A Purely Syntactical Definition of Confirmation

Author(s): Carl G. Hempel

Source: The Journal of Symbolic Logic, Vol. 8, No. 4 (Dec., 1943), pp. 122-143
Published by: Association for Symbolic Logic
Stable URL: http://www.jstor.org/stable/2271053
Accessed: 19/07/2010 22:57

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=asl.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

Association for Symbolic Logic is collaborating with JSTOR to digitize, preserve and extend access to The
Journal of Symbolic Logic.

http://www.jstor.org
TiEc JOuiRNAL OF SYMBOLIC LOGIC
Volume 8, Number4, December1943

A PURELY SYNTACTICAL DEFINITION OF CONFIRMATION

CARL G. HEMPEL

1. Objective of the paper. The conceptof confirmation occupies a central

position in the methodologyof empirical science. For it is the distinctive
characteristicof an empiricalhypothesisto be amenable, at least in principle,
to a test based on suitable observationsor experiments;the empiricaldata
obtained in a test-or, as we shall preferto saty,the observationsentences
describingthose data-may then eitherconfirmar disconfirm the given hypo-
thesis,or theymay be neutralwithrespectto it. To say that certainobserva-
tion sentencesconfirmor disconfirm of
a hypothesis,does not, course,generally
mean that those observationsentencessufficestrictlyto prove or to refutethe
hypothesisin question,but ratherthattheyconstituteavorable, or unfavora-ble,
evidence for it; and the term "neutral" is to indicate that the observation
sentencesare eitherentirelyirrelevantto the hypothesis,or at least insufficient
to strengthenor weaken it.
A precise definitionof the concepts of confirmation, disconfirmation, and
neutralitywhichhave just been looselycharacterizedappears to be indispensable
for an adequate treatmentof several fundamentalproblemsin the logic and
methodologyof empiricalscience; these problemsincludethe elucidationof the
logical structureof scientifictests, explanations,and predictions,the logical
analysisoftheso-calledinductivemethod,and a rigorousstatementand develop-
mentoftheoperationalistand empiricistconceptionsofmeaningand knowledge.2
However, no generaldefinitionof the concept of confirmation and of its cor-
relatesseemsto have been developedso far;and the fewrudimentary definitions
implicit in recent methodologicalwritingsprove unsatisfactoryin various
respects (cf. section 3, end).
It is the objectiveof thisstudyto develop a generaldefinition of confirmation
in purely logical terms. Confirmationwill be construedas a certain logical
relationship whichmaybe consideredas an ideal modelor a "rationalreconstruc-
tion" of the conceptof confirmation as it is used in the methodologyof empirical
science. Apart fromits relevancefor the theoryof science,the definitionof'
confirmation hereto be developedappears to be of interestalso fromthe view-
point of formal logic; forit is of a purelysyntacticalcharacter,3 and it establishes
a relationanalogous in various ways to the relationof consequencein its syn-
tactical interpretation. In fact, confirmationwill prove to be, in a certain

Received September 2, 1943.

' A modified and expanded version of a paper of the same title of which an abstract
appeared in this JOURNAL, vol. 8 (1943), p. 39. The earlier version was scheduled to be
read at the meeting of the Association for Symbolic Logic that was to have taken place at
Yale University in December 1942.
2 A detailed discussion of the various methodological aspects of the problem of confirma-

tion will be given in a separate paper, to be published elsewhere, by the present author and
Dr. Paul Oppenheim. It is the study, with Mr. Oppenheim, of these broader issues which
suggested the problem of definingconfirmationin logical terms.
3 Cf. R. Carnap, The logical syntax of language, 1937, ?1.

122
 A PURELY SYNTACTICAL DEFINITION OF CONFIRMATION 123

sense, a generalizationof the converseof the syntacticalconsequencerelation

(cf. 3.3 below). The developmentof a logical theoryof confirmation might
thereforebe regarded as a contributionto a fieldof study sometimescalled
inductivelogic; a fieldin whichresearchso farhas almostexclusivelybeen con-
cernedwith probabilities(in the sense of the "logical" theories4)or degreesof
confirmation' ratherthan with the more elementarynon-quantitativerelation
of confirmation whichformsthe object of thisessay.
2. The model language L. Notationalconventions. The presentstudy is
restrictedto the case wherethe languageof sciencehas a comparativelysimple
logical structurewhich will now be characterized;it is essentiallythat of the
lowerfunctionalcalculuswithoutidentitysign. Whilemanydifferent languages
it
satisfythe followingstipulations, will be to
convenient formulate the latteras
referring to "the" language L.
Throughoutthis paper, quotation marksto be used forformingdesignations
of expressionswill be omittedwhenthe expressionsare listedin a separateline.
2.1 The followingkinds of signs are permittedin L:
2.11 Four statement connectives:
V . D

2.12 Individualconstants: As such we use the small letters'a', 'b', 'c, 'd', 'e'
with and without positive integersubscripts. It will prove convenient,for
furtherreference,to lay down a certainorderforthese constants,to be called
their alphabetical order, namely:
a, a, , a2, *' * ; b, bi, b2 *..X C)c, )1 C2,*ca d, di , d2,* - e, el , e2,* -.

2.13 Individualvariables:
xiXx1 x2, *... y,
Y7 l Y2) ... y Z. Zl) Z2,1

of any finitedegree;especially:
2.14 Predicateconstants
a) of degree1.
P. PI), P2 .**, Q. Ql, Q2,

b) of degree2,
R. Ri , R2, )

Also, additional predicate symbols-mostly abbreviationsof English words-

will be used when necessary;example: 'Sw' for 'Swan.'
signs,as in:
2.15 Universaland existentialquantifier
(x)(Ey)R(x, y).
2.16 Auxiliary signs: parentheses,comma.
2.2 The sentencesof L or L-sentencesare formedout of the above kinds of

4Such as those of J. M. Keynes, J. Nicoci, H. Jeffreys,B. 0. Koopman, St. Mazurkie-

wicz, and others.
5 Cf. particularly Janina Hosiasson-Lindenbaum, On confirmation,this JOURNAL, vol. 5
(1940), pp. 133-148.
124 CARL G. HEMPEL

signsaccordingto therulesofthelowerfunctional calculus. To indicategroup-

ing,parenthesesratherthan dots are used. Universalityis always expressed
by meansof a universalquantifier, neverby the use of freevariables. A class
of Lhsentences is also called a sentential class.
2.3 The rulesof inference for L are those of the lowerpredicatecalculus
withoutidentitysign. As neithersententialvariablesnor predicatevariables
are providedforin L, we have to assume that primitivesententialschemata
ratherthanprimitive sentenceshave beenlaid down,each schemacharacterizing
a certainclass of -rsentences as primitivesentences. Furthermore, a set of
rules of inferenceis assumedto be given. The particularway in whichthe
primitivesententialschemata and the rulesof inferenceare chosenis not of
importanceforthe subsequentdiscussion;we therefore refrainfromgivingan
explicitstatementof these stipulations.6
2.31 L willbe assumedto containno definedindividualor predicatesigns;
all thetermslistedin 2.12 and 2.14 willbe consideredas primitives. (In effect,
the followingconsiderationsare also applicable to languages providingfor
explicitdefinitions; the above restriction thenamountsto the assumptionthat
in sentences'studied as to confirmation, all definednon-logicalsignshave been
eliminatedby virtueof theirdefinitions.)
A sentenceS -will-besaid to be a consequence of a class K of sentencesif S
can be deduced,by means of a finitenumberof applicationsof the rules of
inference, fromthe class obtainedby addingto K the primitivesentencesof L.
A sentenceS2 is said to be a consequenceof a sentenceSi if S2 is a consequence
of { Si }, i.e., oftheclass containingSi as itsonlyelement.
We note the following theoremforlaterreference:
2.32 Theorem:If K is an infinite sententialclass,and S is a consequenceof
K, thenthereexistsa finitesubclassK' of K such thatS is a consequenceof K'.
This followsfromthe definition of "consequence"in connectionwiththe fact
thateveryruleof inference of the narrower predicatecalculuspresupposesonly
a finitenumberof premises-theentiredeductionof S fromK can therefore
containonly a finitenumberof sentences.
A class K (or a sentenceSi) willbe said to entaila sentenceS2 if S2 is a con-
sequenceof K (or of Si, respectively).
A sentenceS is called analyticif it is a consequenceof the null class of sen-
tences. S is called contradictory if its denial is analytic (in this case, every
sentenceis a consequenceof S). S is calledconsistent ifit is not contradictory.
A sententialclass K is called inconsistent if everysentenceis a consequence
of it; otherwise,it is called consistent.Two sententialclasses will be said to
contradict each other,or to be incompatible witheach other,if theirsum is in-
consistent;otherwisethey are called compatible.A sentenceSI will be said
to be incompatible with,or to contradicta sentenceS2 (or a sententialclass K)
if {Si) + {S2} (or SI I + K, respectively) is inconsistent; otherwise, Si will
be said to be compatible withS2 (or withK, respectively).

