nwo carnar. ‘The Methodological Character of Theoretical Concepts 1. Our Problems Is piscusuioss on the methodology of science, i is customary and we fal to divide the language of scence into two pats the observation Junguage and the theoretia! language. The observation language uses tens designating obserable properties and relations for the descrip tion of obsenable things or events. The theoretical language, on the other hand, contains terms which may refer to unobservable events, n+ obsenable aspects ot features of events, eg, to micropatils ike electrons o atoms, tothe electromagnetic feld or the gravitational fld in physi, to drives and potentials of vious kinds in psychology, etc. In this til I shal try to clarfy the nate of the theoretical language and is relation to the observation language. The obsersation language wil be Briefiy described in Section 2 ofthis paper. Then a more detailed account of the theoretical Tanguage and the connections between the two nguages wil be given in Sections INV. ‘One ofthe main topics will be the problem of a eiterion of sgnif- cance for the thoretial language, ie, ect conditions which terms and sentences of the theoretical language must full in order to have 1 function for the explanation and prediction of observable ‘vents and thus to be acceptable a empirically meaningful. shall eave aside the problem of a cxiteron of significance for the observation lan suage, because thre sem to be hardly any points of serous disagree ‘ment among philosophers today with respect to this problem, at last ifthe observation language i understood in the marow sense indicated above, On the other hand, the problem forthe theoretical language i 4 very serious one. There are not only diageements with respect to the exact locaton ofthe boundary Kine between the meaningful and the ‘meaningless, but some philosophers are doubtful about the very pos 38 ‘TuPORBTICAL. ConcEPTS bility of drawing any boundary line, It is true that empiicis today generally agree that certain cxtetia previously proposed were too nar; for example the requirement that all theoretical terms should be defin- able onthe basis of those ofthe observation langage and that all theo- retical sentences should be trunsatable into the obsevation language. ‘We ate at present aware that these requirements ae to strong because the rules connecting the two languages (which we shall cll “rales of correspondence”) can give only a partial interpretation for the theo retical language. From this fat, some philosophers draw the conclason that, once the ear criteria ae liberalized, we stall find a continuous line from terms which are closely connected with obsenations, 8. ‘mas? and ‘temperature’ though more remote tems like ‘electromag netic Sel’ and ‘peifonction’ in physics, to those terms which have no speifable connection with observable events, eg. terms in speculative retaphycies therefore, meaninglness seems to therm merely a matter of degree. Tis skeptical postion is maintained also by some emits, Heinpel, for instance, has given clear and forceful arguments for this view (se his articles, (14) and (15)). Although be still regards the ‘asc idea ofthe empiricist meaning exiterion 2 sound, he believes that deep going modifications are necessary. First, the question of meaning- falnes cannot, in his opinion, be mised for any single term or sentence but only forthe whol system consisting ofthe theory, expresed in the theoretical language, and the corespondence rules. And secondly, even for this sytem asa whole, he thinks that no sharp distinction between reaningful and meaningless can be drawn; we may, at bet sy some thing aboot its degiee of confmation on the bass of the available observational evidence, or aboot the degree ofits explanatory or pre Active power for obserable evens. "The skeptics donot, of cours, deny tat we can dew an exact bound: ay line if we want to, Bot they doubt whether any boundary line is an adequate explication ofthe distinction which empcicists had orginlly Sn mind. They believe that, i any boundary lie is drawn, it will be tore or less arbitrary; and, moreover, that it will tar out toe either too narrow or too wide. That it is too narow means that some terms ‘or sentences are excluded which are accepted by scientists as menningfl; that i i too wide means that some terms or sentences ae included which sientically thinking men would not accept as meaningful My attitude is more optimistic than that of the skeptics. I believe 30adolf Carnap tha, lio in the theoretic Tague it spoil to daw an adequate boundary Tine which separates the scenifcaly meaningfal fom the eninge. 