ARACTERIZATION OF SENSE-DATA rzg

the confusions that result from treating sense-data

as a species of material things. And while the

adoption of the memory criterion may not in itself

be a source of serious confusion, it introduces a com-plication into the sense-datum language without,so far as I can see, securing any compensatoryadvantage.

The question which remains still to be considered

is whether sense-data can ever really have properties

that they do not appear to have. It has been asserted,for example by Professor Broad, that while " sensa

cannot appear to have properties they do not really

have, there is no reason why they should not have

more properties than we can or do notice in them ".

"'We must distinguish ", he says, " between failing

to notice what is present in an object and' noticing'

what is not present in an object. The former

presents no special difficulty. There may well be

an object which is too minute and obscure for us to

recognize distinctly. Agiin, it is obvious that we

may sense an object without necessarily being aware

of all its relations even to another object that we

sense at the same time. Consequently, there is no

difficulty whatever in supposing that sensa may be

much more differentiated than we think them to be,

and that two sensa may really differ in quality when

we think that they are exactly alike." I The second

of these points has been made more strongly by

Bertrand Russell. " It is important to realize " ,he says, " that two sense-data may be, and must

t Scientific Thought, p. 244.

r3o FOUNDATIONS OF EMPIRICAL KNOWLEDGE tr

sometimes be, really different when we cannot per-ceive any difference between them." t

The ground upon which it is maintained that " asense-datum with which f am acquainted may per-fectly well have parts with which I am notacquainted rr z is, presumably, that if I scrutinize thesense-datum more carefully, or examine it with thehelp of instruments, I may detect features in it thatI have not been able to detect before. But why, wemust ask, should it be supposed that the sense-datumwhich I sense in these new conditions is the same asthat which I was sensing before they were intro-duced ? The answer, in this case also, depends uponthe choice of a convention. If one defines " thegiven " physiologically, then it is reasonable topostulate that a change in the attitude of the observershall not be regarded as involving any change in thenature of the sense-datum that he observes. But ifone rejects the purely physiological criterion, as fhave fdund reason to do, it is preferable to say, notthat a more exact scrutiny reveals parts of a sense-datum that were not apparent before, but rather thatit causes one sense-datum to be replaced by anotherwhich resembles the first but is more highly differenti-ated. Thus, while it can be true of a material thingthat it is too minute afrd obscure for us to recognizeproperly, I deny that this can be true of a sense-datum. By looking through a microscope I may dis-

I Otn Knoutledge of the External Worlil, p. r4r.. C. D. Broad, " Is there 'Knowledge by Acquaintance'? ,'

Aristotelian S ociety Supplmtentary Pro ceedings, rgtg, p. zr8.

II THE CHARACTERIZATION OF SENSE.DATA I3I

cover that some material thing has properties that Ihave not previously detected in it ; but I should notexpress any fact of this kind by saying that I foundsome of my sense-data to have properties that theydid not previously appear to have. For while thematerial thittg remains the same whether or not Imake use of a microscope to observe it, the sense-data do not. And the reason for this is that I chooseto say, not that the use of the microscope enablespeople to detect new features of sense-data whichtJrey were already apprehending, but that it leadsthem to apprehend new sense-data. I adopt theconvention that reinforces the distinction betweensense-data and material things, in preference to onethat would encourage us to confuse them.

In support of his view that two sense-data mustsometimes be really different when we cannot per-ceive any difference between them, Russell arguesas follows : " In all cases of sense-data capable ofgradual change, we mdy find one sense-datumindistinguishable from another, and that otherindistinguishable from a third, while yet the first andthird are quite easily distinguishable. Suppose, forexample, a person with his eyes shut is holding aweight in his hand, and someone noiselessly adds asmall extra weight. If the extra weight is smallenough, no difference will be perceived in the sensa-tion. After a time, another small extra weight maybe added, and still no change will be perceived ; butif both extra weights had been added at once, it maybe that the change would be quite easily perceptible.

t32 FOUNDATIONS OF EMPIRICAL KNOWLEDGE n

Or, again, take shades of colour. It would be easyto find three stuffs of such closely similar shades thatno difference could be perceived between the firstand second, nor yet between the second and third,while yet the first and third would be distinguishable.In such a case, the second shade cannot be the sameas the first, or it would be distinguishable fromthe third; nor the same as the third, or it would bedistinguishable from the first. It must therefore,though indistinguishable from both, be really inter-mediate between them." r