6 For detailed statementsof suitable sets of rules see, forexample, R. Carnap, Formali-

zation of logic,Cambridge,Mass., 1943,p. 135ff.;and especially D. Hilbert and P. Bernays,

Grundlagender Mathematik,Vol. I, Berlin, 1934, ?4.-Carnap's systeminvolves reference
to primitivesententialschemata.
 A PURELY SYNTACTICAL DEFINITION OF CONFIRMATION 125

of
is a sentenceconsisting
A sentenceof atomicform,or an atomicsentence,
by a parenthesized
followed
a predicate, whichconsistsof as many
expression
individualconstants-separatedby commas-as the degreeof the predicate
requires. Examples: 'P(a)', 'R(c, d)'.
A sentenceof basic form,or a basicsentence,is a sentencewhichis either
atomic,or thedenialof an atomicsentence.-Asentenceis said to be a basic
ifit is eithera basicsentence,
conjunction ofbasicsentences.
or a conjunction
or simplya molecule,
A sentenceof molecularform,or a molecularsentence,
is a sentencewhichis eitheratomicor consistsof atomicsentences and state-
mentconnectives.
A generalized is a sentencecontaining
sentence at leastone quantifier.
An expression obtainablefroma sentenceby replacing at leastoneindividual
constant bya variablenotoccurring inthatsentence willbe calleda (sentential)
matrix, a molecular
and in particular matrix ifit is obtainablefroma molecular
sentence.
A sentenceconsistingofa molecular matrixprecededexclusively by universal
or exclusively
quantifiers, by existential
quantifiers, or by at leastoneuniversal
and at leastone existential willbe calleda universal
quantifier or an
sentence,
respectively. A sentencewill be
existentialsentence,or-a--mixedgeneralization,
ifit containsno individualconstants.
generalized
calledcompletely
A full sentenceof a givenpredicateis an atomic sentencebeginningwith that
predicate. A full sentenceof a givensententialmatrixis one obtainable from
the matrixby substitutingsomeindividualconstantforeveryvariablewhich
is notboundby a quantifier, occurrences
different of thesamevariablebeing
replacedby the same constant.
We nowturnto a briefsummaryofthedevicesusedinthispaperto formulate
statementsaboutL; someofthemhavealreadybeenemployed in thepreceding
text.
As meta-languageforL, we use English,enrichedby certainabbreviating
we use:
symbolsand by variablesofcertainkinds. In particular
Variablesfor names of sentencesof L:
formolecules),
M (reserved
StS Its D2t *^ TT T2,
I .

Variablesfor names of sententialclasses:

K, K1, K2,y , K*, K**,
Unlessotherwisestated,thedesignationof an expression in L is formed by
in quotes.
includingthe expression
classofsentences,
ofa finite
To forma designation weshallplacetheirnames,
separatedby commas, betweenbraces. The samedevicewillbe used in con-
nectionwithsententialvariables. Thus,'{Si, S2, S.}' will be shortfor'the
class whose elementsare S1, S2, SS-)
'rS5' of S, and S2,
willdesignatethe denialof Si, 'S1.S2'the conjunction
and 'SevS2','S1DS2' willbe used analogously.The same stipulation applies
126 CARL G. HEMPEL

to all other sentence name variables and will also be used in connection with
names of particular sentences when no misunderstandingseems possible.
The arrow symbol, I', will be used as an abbreviation of "has as a conse-
quence" or "entails," as in 'Si -+ (SYvS2)', 'IS1, S2} -* Si'. Analogously,
the double arrow, '', will be used to designate equivalence (mutual con-
sequence).
Occasionally the statement connective symbols ', 'v', '., 'd', '-' will be
used also in the meta-language; especially in the formulationof theorems,
or of definitionssuch as 3.01, 3.02 below.
Some furthernotational devices will be explained upon their firstoccurrence.
Finally, we have to characterizethe intendedinterpretationof the expressions
occurringin L. While the solution of our problem will proceed along purely
syntactical lines, semantical and pragmatic referencesare indispensable for an
appraisal of the significanceof the problem, and for an evaluation of the ade-
quacy of the proposed solution.
The individual constants of L are thought of as names of particular things
which formthe object of some scientificinquiry. The predicatesare to designate
certain attributes,i.e., propertiesor relations,of those things; and we conceive of
them as having been chosen in such a way that the observationalor experimental
methodsused in the inquirymake it possible to ascertain the absence or presence
of those properties or relations in a given individual or group of individuals.
In other words, we assume that any test report, i.e., any complex of "data"
obtainable by the testing procedures in question, can be formulatedby means
of full sentences of predicates of L. To state this idea in a precise manner,one
mightstipulate that each singledatum obtainable in those tests is expressiblein a
basic sentence,and that everytest reporthas to have the formof a basic conjunc-
tion; or one may, as we shall do here,adopt the moreliberalstipulationthat every
possible test reportis of molecular form. Now, the scientifictest of any hypo-
thesis is based on a confrontationof the hypothesis with empirical data, i.e.,
with a test report. We shall thereforeaim at definingconfirmationas a relation
which obtains, under certain conditions,between a molecular sentence M (rep-
resenting,the observation report) and a sentence S which may be of any form
(representingthe hypothesis). This correspondsto the idea that the crude data
obtained by the testing procedure are always particular in character; but that
they may serve to corroborateor weaken hypotheses of general as well as of
particular form.
3. Restatement of the problem. Criteria of adequacy for its solution. In
view of the preceding considerations,our problem assumes a purely syntactical
character. In its originalformit presentedalso a semantical and a pragmatical
aspect; for it amounted to this question: Under what conditions shall we say
that a report on the outcome of a certain test confirmsa given hypothesis?
Thus, refernc.,aeto actual testing procedures was involved. But clearly, it is
immaterial for the definitionof a logical relation of confirmationwhether the
given report represents actual or hypothetical data. In fact, our definition
will be satisfactoryonly if it enables us to discuss questions of the type: "Sup-
posing that such and such data were obtained, would they confirmsuch and
 A PURELY SYNTACTICAL DEFINITION OF CONFIRMATION 127

such a hypothesis?";"What empiricalfindingswould constitutedisconfirming

evidence for this hypothesis?";etc. Our problem,then,will be to lay down
purelyformalconditionsunderwhicha moleculeM confirms a sentenceS.
Any proposed definitionof a relationCf of confirmation will automatically
determinedefinitionsof the two associated relationsof disconfirmation and
of neutrality,accordingto the followingschemata:
3.01 Dscf(M, S) _ Cf(M, US).
3.02 Ntl(M, S) _ ('-,'Cf(M, S) . '-,Dscf(M, S)).
While the indicationsgiven so far in this essay as to the intendedmeaning
of"confirmation" do not determinethisconceptunambiguously and precisely,it
is clearthat any acceptabledefinition
of confirmation willhave to satisfycertain
criteria of adequacy. These may be divided into two classes, logical and
materialconditionsof adequacy, whichwill now be discussedin turn.
Logical conditionsof adequacy. The logical requirementsare stated in three
groups. In each group,the fulfillment of the firstconditionentails that of all
others. Those other conditionsare mentionedfor two reasons; first,because
most of them representimportantcharacteristicswhich would generallybe
soughtin an adequately definedconceptof confirmation; and secondly,because
some apparentlyreasonable alternativedefinitionswhich we shall examine,
turn out to satisfysome of those weaker conditions,but not the strongestof
each group. Confrontationwith the different requirementsexplicitlystated
will thus provide a yardstickfor the appraisal of what mightbe termedthe
degree of adequacy of a proposed definitionof confirmation.
First group of logical requirements.
3.1 Generalconsequencecondition: If M confirmseverysentenceof a class
K, then it also confirmsevery consequenceof K.
3.11 Special consequence condition: If M confirmsS, then it also confirms
everyconsequenceof S.
3.12 Equivalencecondition: If M confirmsS, then it also confirmsevery
equivalent sentence. (Intuitively,this means that whethera sentenceis con-
firmedby a givenmoleculeor not,dependsonlyon its content,not on its formu-
lation.)
3.13 Conjunctioncondition: If Il confirmseveryone of a finitenumberof
sentences,then it confirmsalso theirconjunction.
Second groupof logicalrequirements.
3.2 Generalconsistency condition: Every consistentmoleculeis compatible
withthe class of all the sentenceswhichit confirms.
3.21 Special consistency condition: The class of all the sentencesconfirmed
by a consistentmoleculeis consistent.
3.211 No consistentmoleculeconfirmsand disconfirms the same sentence.
3.22 No consistentmolecule confirmsa contradictorysentence.
Third group of logical requirements.
3.3 Entailmentcondition: If M -> S, then M confirmsS. (This means
that the relationof entailmentin L, with its domain restrictedto molecular
 128 CARL G. HEMPEL