1 stall propose cteria of sigieancsy he citeion for theoreti ten willbe fered a Section Vand the question of its adequacy wl he examined in Section VII; he eiterion for theoreti ca sentences wl be gien in Sesion VIL “Two alternative forme forthe introduction of scenic concepts into cour twolahguge jtem will be explained and their compuatve we fais esmined (Sections IX and X). One kind is that of thetic concepts introduced nto the theoeal language by postults, The other Kind Tell "dspotion concepts” They may Be ntrodced into fn extended obeewatonlnguge. Concepts deine by scaled ope. Sonal dfnions andthe socalled intervening vara belong to this End. I salty to show that the ftedction fo the fom of thera concepts is mre uit anethedbecase i allo greater teed in the coe of concptal forms; mare it seams more in aod with the way the sient actualy wether concepts Tn te last sein, 1 est briely the pris and advantages of the we of theoretical concep in pyle. IL. The Observation Language Lo ‘The total language of science, L, is considered 28 consisting of two parts, the observation language Lp and the theoretical linguage Lz. 1 shall hete briely indicate the nature of Lo; the later discasion wll chiefly concem Ly and its relations to Le, Without actually specifying it, we assume that the logical structure of Lo is given. ‘This would in- clude a specification ofthe primitive constants, divided into logical and Aesciptive (ie, nonlogical) constants. Let the observational vocabulary Vo be the cass of the descriptive constants of Lo. Fusther, for each language patt the admitted types of variables are specifed. In Lo it may suice to se only individual variables, with observable events (including thing moments) taken as individuals. Thea rules of formation, which specify the admitted forms of sentences, and rules of logical deduction sxe given. ‘Let us imagine that Lis ued by a certain language community asa ‘means of communication, and that all sentences of Lp are understood by all members of the group in the same sense. Thus a complete in- texpreiton of Le is given. 0 ‘The terms of Vo ate predicates designating observable propertcs of events of Hhings (eg, “blue,” “hot” “lage,” et.) or observable rela tions between them (eg. “sis warmer than y” “x i contiguous to 3.” ete). ‘Some philosophers have proposed certain principles which restrict cither the forms of expression or the procedures of deduction in “the language,” in order to make sore that everything said in the language is completely meaningful. It seems to me thatthe justification of such requirements depends upon the purpose for which the language in ‘question is used. Since Ly is intended for the description of observable ‘events and therefore is meant to be completely interpreted, the require- ‘ments or at least some of them, seem to have merit, Let us consider ‘the most important requirements that have been proposed for some ‘or any language L. 1. Requirement of observability for the primitive descriptive tems. 2 Requirements of vasious degrees of stictnes for the nonprimitive deserptive terms. (a) Exphch deat (b) Reluebilty by conditional definitions (eg, by reduction sen- {ences se proposed in (5) )- ; 3. Requirement of nominal: the values of the variables must be con- crete, obserable entities (eg, observable events, things, or Hinge ‘moments 4, Reguoment of ftom, in one of hee forms of iaeesing tit fa) The rules of the Language L do not state or imply that the basic (0) Ganamn (Se ange ef of he aida ables) nite In technical terms, Eas at least one nite model. (b) Litas only fnite models. {e) There is a finite nomber n such that no model contains more than a individuals 5, Requirement of constructivism: every value of any vardble of Lis designated by an expresion in L. 6, Requirement of extensionlity. The language contains only truth functional connectives, no terms for logical or causal modal (necessity, posbility, et.) ‘Any Timguage fuliling these requirements is more dzeetly and more completely understandable than languages transgressing these limits ‘tons, However, for the language as a whole, the requirements are not justied; we shell reject them later for the theoretical language Lr. aRudolf Camap Since then we have ia the prt Lx all the freedom of expresion desired, ‘we may well accept some or ll of these requirements for Lo. ‘We have already accepted requirements 1 and 3. The dessin about requirement 2 depends upon out intention concerning disposition terms (eg, “soluble” “fragile” “Bexble"). We shall not include them in Lo ‘tel; thus Lo is here taken asa restricted observation languge falling ‘the stronger requirement 2(a). Later (in Section IX) the possibility of an extended observation language Lo, which alows the introduction ‘of disposition terms, will be explained. Another method consists in representing the disposition