There is no reason to dispute the empirical factsthat are set forth in this argument ; but the con-clusion Russell draws does not follow from them.The question is whether the relation of exactresemblance, in respect of colour, or weight, or anyother sensible property, is to be regarded as applyingto sense-data in the same way as it applies to materialthings.- When it is applied to material things thisrelatiori is understood to be both symmetrical andtransitive. That is to say, if a material thing z4 ischaracterized, for example, by exactly the same shadeof colour as another material thing B, then it followsnot only that B is characterized by exactly the sameshade of colour as A, but also that if B is of exactlythe same shade as a third thing C, A too is of exactlythe same shade as C. But it muSt not be assumedthat this principle will hold good also for sense-data.For the logical counterpart of a relation betweensense-data is not a relation that really characterizes

t Our Knowledge of the External World, pp, r4r-2,

rt THE CHARACTERIZATION OF SENSE-DATA r33

material things, but only one that appears to charac-terize them. To say that a sense-datum a, whichbelongs to a material thing A,has the relation R to asense-datum b, which belongs to a material thing B,is not to say that A really has R to B, but only thatit appears to have R to B. Accordingly, the modelthat we must take for the relation of exact resemblancebetween sense-data is not the relation of exactresemblance between material things, but the relationbetween material things of appearing to be exactlyresemblant; and this relation, though it too is sym-metrical, is not transitive. For it is admitted that amaterial thing A can appear exactly to resemble asecond thing B, and that B can appear exactly toresemble a third thing C, in respect of some sensibleproperty, without its being the case that A and Cappear in this respect exactly to resemble one another.We must therefore conclude that the relation ofexact resemblance is not a transitive relation when itis applied to sense-data. Oi, if it be objected that anon-transitive relation of exact resemblance is a con-tradiction in terms, we must say that the relation ofexact resemblance does not apply at all to sense-data,and then we must find some other name for therelation that obtains between two sense-data whenthe corresponding material things appear exactly toresemble one another. In either case, we destroythe ground for holding that sense-data must some-times differ in ways that are not apparent. For theassumption that there really obtains between sense-data, as between material things, a relation of exact

r34 FOUNDATIONS OF EMPIRICAL KNOWLEDGE rr

resemblance which is both symmetrical and transitiveis indispensable to Russell's argument.

We see then that the facts that have led certainphilosophers to hold that the distinction betweenappearance and reality must apply to sense-data, aswell as to material things, do not necessitate thisconclusion. It is indeed open to these philosophersto formulate the rules of the sense-datum languagein such a way as to admit this distinction, if theychoose ; but I have shown that it is much more con-venient to exclude it. From the conventions whichI am adopting it follows that one's awareness of asense-datum cannot be delusive in the way that one'sperception of a material thing can. If one knowswhat properties a sense-datum appears to have, oneknows what properties it really has. This does notprevent us, however, from expressing false proposi-tions even about the sense-data that we are actuallysensing. For it is always possible, whether from anintentidn to deceive or else because of a verbal error,to misdescribe the properties that they appear to have.

A further point in favour of this way of conceiving" the given " is that it imposes no a priori restrictionsupon the possible scope of perceptual knowledge. Itmight be thought that in sayrng that perceiving amaterial thing always fnvolved being directly awareof a sense'datum, I was making it a necessary condi-tion for a material thing to be veridically perceivedthat it should exercise some physical stimulus upona sense-organ, and was thus excluding the possi-bility of clairvoyance as a method of perception.

tr THE CHARACTERIZATION OF SENSE-DATA t3S

But this is not the case. It may indeed be questionedwhether much of what is now referred to as " extra-sensory perception " t ought properly to be called" perception " at all. I at least should prefer not toextend the use of the term " perception " to coverthose cases in which the object that is supposed tobe perceived is not located by the percipient any-where in physical space. But if there are experiencesthat are held to be extra-s€rsorlr not because theyseem to differ intrinsically from sensory perceptions,but because of a difference in the conditions of theiroccurrence, then I think they may reasonably becounted as perceptions. And if one allows them tobe perceptions, then one may describe them in termsof sense-data. For I am using the word " sense-datum " in such a way that in every case in which itcan legitimately be said that a material thing is beingperceived it can also be said that a sense-datum isbeing sensed. It makes no difference what the causesof the perception may be. ' In determining theexistence and character of sense-data we must con-sider only what it is that is actually observed ; thequestion how it comes to be observed is irrelevant.In this way we ensure that " the theory of sense-data" does not involve more than the elaboration ofa special terminology for describing our perceptualexperience. It must not be regarded as presupposingthe validity of any particular theory either about thecauses or about the character of what we perceive orsense.

I Cf. J.B. Rhine, Extra-Sensory Perception.

Ayer the Caracterization of Sense-data

Documents

Transcript of Ayer the Caracterization of Sense-data