sentences,is containedin the relationof confirmation, and that, in this sense,

the conceptof confirmation is a generalizationof that of entailment.)
3.31 Any molecule confirmsany analytic sentence.
3.32 A contradictory moleculeconfirms everysentence.
It mightbe thoughtplausiblethat3.3 shouldbe supplementedby stipulating:
If S -* M, then M confirmsS. Indeed, if this conditionis satisfied-as, for
instance,in the case whereM = 'P(a) . P(b) . P(c)' and S = '(x)P(x)'-then,
it mightbe argued,all the information conveyedby M bears out the hypothesis
S. However, the stipulationunder considerationis far too comprehensive:
togetherwith 3.11, it would make any moleculeconfirmeverysentencewhat-
oever. For, let S be an arbitrarysentence,then,by the contemplatedstipula-
.:on, M would confirmM.S, and consequently,by 3.11, also S.
Let us note that as a consequenceof 3.01, 3.02, and 3.211, everyconsistent
moleculeM determinesa divisionof all sentencesof L into threemutuallyex-
clusive classes: those whichM confirms, those whichit disconfirms, and those
withrespectto whichit is neutral.-If M is inconsistent, then it entails every
sentenceand thus,because of3.3, both confirms and disconfirms everysentence.
The materialconditionof adequacyamounts,briefly,to the requirementthat
the proposeddefinition of confirmation shouldbe in sufficiently close agreement
withthe customarymeaningof the conceptof confirmation as it is used in the
methodologyof empiricalscience. Naturally,this requirementis vague, and
it refersto certainstandardswhichcontaina fairlystrongsubjectiveelement:
what appears as a materiallyquite satisfactorydefinitionof confirmation to
one person,may be judged artificialand too comprehensiveor too narrowby
another; also, opinionas to the adequacy of a proposeddefinitionmay vary
withone and the same personin the courseof his occupationwiththe problem.
Consider,forexample,the hypothesis'All swans are white,'whichwe sym-
bolize by '(x)(Sw(x) D Wh(x))', and the molecules'Sw(a). Wh(a)' and '-%'Sw(b)
--.Wh(b)'. It mightbe felt-and would probablybe argued by many-that
undera materiallyadequate definition of confirmation, the firstmoleculeshould
be confirming, the second neutralwith respectto the hypothesis. However,
accordingto the same intuitivestandards,the secondmoleculewould constitute
confirming evidencewith respectto the hypothesis'(x)('%.'Wh(x) D Sw(x)).
But the two hypotheseshere mentionedare equivalent,and adherenceto the
intuitivestandards exemplifiedin this illustrationwould thereforemake the
confirmation of a hypothesisby a certainbody of evidencea matternot onlyof
the contentof the hypothesis,but also of the formin which it is expressed.
This featureappears intuitivelyintolerableand was, in fact,explicitlyruledout
by the equivalence condition3.12. As this case shows, some aspects of the
intuitiveidea of confirmation-eachof which separately may appear quite
reasonable-are incompatiblewith each other. It is thereforeto be expected
that certain apparentlyvery natural intuitiverequirementswill have to be
sacrificedin the interestof others which appear to have greatersystematic
importance.
In additionto beingvague in the sensejust illustrated,the materialcondition
of adequacy is also indefinitein the specificsense that the materialadequacy
 A PURELY SYNTACTICAL DEFINITION OF CONFIRMATION 129

of any proposedcriterionof confirmation can be ascertainedonly inductively,

as it were, namelyby applyingthe criterionto certainconcretecases which,
on intuitivegrounds,appear as unquestionableinstances of confirmation(or
of disconfirmation or neutrality,as the case may be), and by checkingwhether
the formalcriterionin questionyields the desiredresult. But clearly,even if
a givendefinition ofconfirmation passes any numberof testsofthistype,we can
never be quite certain that it does not possess some featureswhich,if only
recognized,wouldstampit materiallyinadequateor at least awkward. Despite
its indefiniteness,the questionof materialadequacy cannotbe disregarded,and
it will be taken into considerationin every step of the subsequent develop-
ment of a definitionof confirmation.
Certainconceptionsof confirmation whichseemto be reflected in recentmeth-
odological discussionswould have to be ruled out on the basis of the above
requirementsof adequacy.
Thus, one customarycriterionof confirmation, if stated explicitly,would
lead to the followingdefinition:A confirming instancefora hypothesisof uni-
versal conditionalform-say, '(x)(y)(R1(x, y) D R2(x, y))'-consists in the
conjunctionof two fullsentencesobtainedby replacingeach individualvariable
in the molecularmatrixwhich constitutesthe antecedentby some individual
constant, and by performingthe same substitutionin the consequent.7 In
theabove illustration, 'R1(a, b) . R2(a, b)' wouldbe a confirming instance.-How-
ever, this criterionis open to various objections: (a) It violates the formal
requirements3.1, 3.11, 3.12.8 This is readilyseen fromthe earlierdiscussion
referring to the hypothesis'(x)(Sw(x) D Wh(x))'. (b) An interestingaspect
of the violationof 3.12 is the following:Certainuniversalconditionals,such as
'(x) ((P(x) . -'Q(x)) D (Q(x) . -'Q(x)))', wherethe consequentis contradictory
(but the antecedentnot analytic),clearlycannothave any confirming instances
under the criterionin question; and yet these universalconditionalsare not
contradictory;the one just mentioned,for example, is equivalent with the
hypothesis'(x)(P(x) D Q(x))'. (c) The criterionunder considerationis re-
stricted to hypothesesof universal conditional form,whereas an adequate
definitionof confirmation should be applicable to hypothesesof other forms
as well.
An alternativeapproach mightbe suggestedby the considerationthat the
test of a scientifichypothesisconsistsin examiningits predictivepower,i.e.,
in deducingfromit and certainobservationsentencescertainnew observation
sentences,and in seeingwhetherthese predictionscome true. This idea could
be reflectedin a syntacticaldefinition of the followingtype: A class K of mole-
cules (which,incidentally,is equivalent to one moleculeM) confirmsa given
sentenceS if K can be exhaustivelydividedinto two mutuallyexclusiveclasses
K1, K2 such that K2 is non-null,and every sentencein K2 is a consequenceof

7 A criterion to essentially this effectwas formulated by Jean Nicod in Foundations of

geometry and induction, London 1930, p. 219.
8 With respect to the equivalence condition, this difficultywas exhibited already in the

author's Le problemede la vkritt,Theoria (Gkteborg), vol. 3 (1937), pp. 206-246, esp. p. 222.
130 CARL G. HEMPEL

K1 + {S but not of K1 alone.-But this type of definitionclearlyviolates the

requirements 3.1 and 3.11; also, it is materiallyinadequate in that,forinstance,
themolecule'P(a)' wouldnot by thiscriterionconfirm the hypothesis'(Ex)P(x)'
(this featurealso violates 3.3); and more generally,it would notprovide any
possibilityof confirmation at all forexistentialsentencesbecause these cannot
satisfy the above condition.-Finally, the consistencycondition would be
violated; thus, e.g., the class {'Pi(a)', 'P2(a)'} would confirm (x)(P1(x) D
(P2(x) . P3(x)))' as well as, say '(x)(Pi(x) D (P2(x) . P3(x)))'.