concepts by theoretical terms in Ly (See- tion X) Theva utente) lim fle in, Tle its easly posible to satisfy requirement 5. Further, we tke Lp as an extensional language; thus requirement 6 is fullled. III. The Theoretical Language Lx ‘The primitive constants of Le are, like those of Lo, divided into logi- cal and descriptive constants. Let the theoretical vocabulary Ve be the class of the descriptive primitive constants of Ly. We shall often call these constants simply “theoretical terms.” (They are often called “theo- retical constructs” of “hypothetical construct.” However, since the ‘term “construct” was originally used for explicitly defined texms or con- ‘cepts, it might be preferable to avoid this term here and use instead {he neutral phrase “theoretical term” (or “theoretical primitive"). This tse seems to be in better accord with the fact that iti, in general, not posible to give explicit definitions for theoretical terms on the basis of Lo.) ‘We may take it for granted that Ly contains the usual truthfunctional connectives (eg, for negation and conjunction). Other connectives, ‘egy signs for logical modalities (eg, logical necesity and strict impli- cation) and for causal modalities (eg, causal necesity and causal im plication) may be admitted if desired; but thei inclusion would require 4 considerably more complicated set of rules of logical deduction (as syntactical or semantial rules). The most important remaining problem for the specification ofthe logical structure concerns the ranges of values for the variables to be admitted in universal and existential quantifies, and thereby the kinds of entities dealt with in Le. This problem will be discussed in Section IV. 2 ‘THEORETICAL CONCEPTS A theory is given, consisting of a finite number of postulates formu- lated in Ly. Let T be the conjunction of these postulates. Finally, cor respondence rules Care given, which connect the terms of Vz with those (of Vo. These rules will be explained in Section V. IV. The Problem of the Admissibility of Theoretical Entities Tt seems that the acceptance of the following three conventions CI- (CB is suficient to make sure that Ly includes all of mathematics that is needed in science and also all kinds of entities that customarily occur in any branch of empiri Gonrentione soot the doin D of entities adit at vals of Cl. D includes a denumerable subdomain I of entities. C2. Any ordered n-tuple of entities ia D (for any finite n) belongs ako toD. C3, Any cass of entities in D Belongs also to D. 1 shall now indicate briefly how these conventions yield all the cus. tomary Kinds of entities referred to in scientific theories. To facilitate the understanding, I shall fist use the customary way of spesking and the customary terms for certain kinds of entities, and only later add a ‘warning against a possible misinterpretation of these formulations. Fisst about mathematical entities. Since the subdomtin I stipulated in Cl is denumerable, we may regard its elements a the natural num- bers 0,1, 2, ete. IF R is any relation whose members belong to D, then R may be construed as a clas of ordered paits of its members. There: ore, according to C2 and C3, R belongs also to D. Now’ the (postive and negative) integers can, in the usual way, be constructed as relations of natural numbers. ‘Thus, they belong also to D. Analogously, we pro- eed to rational numbers as relations emong integers, to real numbers 8 clases of rational numbers, and to complex numbers as ordered pais of real numbers. Furthermore, we obtain clases of numbers of these kinds, relations among them, functions (as special kinds of relations) whose arguments and values are numbers, then clases of functions, functions of functions, et. Thus D includes all those kinds of entities needed in the purely mathematical part of Lr ‘Now we proceed to physics. We assume that Lr is based upon a pat ticular spacetime coordinate system; thus the spacetime points are ‘ordered quadruples of real numbers and hence, according to C2, belong 8Rudolf Carnap to D. A spacetime region i a clas of spacetime points. Any particular physical system of which a physicist may speak, eg, a material body (ora process of radiation, occupies a certain spacetime region. When a physcit describes a physial system or 2 process occurring in it fora momentary state of i, he ascribes values of physical magnitudes (cg. mass, electric charge, temperature, electromagnetic Seld intensity, ‘energy, and the like) cither to the spacetime region at a whole or *0 its points. The values of a physical magnitude are either re names for ntuples of such. Thus a physical magnitude is a function whose arguments are either spacetime points or regions and whose values are either real