4. The C-development of a sentence. The basic idea of the definition of

confirmation hereto be developedmay be illustratedby the followingexample
If S and M are the sentences'(x)(P (x) D Q(x))' and 'P(a) . Q(a) . P(b)
Q(b) . Q(c) . P(c)', respectively,then M can be said to confirmS in the
followingsense: S assertsthat the extensionof P is includedin the extension
ofQ (and, consequently,the extensionof -Q in thatof U-%.P); and forthe objects
referred to in M, namely,a, b, and c, it is indeedthe case that all thosereported
in M as belongingto the extensionof P are also reportedas belongingto the
extensionof Q, and those reportedas belongingto the extensionof -Q are
also reportedas belongingto the extensionof -P.9
The problem now presentsitselfof developingan explicit criterionwhich
embodiesthis idea, and whichis applicable also to hypothesesthat cannot be
interpretedas assertinga relation of inclusion,as is the case, in particular,
with existentialsentencesand mixed generalizations. We thereforestate the
inclusioncriterionin a modifiedversionwhichlends itselfmore readilyto the
necessaryexpansion.
The above molecule M determinesa certain class of individuals {a, b, c};
and we may say thatM confirms S because fromtheinformation containedin M
it can be inferredthatin {a, b,c} S is completelysatisfied;or,to use a metaphor.
in a world containingexclusivelythe individualsa, b, and c, the sentence S
wouldbe true,accordingto the information containedin M. Now, fora world
of that finitekind, the contentof S could be expressedin termsof a certain
molecularsentence,namely'(P(a) D Q(a)) . (P(b) D Q(b)) . (P(c) D Q(c))';
and the assertionthat S would be truein that world,accordingto the informa-
tion containedin M, can be expressedmorepreciselyby sayingthat the above
molecularsentenceis a consequenceof M. We shall call that molecularsen-
tence the IM-development of S (i.e., the development of S for the individual
constantscontainedin Ml). The idea of IM-developmentcan readilybe de-
finedin a generalmanner,so as to be applicable to sentencesof any form;thus,
forexample,the developmentof the sentence'(x)(Ey)R(x, y)' forthe domain
determined b) . R(c, b) . R(b, a)' is thefollowing:
by themolecule'R(a, a) . -%.'R(a,

9 This idea of definingconfirmationin terms of an "inclusion criterion" was suggested

to me by Dr. Nelson Goodman as offeringconsiderable advantages over an attempt which I
had made before to defineconfirmationin terms of "inductive attainability" of S fromB-
an approach which is not discussed in the present paper. Dr. Goodman's idea proved an
invaluable help for the present investigation; in fact, it initiated all the following con-
siderations.
 A PURELY SYNTACTICAL DEFINITION OF CONFIRMATION 131

'(R(a, a) v R(a, b) v R(a, c)) . (R(b, a) v R(b, b) v R(b, c)) . (R(c, a) v R(c, b) v
R(c, c))'. In termsof this concept,whose generaldefinitionwill be given
presently, thecentralidea ofthedefinition of confirmation hereto be developed
assumesthis form: A sentenceS is confirmed by a moleculeM if M entails
theIM-developmentof S. As willbe seenbelow,thisidea stillrequirescertain
modifications if it is to yieldan acceptableconceptof confirmation.
In the above illustrations of the conceptof the development of S we referred
to a domain of objects, and to theirsatisfyingcertainconditions. Thus we
leftthe sphereof purelysyntacticalanalysisand made use of semanticalrela-
tions. However,it provespossibleto expressthe intendedidea in exclusively
syntacticalterms. This is done in the followingdefinition:
4.1 Df. Let C be a finiteclass of individualconstants(not individuals)
and S a sentence. Then the C-development of S is a sentenceDC(S), whichis
determined by the following recursivedefinition:
A) If C is empty,thenDC(S) = S.
B) If C is not empty,then:
I a. Dc(,-S) = c(S).
b. Dc(S1 v S2) = Dc(S1) v DC(S2).
c. Dc(S1. S2) = Dc(S1). DC(S2).
d. Dc(S1 D S2) = Dc(S1) D DC(S2).
II a. If S is atomic,thenDC(S) = S.
b. If S is oftheform()4) wheret is an individualvariableand4 a sentence
or a matrix,then:
bl. If 4)containsno freeoccurrence of(, thenDc(S) = Dc(D).
b2. If c1containsfreeoccurrences of(, thenlet

Yee
TfEC

be the conjunctionof the following sentences,in the orderheredescribed:

The sentenceobtainedfrom4) by replacingall freeoccurrences of t by the
alphabeticallyfirst(cf.2.12) individualconstantin C; thesentenceobtained
analogouslyby means of the alphabeticallysecond constantin C; and so
on. Then
Dc(S) = Dc(H CY).
7EC

c. If S is of the form(Et)4, then:

cl. If 4 containsno freeoccurrenceof (, thenDc(S) = Dc(D).
c2. If 1 containsfreeoccurrencesof (, let

be the disjunctionof the sentencesdescribedunder2.2, takenin the same

order. Then
Dc(S) = Dc(Z b).
7fC
132 CARL G. HEMPEL

Notational conventions: (a) When names of particular sentences occur

as argumentsof 'Dc', the name-forming quotes will be absorbedby the frame
'Dc( )'. (b) Also, in sentencesof the formDc(...) = - - -, where on the
righthand side a quoted sententialname would have to appear, we shall have
the quotes absorbedby the identitysign.
Thus, e.g., if C = {'a', 'b'), we shall write
Dc((x)P(x)) = P(a) P(b)
instead of
Dc('(x)P(x)') = 'P(a) . P(b)'.
The left-handside is simplifiedaccordingto part (a), the right-handside ac-
cordingto part (b) of the notationalconventions.
Illustrationsof the above definition.
4.11 Let S = '(x)(Ey)R(x, y)'; C = ('a', 'b', 'c'}; then:
Dc(S) = Dc((Ey)R(a, y). (Ey)R(b, y) . (Ey)R(c, y)) (by B II b2)
= Dc((Ey)R(a, y)). Dc((Ey)R(b, y)) . Dc((Ey)R(c, y)) (by B I c)
= Dc(R(a, a) v R(a, b) v R(a, c)) . Dc(R(b, a) v R(b, b) v R(b, c))

. Dc(R(c, a) v R(c, b) v R(c, c)) (by B II c2)

= (R(a, a) v R(a, b) v R(a, c)) . (R(b, a) v R(b, b) v R(b, c))
. (R(c,a)vR(c,b)vR(c,c)) (byBIbandBII a).
This,incidentally,is thesentencewhichwas givenabove as theIM-development
of S forM = 'R(a, a) . --'R(a, b) . R(c, b) . R(b, a)'.
4.12 Let C = {'a', 'b'}, S = '(x)(P(x) v Q(c))'. Then, as is readilyverified:
Dc(S) = (P(a) v Q(c) . (P(b) v Q(c)).
4.13 For everymolecularsentenceM and everyfiniteclass C of individual
constants,Dc(M) = M.
Subsequently,the expressions'c.g. sentence' and 'c.g. class' will serve as
abbreviationsof 'completelygeneralized sentence' and 'class of completely
generalizedsentences,'respectively.
We note the followingtheoremson C-development:
4.2 Theorem. If C is a finiteclass of individualconstantsand Si, S2 are c.g.
sentencessuch that Si -+ S2,thenDc(Si) -+ Dc(S2).
Proof. If Si -+ S2, then Si DS2 is analytic,'0and hence identicallytrue for
any finitedomain;11i.e., Dc(S1DS2) is an analyticsentenceforeveryfiniteclass C
of individual constants. Now Dc(S1DS2) = Dc(S1) D Dc(S2) (4.1 B I d);
therefore,thelattersentenceis analytic,and by virtueofthisfactand the modus
ponensrule,Dc(S1) -+ Dc(S2).

10 Cf. Hilbert and Bernays, loc. cit., p.

155, or Carnap, loc. cit. (see footnote 6), p. 142,
T 28-11.
11Cf. Hilbert and Bernays, loc. cit., p. 121, Theorem 1.
 A PURELY SYNTACTICAL DEFINITION OF CONFIRMATION 133

4.21 Theorem. If S,, S2 are equivalent c.g. sentences,then Dc(S1), D0(S2)

are equivalentforeveryfiniteclass C of individualconstants. (From 4.2.)
4.3 Theorem. Let C be a finiteclass of individual constants,K a finite
class of c.g. sentences,Dc(K) the class of the C-developmentsof the elements
of K, and S a c.g. sentencesuch that K -> S; thenDc(K) -> Dc(S).
Proof. Let
IIT
TeK

be the conjunctionof the elementsof K takenin any one order,then

II
TeK
T-S;

hence,. by 4.2,

Dc(][I T)
TeK
Dc(S);

furthermore,
by 4.1,
Dc(HJ T)
TeK

is a conjunctionof the C-developmentsof the elementsof K, and therefore

followsfromDc(K); thus we have
DC(K) -> Dc(lI T) -Dc(S)
TeK

which proves the theorem.

4.4 Theorem. If a class K of c.g. sentencesis inconsistent,then,forevery
finiteC, the class oftheC-developments ofthesentencesofK is also inconsistent.
Proof. K mustcontainat least one sentenceto be inconsistent. Let S e K.
Then, since K is inconsistent,K --+ -S. Hence, by 4.3, Dc(K) -* Dc(-S), or,
by virtue of 4.1, Dc(K) -+ -Dc(S). This shows that Dc(K) is inconsistent,
since also Dc(S) e Dc(K).
4.41 Theorem. For every finiteC, a contradictoryc.g. sentence has a
contradictoryC-development. (From 4.4)
4.42 Theorem. For everyfiniteC, an analyticc.g. sentencehas an analytic
C-development.
Proof. If S is an analytic c.g. sentence,then -S is a contradictoryc.g.
sentence;now,by 4.1, Dc(S) is equivalentwith -Dc(-S); and since,by 4.41,
Dc(-S) is contradictory, Dc(S) is analytic.
4.51 Note. The converseof theorem4.4 and its corollariesdoes not hold:
it may happenthat fora certainclass C ofindividualconstantsa c.g. sentenceS
has an analytic or a contradictoryC-developmentwithoutbeing analytic or
contradictory itself. For example,let S1 = '(x)P(x) v (x)--P(x)', and S2 =
(Ex)P(x). (Ex)-P(x)'. Then,ifC containsexactlyone element,Dc(Sl) will be
analytic. Thus, if C = {'a'), Dc(S1) is 'P(a) v -P(a)'. On the otherhand,
S2 can be true only in a domain of at least two individuals;fora class C con-
tainingonlyone elementits C-developmentis contradictory; thus,if C = {'a'},
then Dc(S2) = P(a) . P(a).
134 CARL G. HEMPEL

By developingthese illustrationsa little further,it can also readilybe seen

thatfora givenC and twocompletelygeneralizedsentencesSi, S2, it may happen
that Dc(S1) -> DC(S2), whileit is not the case that Si -> S2. This shows that
the converseof 4.2 does not hold.
4.52 Note. None of the theorems4.2 through4.42 holds forall generalized
sentences,i.e., includingthose which contain individual constants. For let
Si = '(x)P(x)', S2 = '(x)(P(x) . P(a))', C- { 'b'}; then Si <-+ S2; but Dc(S1) =
'P(b)'; DC(S2) = 'P(b) P(a)',
. and thus neither Dc(S1) <-+ DC(S2) nor even
Dc(S1) -? DC(S2). This providescounter-examples for4.2-and thus for4.3-
as well as for4.21; to obtain a counter-examplefor4.41-and thus for4.4-
let S = '(x)P(x).-P(a)', C = {'b'}, so that Dc(S) = 'P(b) . -P(a)', whichis
non-contradictory; finally,as counter-examplefor 4.42, choose S = 'P(a) D
(Ex)P(x)', C = {'b'}, whichyields the non-analyticDc(S) = 'P(a)DP(b)'.
5. Preliminaryremarks on the subsequent definitionsof confirmation.
We now turn to the systematicconstructionof a syntacticalconcept of con-
firmation. We shall beginby formulating a firstdefinitionin strictaccordance
withthe idea outlinedin the beginningof section4: M willbe said to confirmS
if M entailsthe IM-developmentof S. Closerexaminationof the conceptthus
determinedwill reveal certain inadequacies which will then be removed by
constructinga second, revised definition;the latter,in turn,will be replaced
by a modifiedand moresatisfactoryfinalversion. Lest the readerbe alarmed
by the propsectof being needlesslyled astray by a study of certaintentative
definitionswhichwill laterbe abandoned,it may be well to emphasizethat it is
not intendedto presenthere all the various attemptsat definingconfirmation
whichweremade in connectionwiththis study. The fewvariantsthat will be
consideredherehave been selectedforsystematicreasons,namelybecause they
represent,as it were,successive approximationsof the definitionfinallyto be
proposed;in fact,everydefinition subsequentlyto be consideredpresupposesthe
precedingones or certaintheoremsprovedin connectionwiththem.
For reasons whichwill be exhibitedin the followingsection,the relationof
confirmation, Cf, will, in each of the successive stages of our discussion,be
definedin termsof a narrowerrelationof directconfirmation, Cfd. The manner
in whichCf is definedin termsof Cfd will remainthe same throughout,and the
gradualmodifications referred to will concernthe definitionof Cfd.
6. A firstapproximation:Cfdl and Cfj. The followingdefinitionof direct
confirmation embodiesthe idea presentedin thebeginningofsection4:
6.1 Df. Cfd1(M,S) if and only if: (a) M is a molecule; (b) S is a c.g. sen-
tence; and (c) M -> DIM(S), whereIM is the class of thoseindividualconstants
whichoccurin IMl.
The followingtheoremshold for Cfdi:
6.11 Theorem. Within the class of all c.g. sentences,Cfdi satisfies the
general consequencecondition3.1; i.e., if K is a c.g. class and M a molecule
such that Cfd1(M, T) for every T e K, and if S is a c.g. sentencesuch that
K -* S, then Cfd1(M, S).
Proof. In view of 2.32, K may be assumed to be finite. Let DIM(K) be the
 A PURELY SYNTACTICAL DEFINITION OF CONFIRMATION 135

class of the IM-developmentsof the elementsof K. Then, since K -f S, we

have DIM(K) -* DIM(S), by 4.3; furthermore,by hypothesis,MIll * T for every
T e DIM(K); hence: M -> DIM(S), and thus Cfd1(M, S).
6.12 Theorem. Cfdl satisfiesthe general consistencycondition3.2; i.e.,
if M is a consistentmoleculeand K* the class of all S such that Cfd1(M,S).
thenK*+ {M} is a consistentclass.
Proof. SupposethatK*+ {M} is inconsistent.Then thereexistsa sentence
T such that K*+ {AMI- T, K*+ {M' T. In view of 2.32, thereexists
-

even a finitesubclassK of K* such that K+ {M} -* T, K+ {M} - T. Let

SK be the conjunction of the elementsof K. Then SK. 111wouldbe a contra-
dictorysentence. This will now be shownto be impossible. The coreof this
proofis the followingconsideration:If SK . M is a contradictory sentence,then
it cannotbe satisfiedin any domain. But in the domainconsisting exclusively
oftheindividualsmentioned in M, clearlyM is satisfiableif,as was presupposed,
it is consistent. And if M is satisfied,thenso is SK; forin the finitedomainin
question,SK is equivalentwithDIM(SK), and thelattersentenceis, by hypothe-
sis,a consequenceofM, and thusis satisfiedwheneverM is.
This idea can be expressed more precisely as follows: Let IM = 'al', 'a2',
- ak'1;and let 'm', 'Sk', 'dsjc'be abbreviations
ofthesentencesM, SK, DIM(SK)
respectively. (Note. 'in' is an abbreviation
for the sentencedesignatedby
'M ', etc. Theseabbreviationscannotoccurin L, as no definitions
are allowedin
that language;but we mayintroducethemintothe meta-language.) The idea
that in a world containingonly the individualsa,, *.. , ak, the sentenceSK
holdsif DlJ(SK) does, can nowbe expressedby the statementthatthesentence
(x)((x=a1) v (x=a2) v ... v (x=ak)) D (dsk D Sk)

is analytic(not in L, whichdoes not containtheidentitysign,but in the meta-

language,whose logic has, of course,to includethe lowerpredicatecalculus
withidentitysign). Now,sinceby hypothesis M standsin Cfdlto everyS in K,
we have M -+ DIM(SK); thus,'m D dsk'is analytic;hence,finally,
((x) ((x = a,) v (x = a2) v v (x=Uak)) .n) D (Sk * rC)

is analytic. Now,ifSK. Al,and thus'skA.m' werecontradictory,

then
(x)((x=a1) v (x=a2) v ... v (x=ak)) . m
would be contradictory, and hencecould not be satisfiedin any finitedomain;
i.e., taking 'a ''a, , 'ak', and all the extralogical predicates occurringin M
as uninterpreted parameters,therewouldnotexista finitedomainin whichthose
individualand predicateconstantscould be so interpreted as to make the last
formulaa truesentence. But actually,thesentenceis satisfiable in a domainof
k individuals. We only have to construe'al', 'a2', , 'a' as namesof any k
individuals,and to interpretthe predicatesoccurringin M in such a way that
M becomes a true sentenceunderthat interpretation.And that is possible
because by hypothesisM is consistent.