numbers oF n-tuples of such. Thus, on the bass of our con ‘ventions, the domain D contains space-time points and regions, physical ‘magnitudes and their values, physical systems and their sates. A physi- cal system itself is nothing else than a spacetime region characterized in terms of magnitads. In a similar way, all other entities occuring in ‘Physical theories can be shown to belong to D. Peychologial concepts are properties, eelations, or quantitative mag nitudes ascribed to certain spacetime regions (usualy human organisms ‘or classes of such). Therefore they belong to the same logical types a, ‘concepts of physics, iespective of the question of ther difference in meaning and vay of definition. Note thatthe logical type ofa payeho- Topical concept is alo independent of its methodological nature, ¢, ‘whether based on observation of behavior or on introspection; plioso- ‘hers seem sometimes not to realize this. Thus the domain D includes ako all entities refered to in psychology. The same holds for all social ‘We have considered some of the kinds of entities refered to in mathe- anatics, physics, psychology, and the socialsciences and have indicated ‘that they belong to the domain D. However, I wish to emphasize here that his al about the admision ofthis o that kind of entity as vas ‘of variables ia Leis only @ way of speaking intended to make the use ‘of Ly, and especialy the we of quantied variables in Ls, more easly ‘undentandable. Therefore the explanations just given must not be ‘undentood as implying that those who accept and use a language of ‘the kind hete described are thereby committed to certain “ontological” doctrines in the traditional metaphysical sense. The usual ontological ‘questions about the “reality” (in an alleged metaphysical sense) of num: ‘ers, classes, spacetime points, bodies, minds, et, are pseudo questions 4 ‘THIEORETICAL. CONCEPTS without cognitive content. In contrast to this, thee i @ good sense of the word “rea,” viz, that used in everyday language and in science. It ‘may be useful for our present discussion to distinguish two kinds of the meaningful we of "real," viz, the common sense use and the scientific ‘use, Although in actual practice there is no sharp line between these ‘wo uses, we may, in view of our partition of the total language L into ‘the two parts Lo and I, distinguish between the use of “real” in con- nection with Lo, and that in connection with Lx. We assume that Lp contains only one kind of variable, and that the values ofthese variables are possible observable events. In this context, the question of reality can be raised only with respect to possible events, The statement that a specified possible observable event, eg, that ofthis valley having been a lake in earlier times, is real means the same as the statement that the sentence of Lo which describes this event is tue, and therefore means this sentence itself: “This valley was a lake.” of realty in connection with Lr the situation is in certain respects more complicated. If the question concems the reality of an event described in theoretical terms, the situation is not much different from the ealir one: to accept a statement of ralty ofthis kind i the same as to accept the sentence of Le describing the event However, a question about the reality of something like electrons in general (in contzadistinction to the question about the reality of loud of electrons moving here now in a specified way, which is a question ofthe former kind) or the electromagnetic field in general is ‘of a diferent nature, A question of this kind isin itself rather ambigu fous, But we can give ita good scientific mening, eg, if we agice to ‘understand the acceptance ofthe reality, sy, ofthe electromagnetic eld inthe classical sense asthe scceptance ofa language Ly and in ita term, say ‘Ey and a set of postulates T which includes the classical laws of the electromagnetic eld (sy, the Maxwell equations) as postulates for.” For an observer X to “accept” the postulates of T, means here not simply to take T as an uninterpreted calculus, but to use T together with specified rules of comespondence C for guiding his expectations by deriving predictions about fatore observable events from observed events with the help of T and C. 1 said previously that the elements of the basic domain T may be regarded 25 natoral numbers. But I warmed that this remark and the ‘others about real numbers, ec, should not be taken literally but merely 45Rudolf Camap 45 didactic help by ataching familiar labels to certain kinds of entities 0% to say it ina stil more cautious way, to cetain kinds of expressions in Ly Lat the expessionscoresponding tothe domain Ibe “0,” “0%” “07 et. To say tht “O” desgoats the number zero, "0" the