12 Cf. Hilbert and Bernays, loc. cit., pp. 185-186.

136 CARL G. HEMPEL

6.13 Cfd1fails to satisfythe entailmentcondition;thus, e.g., Cfd1(M, M)

cannothold forany moleculebecause of the clause 6.1(b).
For the same reason, Cfd1violates the generalconsequencecondition;thus,
'P(a)' standsin Cfd1to '(x)P(x)', but not to 'P(a)'.
These shortcomings,however, can be remedied by defining,in terms of
'Cfd1',a broaderconceptof confirmation, 'Cf1',as follows:
6.2 Df. Cf1(M, S) if and only if (a) M is a molecule,and (b) S is a con-
sequence of a sententialclass K each element T of which satisfiesone of the
followingconditions:
1. M-+ T; 2. Cfd,(M, T).
Clause (b) can be symbolizedas follows:
(EK)((K - S) . (T)(T e K D ((M.- T) v Cfd1(M,T))))
6.21 Theorem. Cf1satisfiesthe generalconsequencecondition.
6.22 Theorem. Cf1satisfiesthe generalconsistencycondition.
6.23 Theorem. Cf1satisfiesthe entailmentcondition.
6.24 Theorem. Cfdj is a propersubrelationof Cfj.
Proofsof these theoremsfollow.
Proofof 6.24. (a) Cfd1is a subrelationof Cf1. Let Cfd1(M,S); thenthere
exists a K whichsatisfies6.2 (b), namely {S}; hence Cf1(M, S). (b) Cf1is
not a subrelationof Cfd1. Thus, e.g., 'P(a)' stands in Cf1 (via '(x)P(x)') but
not in Cfd1to 'P(b)'.
Proofof 6.23. If M -- S, thenthereexistsa K whichsatisfies6.2 b, namely
{M}.
Proofof 6.21. Let M be a moleculeand K1 a class such that Cf,(M, T) for
every T e K1; let K1 -- S. We have to prove that Cf1(M, S). By hypothesis
and 6.2(b), every T e K1 is a consequenceof a class KT such that M entails or
stands in Cfd1to everyelementof KT. Let , Kr be the sum of these classes.
Then, clearly,S is a consequenceof E KT, i.e., of a class such that M entailsor
stands in Cfdi to each of its elements. Hence, by 6.2, Cf1(M, S).
Proofof 6.22. Let M be a consistentmoleculeand K** the class of all sen-
tencesto whichM standsin Cf1. AssumeK**+ {M) to be inconsistent. Then
K** would be inconsistent,for M e K** by virtue of 6.23. Now, for every
T e IK**,we have by hypothesisCf1(M, T); i.e., T is a consequence of a sen-
tentialclass KT such that M eitherentails or stands in Cfd1to each of its ele-
ments. If, therefore, K* is the class of all sentencesto whichM stands in Cfd1,
theneverysentencein K** is certainlya consequenceof {M} +K*; and if K**
were inconsistent,then {M}I+K* would have to be inconsistent. But that is
impossiblein view of 6.12.
The concept of confirmation determinedby 6.2 thus satisfiesall the formal
conditionsof adequacy set up above. As regardsits materialadequacy, how-
ever, it mightbe argued that while the basic idea of the definitionappears
satisfactory,its formalization in 6.2 involvesan unnecessaryrestriction. Thus,
if a test reportcontainsa certainamountof definitelyfavorableevidencefora
hypothesis,and in additionsome entirelyirrelevantstatements,thenthe report
 A PURELY SYNTACTICAL DEFINITION OF CONFIRMATION 137

mightwell be consideredas confirming the hypothesis. Not generallyso,

however,accordingto theabove definitions of'Cfd1'and 'Cfl'. Let,forexample,
S = '(x)P(x)'; M1 = 'P(a) . P(b)', M2 = 'P(a) . P(b) Q(a)', M3 = 'P(a)
P(b) . Q(c)'. Then DIM1(S) = DIM2(S) = 'P(a) . P(b)', and thusCfd1(Mi,S)
and Cfd1(M2,S); and, by 6.24, Cf1(M1,S) and Cf1(M2,S); but DIM3(S) =
'P(a) . P(b) . P(c)', and thusM3 does notstandin Cfd1to S; nordoes it standin
Cf1to S either. (This latterstatementcan be provedby meansof the method
used in theproofoftheorem6.3 below;we omitthedetailshere.)
One mightfeelinclinedto changethissituationby defining 'M confirmsS' by:
'M has a consequenceof molecularformwhichstands in Cf1to S'. By this
standard,M3 wouldconfirm S. However,thenewcriterion is muchtoo liberal:
Accordingto it, themolecule'P(a) . --P(b)' would,by virtueofits consequence
'P(a)', confirm'(x)P(x)'; and, by virtueof its consequence"--'P(b)', it would
confirm '(x)--P(x)'; thus,the consistency requirement wouldbe violated. But
uponsomewhatcloserinspectiontheintuitivedifficulty whichthe contemplated
modification was designedto overcomeappears anyhowto be of minorsig-
nificance;the concept definedin 6.2 proves to be somewhatnarrowerthan
intuitiveusage would require;but clearlyit has to be expectedthat a precise
redefinition of a customarily vague conceptwill to some extentbe at variance
with the intuitivemeaningof the original. Besides, the conceptintroduced
in 6.2 providessufficient meansforstatingin whatsenseM3 constitutes, "on the
whole,"as it were,favorableevidenceforS; namelythus: M3 has consequences
of molecularformwhichstandin Cf1to S1 but nonewhichstandin Dscf1to S
(i.e., in Cf1to --S; cf.3.01).
But thereis anotherconsequenceof the definitionof 'Cf,' whichrequires
considerationhere: The conditionswhichM has to satisfyto confirma gen-
eralized,but not completelygeneralizedhypothesisappear to be too rigorous.
Let,forexample, S = '(y)R(a, y)', M1 = 'R(a,a) . R(a,b)'. Nowitmaybeargued
that by the same tokenby which'P(a) . P(b)' confirms '(x)P(x)', M1 shouldbe
designatedas confirming S; analogously,M2 = 'R(a, a) . R(a, b) . R(a, c)', etc.,
should representconfirming evidence for S. But, while 'P(a) . P(b)' does
stand Cf1 '(x)P(x)', neitherM1 nor M2 nor any of the analogouslonger
in to
moleculesstandsin Cf1to S. This will now be provedforM1; the proofcan
readilybe extendedto theothercases.
6.3 Theorem. M1 = 'R(a, a) . R(a, b)' does not stand in Cf1 to
S = '(y)R(a, y)'.
Proof. Note that clause (b) in 6.2 is equivalentwith
(EK)((T) (T e K D Cfd1(M,T)) . (K+ {M} S
S)).

Hence,ifthetheoremis false,thenthereexistsa classK-in viewof2.32it maybe

assumed finite-suchthat K+ {M1} - S and Cfd1(M1,T) forevery T e K.
Let SK be theconjunctionoftheelementsofK and let 'Sk' be an abbreviationof
thatconjunction. Then SK . M1 -* S; i.e., 'Sk . R(a, a) . R(a, b)' --'(y)R(a, y)'.
Therefore,'(Sk . R(a, a) . R(a, b)) D (y)R(a, y)' is an analyticsentence;'3also

13 Cf. Hilbert and Bernays, loc. cit., p. 155.

138 CARL G. HEMPEL

'Sk ((R(a, a) . R(a, b)) D (y)R(a, y))' is analytic. But then,in view of the
D
modus ponensrule,SK -* '(R(a, a) . R(a, b)) D (y)R(a, y)', and thus SK -r
'(y)R(a, y) v -R(a, a) v -R(a, b)'. But since SK, being completelygen-
eralized,containsneither'a' nor'b', it follows"4fromthelast statementthateven
SK -* '(x)(z)((y)R(x, y) v R(x,x) v -R(x, z))'; henceSK -R '(X)((Y)R(X,y) v
-R(x, y) v (z)-R(x, z))', and finally,again replacingthe sentenceon the right
handside ofthearrowby an equivalentone, SK -+ '(x)((y)R(x, y) v '-'R(x, x))'.
Let S'K be the sentenceon the righthand side of the last formula. Then, as
Cfdl(MI, T) forevery T e K, we have M1 -- DIM1 (SK); hence, since SK -* S'K,
Ml -- Dj1,(S'K) (by 4.2). But this last statementis false; for DIM1(S'K) is
equivalentto
((R(a, a) . R(a, b)) v --'R(a, a)) . ((RJ(b,a) IR(b, b)) v -'R(b, b)).
And the secondcomponentof thisconjunctiondoes not followfromMI. This
concludesthe proof,whichcan be extendedwithoutdifficulty to M2 = 'R(a, a).
R(a, b) . R(a, c)' and all analogouslybuiltlongermolecules.
This narrownessin the definition of 'Cf1' is obviouslydue to the factthat,
accordingto 6.2,a sentencewhichis notcompletely generalizedcan be confirmed
by a moleculeM onlyby virtueof beinga consequenceof M and directlycon-
firmedc.g. sentences;the relationof directconfirmation having so far been
to
restricted c.g. sentences. We shall now proceed to definea relationof direct
confirmation, Cfd2,whichcontainsCfd1as a subrelation, and whichis applicable
also to not-completely-generalized sentences. In termsof it, a corresponding
relationofconfirmation, Cf2.willthenbe defined.
7. A second approximation: Cfd2and Cf2. A definition of directconfirma-
tion whichis applicableto sentencesof any formcan be obtained fromthe
definitionof Cfd1(6.1) by simplydroppingthe requirement that S be a c.g.
of
sentence. This is possible because the concept C-development, which is
crucialforthe definitionof directconfirmation,
is definedforsentencesof any
form. Thus, we obtainthe definition:
7.1 Df. Cfd2(M,S) ifand onlyif (a) M is a molecule,and (b) M -* DIM(S).
In analogyto 6.2, we thendefine:
7.2 Df. Cf2(M, S) if and only if:
(a) M is a molecule;and
(b) (EK)(K -*S) . (T)(T e K D ((M -+ T) v Cfd2(M,T)))).
The new relationCf2is freefromthat materialinadequacyof Cf1 whichwas
exhibitedin the end ofthe precedingsection. Thus, e.g., each ofthe molecules
'R(a, a) . R(a, b)', 'I(a, a) . R(a, b) . R(a, c)', etc. stands in Cf2 (and indeed in
Cfd2)to '(y)R(a, y)'.
7.11 Note. Though the relationCfd2is morecomprehensive than Cfd1-
it containsthelatteras a propersubrelation-itcannotserveas a generalrelation
of confirmation, forit does not satisfyall the formalrequirements of adequacy.
Thus, e.g., 'R(a, a) . R(a, b)' standsin Cfd2to the first,but not to the second

14 Cf. Hilbert and Bernays, loc. cit., p. 106, schema a.

 A PURELY SYNTACTICAL DEFINITION OF CONFIRMATION 139

of the equivalentsentences'(y)R(a, y)' and '(y)(R(a, y) . R(a, c))'. This shows

that the requirements3.1, 3.11, 3.12 are violated and gives a preliminary justi-
ficationforthe introductionof the furtherdefinition7.2 above.
We note a fewtheoremsconcerningCfd2whichwill be needed in subsequent
proofs.
7.12 Theorem. Cfd2satisfiesthe conjunctioncondition3.13.
Proof. Let M be a moleculeand K a finitesententialclass such that Cfd2(M,
S)-and thusM -* DIM(S)-for everyS e K. Let SK be the conjunctionof the
elementsof K; then,by 4.1 B I c, DIM(SK) is the conjunctionof all DIM(S)
withS e K; henceM -> DIM(SK) and thusCfd2(M,SK).
7.13 Theorem. Cfd2(M, M) for every molecule M.-This followsimme-
diatelyfrom7.1 consideringthat DIM(M) = M (cf.4.13).
7.14 Theorem. Cfd2satisfiesthe generalconsistencycondition.
Proof. Let M be a consistentmolecule,and K* the class of all S such that
Cfd2(M, S). We have to show that {M}+K* is a consistentclass. Since
Ml K* (7.12), thisreducesto provingK* consistent.
E

Now, ifK* wereinconsistent,thenit would contain,by virtueof 2.32, a finite

inconsistentsubclass K. Let SK be the conjunctionof its elements;then,by
7.12, M -> DIM(SK), while SK is contradictory. This will now be shownto be
impossible.
IM will containa finitenumberof constants,say 'a,', 'a2', etc. SK may con-
tain individual constants; they fall into two classes (eitherof which may be
empty): the class C, of those constantswhichare elementsof IM, and the class
C2 of all others. (If, forinstance,M = 'R(a, a).R(a, b)' and SK = '(y)(R(a, y) v
U(a, c, d, y)', thenIM = {'a', 'b'}, C, = {'a'}, C2 = {'C', 'd'}.) Every element
of C2 will,of course,also occurin DIM(SK). Now since none of the elementsof
C2 occurs in M, and yet M -* DIM(SK), clearlyM must likewiseentail DIM(S)
for any S obtainable fromSK by replacingthe constantsbelongingto C2 by
arbitraryotherconstants.15 (Thus, in the previousillustration,M -> DIM(SK),
but also M -> DIM((y)(R(a, y) v U(a, a, e, y))), etc.) Now let us replace in
SK everyelementof C2 (ifany) by some one elementof IM, say by 'a,'. Let Sa,
be the DIM of the sentencethusobtained. Then clearlyM -> Sa.
Now, if the sentenceSK were contradictory, then it would not be satisfiable
in any finitedomain.'6 But fromthe precedingconsiderationsit followsthat SK
is satisfiablein a domainwhichhas the same numberof elementsas IM, say m.
Indeed, we can firstchoose a domain of m individuals, a,, a2, ... , am and define
each predicate occurringin M in such a way that on that interpretationM
becomes true. (This is possiblebecause of the presupposedconsistencyof M.)
Now, the followinginterpretation of the extra-logicalconstantsin SK will make
SK a truesentencein thedomainunderconsideration:'a,', 'a2', * , 'am', in so far
as they occur in SK, are to be names of a,, a2, ... , am respectively. All the
constantsof C2 in SK ('b,', 'b2',etc.) are interpretedas names of a,. For the

15 Cf. Hilbert and Bernays, loc. cit., p. 106, schema oa.

16
Cf. Hilbert and Bernays, loc. cit., p. 121, Theorem 1, and the statement on satisfi-
ability, p. 128, second paragraph.
140 CARL G. HEMPEL

predicatesin SK, choose the interpretation mentionedbefore,whichmakes M a

true sentence. Under this interpretation, the assertionof SK for the domain
la1, ** , am} clearly becomes equivalent withthat of S.. But underthe given
interpretation, Sa is true since M -> Sa. Therefore,thereis at least one finite
domainin whichSK is satisfiable;hence SK cannotbe contradictory.
The subsequenttheoremsconcerningCf2can now readilybe proved:
7.21 Theorem. Cf2satisfiesthe generalconsequencecondition.
7.22 Theorem. Cf2satisfiesthe generalconsistencycondition.
7.23 Theorem. Cf2satisfiesthe entailmentcondition.
7.24 Theorem. Cfd2is a propersubrelationof Cf2.
The proofscan be omittedhere: theyare exactlyanalogous to those of 6.21,
6.22, 6.23, 6.24; the proofof 7.22 makes use of the fact that Cfd2satisfiesthe
generalconsistencycondition(cf.7.14).
7.3 Theorem. Cf, is a propersubrelationof Cf2.
Proof. a) Let Cfl(M, S), i.e.,
(EK)((K ->S) . (T)(T e K D ((M -+ T) v Cfd1(M, T)))).
Then, since Cfd,is a subrelationof Cfd2(cf. 7.11),
(EK)((K -+S) . (T)(T e K D ((M -+ T) v Cfd2(M, T))));
hence Cf2(M, S).
b) Cf2is not a subrelationof Cf,: 'R(a, a). R(a, b)' stands in Cf2to
'(y)R(a, y)', but not in Cf,,as provedin 6.3.
7.4 Note. The question mightarise whetheran even more comprehensive
relationof confirmation, say Cf3,mightnot be obtained fromCf2by the same
procedurethat led fromCfd2to Cf2,i.e., by means of the followingdefinition:
Cf3(M,S) if and onlyif M is a molecule,and
(7.41) (EK)((K S-+ ) . (T)(T e K D ((M -+ T) v f2(M, T))))
And a reiterationof this proceduremightpromisea furtherbroadeningof the
relationof confirmation.Actually,however,all the relationsthus obtainable
are coextensivewith Cf2. It sufficesto show this forCf3.-If Cf2(M, S), then
certainlyCf3(M, S), forthe class K = {SI satisfiesthe condition7.41.-For
provingthe converse,note firstthat by virtueof 7.23, the clause '(M -> T) v
Cf2(M, T)' in 7.41 can be replaced by the equivalent 'Cf2(M, T)'. Now let
Cf3(M, S); then S is a consequenceof a sententialclass K such that forevery
T c K, Cf2(M, T); and this,by the definitionof 'Cf2',means that every T in K
is a consequenceof a class KT such that M entailsor standsin Cfd2to each ofits
elements. Thus, S is a consequenceof the sum of those KT, i.e., of a class such
that M entails or stands in Cfd2to each of its elements;but this means that
Cf2(M, S).
The followingexamplesare intendedto illustratethe characterof the relations
Cfd2and Cf2.
7.61 'P(a)' standsin Cf2to thesentence'(x)P(x)', and to all its consequences,
such as
a) 'P(a)', 'P(b)', 'P(c)', and any otherfullsentenceof 'P';
b) '(x)((P(x) v Q(x))', '(x)(Q(x) D P(x))', '(x)P(x) v (Ey)Q(y)';
c) '(Ex)P(x)'.
 A PURELY SYNTACTICAL DEFINITION OF CONFIRMATION 141