num ber one, ete, gives merely the psychological help of connecting these expressions forthe reader with wsefl associations and images, but should not be regarded at specifying prt of the interpretation of Ly. All the interpretation (inthe strict sense ofthis term, ie, observational iter. pretation) that an be given for Le is given in the Crules, and thee function is essentially the interpretation of certain sentences containing Acsriptive tems, and thereby inditety the interpretation of the de seriptivetenns of Ve. On the other hand, the essential sevice that the cexpresions “O” ec. give, consists inthe fact that Chey represent pac ticular kind of structure (vizy a sequence with an inital member ut no terminal member). Thos the structure can be uniquely specified but the elements of the stracture cannot. Not because we ate ignorant of, their mature; they, there is no question of their nature. But then, since the sequence of natural numbers i the most clementary and familit example ofthe sequential stracture here in question, no harm is done in saying that those expressions designate entities and that these entities ate the natural number, as long as We ae not misled by these formula: tions into asking metaphysical pseudo question Inthe exer discussion of the observation language Lo (Section 1), ‘we considered certain restive requirements, Ike those of nominalisn, fnitsm, etc, and found them acceptable. However, the situation with, respoct tothe theoretical language is ently diferent. For Le we do “not elim to havea complete interpretation, but only the indirect and partial interpretation given by the corespondence cules. Therefore, we shoal fel ice to choos the lgial structure ofthis Ianguage ait best fits our needs forthe purpose for which the language i constructed ‘Thus here in Le there is n0 reson aginst the three conventions, although ther acceptance volts the fst Sve requitements mentioned in Section I. Fits, before the Crates are given, Ly, with the postalates TF and the rules of deduction, is an uninterreted calculus. Therefore te earlier requirements cannot be applied toi, We are fee fn the com struction of the ealculus; there is no lack of earity, provided the rales, ‘ofthe callus are clearly given. Then the Cues ae added. All they do sin fect, to permit the derivation of certain sentences of Ly from 6 ‘THEORETICAL. CONCEPTS certain sentences of Le or vice versa. They serve indirectly for derivar tions of conclusions in Lo, eg, predictions of observable events, from given premises in Le, eg, reports of results found by observation, or for the determination of the probability of a conclusion in Ly on the basis of given premises in Ly, Since both the premises and the conclusion Delong to Lo, which fulils the restricting requirements, there can be ‘no objection against the use of the C-rues and of Ly, as far asthe mean- ingfulness of the results of the desivation procedure is concerned. V. The Correspondence Rules C ‘There is no jndependent interpretation for Lx. The system T is in itself an oninterpeted postulate system. The terms of Vr obtain only an indtest and incomplete interpretation by the fact that some of them are connected by the rules C with observational terms, andthe remain ‘ng tems of V-are connected withthe fst ones by the postulates of T. ‘Thus it is clear that the rules C are essential without them the terms ‘of Vs would not have any observational significance. These rules must be such that they connect sentenoes of Lo with certain sentences of Ley for instance, by making a detivation in the one or the other direction possible. The paticular form chosen for the rules C is not essential “They might be formulated as rules of inference or as postulates. Since ‘we asume that the logical structute of the language is sufficient rch to contain all necessary connectives, we may assume that the rules C ‘are formulated as postulates. Let C be the conjunction of these core- spondence postulates. As an example, we may think of Le as a language ‘of theoretical physics, ased on a spacetime coordinate system. Among ‘the rules C there willbe some basie ones, conceming spacetime designa- ‘ions. They may specify a method for finding the coordinates of any ‘obserationally specified location, eg, the method wed by navigators for determining the position (the spatial coordintes: longitude, lat ‘tude, and altitude) and time. In other words, these Cues specify the relation R which holds between any observable location w and the co condinates x, y, 2 where x,y,z are the spatial cooedinates and & is the time coordinate of u, More actly speaking, the ration R relates to an observable spacetime region u, eg, an obserable event or thing a class u’ of coordinate quadruples