Also,themolecules'P(a) . P(b)', 'P(a) . P(b) . P(c)', etc.standin Cf2to all ofthe

above sentences.
7.62 'R(a, a). R(a, b) . R(b, a) . R(b, b)' standsin Cf2to '(x)(y)R(x, y)', and
so does 'R(a, a)' and again 'R(b, b)'-but no otherpartial conjunctionof the
firstmolecule. (Thus,whilea fullsentenceof a predicateof degree1 standsin
Cf2to any fullsentenceof the same predicate-cf. 7.61 a-an analogousrule
does not holdforpredicatesofhigherdegree.)
7.63 Among others, each of the followingmolecules stands in Cf2 to
'(x)(P(x) D Q(x))': '--P(a)', 'P--fa) v Q(a)', 'Q(a)', 'P--fa) * Q(b)'.
7.64 Each of the molecules'R(a, a)'; 'R(a, a) . R(b, a)', 'R(a, b) . R(b, a)
R(c, c)' standsin Cf2to '(x) (Ey)R(x, y)', but 'R(a, a) . R(a, b)' does not.
7.65 Cf2does not generallysatisfythe rule that if each of two molecules
separatelyconfirmsa hypothesis,then so does their conjunction;for while
intuitivelyplausible, this rule is incompatiblewith the special consistency
condition3.21. Thus,e.g.,ifM1 = 'P(a) . P(b)', M2= 'P(c)', 1 = '(x)P(x)
v (x)--.P(x)', then,as is readily verified,Cf2(Ml,SI) and Cf2(M2,Si); but not
Cf2(Ml . M2, Si), because M1.M2standsin Cf2to S2 = (Ex)P(x) . (Ex)-'P(x)',
whichis incompatible withSi, and Cf2satisfies3.21.
Considernow the followingmore liberal definitionof confirmation which
readilysuggests itself: Let Cfd3(M, S) ifeither Cfd2(M, S) or M is a conjunction
ofmoleculeseach ofwhichstandsin Cfd2to S; and let Cf3be definedin termsof
Cfd3in exact analogy to 7.2. Our last illustrationmakes it clear that this
intuitivelysatisfactory, more comprehensive relationwould violate the con-
sistency condition; for Ml.M2 would stand in Cfd3 (and thusin Cf3)to eitherof
the incompatiblesentencesSI and S2.
This case illustratesonce morethat one of the main difficulties in defining
confirmation lies in strikinga balance betweenthat liberalitywhich seems
desirableon intuitivegroundsand the formalstandardsof adequacy,especially
the consistencycondition.
7.66 There is howeverone otherintuitiveinadequacyinherentin the con-
ceptsofCfd2and Cf2whichcan be remediedby a slightmodification ofthedefini-
tions7.1 and 7.2. As was pointedout in 4.51, a generalizedsentencemayhave
an analyticor a contradictory C-development withoutbeinganalyticor contra-
dictoryitself. As a consequence of this fact, a moleculeM may stand in Cfd2
(or in Dscfd2,i.e., the corresponding relationof directdisconfirmation; cf.3.01)
to a generalizedsentenceS simplyby virtueofthefactthatthecardinalnumber
of IM is so small as to make DM,,(S) analytic(or contradictory, respectively),
whilethecontentofM, intuitively speaking,neitherstrengthens norweakensS.
Let, forexample,S, = '(x)P(x) v (x)'-.'P(x)', S2 = '(Ex)P(x) . (Ex)P(x)',
and M = 'P(a)'. Then Dlm(S1) = P(a) v '-'P(a), DIV(S2) = P(a) . P(a);
henceCfd2(M,Si) and Dscfd2(M,S2), whileby intuitivestandardsM will be
nordisconfirming
consideredas neitherconfirming any of the two sentences.
Similarcases can be constructedinvolvinghypothesesin severalvariables.
Thus the sentence
S, = '(Ex)(Ey)(Ez)(W(x, y,z). '-..W(x,z, y).-%sW(y,x, z).- -.'W(z,y,x))'
142 CARL G. HEMPEL

cannot be satisfiedby any interpretationof 'W' in a domain of less than 3

individuals;indeed,if C containsexactlyone or exactlytwoindividualconstants
then Dc(S1) is readilyfoundto be contradictory. This has the awkwardcon-
sequence that any molecule M which contains no more than two individual
constants-no matterwhetherit contains'W' or not-stands in Dscfd2to Si,
since DIM(S1) is contradictory.
of directconfirmation
These inadequacies of our definition can be eliminated
by a slightmodification,to whichwe now turn.
8. Final versionof the definitionof confirmation.The followingpair of
definitionsembodiesthe modifications in question:
8.1 Df. Cfd (M, S) if and only if (a) M is a molecule,(b) DIM(S) is not
analyticor S is analytic,and (c) M -+ DIM(S).
8.2 Df. Cf(M, S) ifand onlyif:
(a) M is a molecule;and
(b) (EK)((K-+ S) . (T)(T e K D ((M -+ T) v Cfd(M, T)))).
The previousillustrationsmightseem to suggestthat the definiensof 'Cfd'
ought to contain,in addition to the clause (b) an analogous provisionto the
effectthat DIM(S) is not contradictory, or S is contradictory. This restriction,
however,is unnecessary. For while it may happen-as in the case S =
'(Ex)P(x) . (Ex)-P(x)', M = 'P(a)'-that DIM(S) is contradictorywhile S
is not, we neverthelessdo not have Cfd(M, S) unless M is contradictory itself;
and the consequence that a contradictorymolecule confirmsevery sentence
appears as quite reasonableand is, anyhow,impliedby the entailmentcondition
(cf. 3.32).
As is readilyseen, the modifieddefinitionis freefromthose intuitivelyun-
desirablefeaturesof the previousdefinitions of confirmation whichwerepointed
out in 7.66; in particular,'P(a)' neitherstands in Cfd to '(x)P(x) v (x)P(x)'
norin Dscfd to '(Ex)P(x) . (Ex) -P(x)'; and, in the case oftheabove sentenceSi
containingthe predicate'W', neitherM1 nor M2 stand in Dscfd to SI because
they do not stand in Cfd to -Si; and this is so because -SI is not analytic,
whileboth DIM1Q(-Si) and DIM2(-Si) are.
8.11 Theorem. Cfd-isa propersubrelationof Cfd2.
8.12 Theorem. Cfd satisfies the general consistency condition.-This
followsfromthe analogoustheoremforCfd2(7.14) in viewof8.11.
8.21 Theorem. Cf satisfiesthe generalconsequencecondition.
8.22 Theorem. Cf satisfiesthe generalconsistencycondition.
8.23 Theorem. Cf satisfiesthe entailmentcondition. The proofsof these
threetheoremsare analogousto thoseof the corresponding theoremsforCf2.
8.3 Theorem. Cf(M, S) if and onlyif M is a moleculesuch that
(EK)((K -+ S) . (T)(T e K -+ Cfd(M, T)))
Proof. By virtue of 8.13, 'M - T' entails 'Cfd(M, T)'; hence the clause
(M-+ T) v Cfd(M, T)' in 8.2 (b) is equivalentwith'Cfd(M, T)'.
8.41 Theorem. If M is an analyticmolecule,thenCf(M, S) if and onlyif S
is analytic.
Proof. If S is analytic,thenM -+ S, henceCf(M, S) by 8.23; and ifCf(M, S),
 A PURELY SYNTACTICAL DEFINITION OF CONFIRMATION 143

then,by virtueof 8.3, S is a consequenceof a class K such that 111standsin

Cfd to each T e K. Since this implies that M -* DIM(T), DIM(T) must be
analytic;and by 8.1(b), everyT of thiskindmustbe analytic;hencealso S.
8.42 Theorem. A moleculeM standsin Cf to everyS if and onlyif M is
contradictory.(From 8.23 and 8.22.)
Finally,all the illustrations and generalcommentsconcerningCfd2and Cf2
whichare includedin 7.61 through7.65 applylikewiseto Cfdand Cf.
Thus, the conceptof confirmation determined by the definitions 8.1 and 8.2
satisfiesall our formalconditionsof adequacy and at least all those tests of
materialadequacywhichhave beenreferred to in thediscussionofanyone ofthe
definitionsof confirmation previouslyconsideredin thisarticle.
The presentstudyrepresents only a firstattemptto arriveat a systematic
logical theoryof confirmation.Its main objectiveswere to characterizethe
issueand itssignificance as clearlyas possible,to suggestcertainconditions which
any adequate solutionshouldsatisfy,and to prove that the problemthus de-
terminedcan be solvedin purelysyntacticalterms.
However,the proofas embodiedin the above construction of a syntactical
definitionofconfirmation is restrictedto languagesofthe comparatively simple
logical structureof the lower functionalcalculus withoutthe identitysign.
From the viewpointof formallogicas well as of the logicalanalysisof science
it appears highlydesirableto generalizethe definition of confirmation in two
respects. First, it should be so expanded as to become applicable to more
complexlanguage forms,such as the lowerfunctionalcalculus with identity
sign, and even the higherfunctionalcalculus; and secondly,it would seem
desirableto freethe definition of confirmation fromthe restricting condition
that the confirming sentencehas to be of molecularform. The generalization
of the conceptof confirmation in thesedirectionsrepresents perhapsthe most
importantopen problem for further research in this field.
QUEENS COLLEGE, FLUSHING, N. Y.