which may be specie by intervals around the coordinate values x, 25, ‘On the bass ofthese Cues for spacetime designations, other Cues, aRudolf Carnap are given for terms of Vr, €. for some simple physical magnitudes like mass, temperature, and the like. These roles ae spatiotemporlly gen- eal, ie, they hold forany spacetime location. They will usally connect only very special kinds of valuedistributions of the theoretical magni- tude in question with an observable event. For example, a role might, refer to two material bodies w and v (ie, observable at locations uw and vs they must be neither too small nor too large for an observer to see them and to take them in his hands, The rule may connect the theo- retical term “‘mast” with the observable predicate “heavier then” as follows: “If u is heavier than y, the mass of u’ (i, the mass of the coordinate region u’ corresponding to u) is greater than the mass of v.” ‘Another rule may connect the theoretical term “temperature” with the observable predicate “warmer than’ in this wa then the temperature of u’is higher than that of ‘As these examples show, the Cres effect a connection only between ‘certain sentences ofa very special kind in Ly and sentences in Lo. The carler view, that for some terms of Vz there could be definitions in terms of Vo, called cither ‘conclative defnitios’ (Reichenbach) or ‘operational definitions’ (Bridgman), has been abandoned by most em- pivicits as an oversimplification (see Section X). The essential incom- pleteness of the interpretation of theocetical terms was pointed out in ry Foundations of Logic and Mathematics (6) and is discussed in detail by Hempel in (15, §3) and (16, §7). Moreover, it cannot be ‘required that there isa Cale for every term of Vr. If we have Cxvles for certain terms, and these terms are connected with other terms by the postulates of T, then these other terms thereby also acqite observa tional significance. ‘This fact shows that the specication, not only of the rules C, but ao of the postulates , is essential for the problem of ‘metningfulness. The definition of meaningfuiness mast be relative to a theory T, because the same term may be meaningful with respect to cone theory but meaningless with respect to another. Jn orde to have a more concrete picture, we may think ofthe terms of Vp as quantitative physical magnitudes, eg, functions from space- time-points (or Snite space timeregions) to real numbers (or ntuples, of real numbers). The postulates T may be conceived of as representing the fundamental laws of physics, not other physical statements, however well etablihed. Let us think of the postulates T and the rules C a6, boeing completely general with respect to space and time—that is 6 ‘THEORETICAL CONCEPTS as'not containing references to any particular position in space or in time. In the above examples, the C-rules have the form of universal postu- lates. A more general form would be that of statistical laws involving the concept of statistical probability (which means roughly, relative fre quency inthe long run). A postulate ofthis kind might say, for example, that, iF a region has a cetain state specie in theoretical terms, then there i a probability of 08 that a certain observable event occurs (which means that, on the average, in 80 per cent of those cases this event coceurs). Or it might, conversely, state the probability for the theoretical property, with respect to the observable event. Statistical correspondence rules have so far been studied very little. (The probability conception of the psifunctions in quantum mechanics might perhaps be regarded as an example of probabilistic Gules, as some customary formulations by physicists would suggest. I think, however, that this conception con- stitutes a probability connection within Le rather than between Lp and ‘Lo, What physicists often call “observable magnitudes,” eg, mass, posi tioa, velocity, energy, frequency of waves, and the like, are not “obser able” in the sense customary in philosophical discussions of methodol- ‘ey, and therefore belong tothe theoretical concepts in our terminology) For the sake of simplicity, in most of my discussions here I shall think ‘of the Crules as postulates of universal form. VI. A Criterion of Significance for Theoretical Terms My task isto explicate the concept of the empirical meaningfulness of theoretical terms. I shall use “empirical significance” or, for short, “significance” as a technical expression for the desired explication. In preparation for the task of explication, et me try to clarify the expl- candum somewhat more, ie, the concept of empirical meaningfulness in its presystematic sease, Let ‘M’ be a theoretical term of Vz; it may esignate a physical magnitude M. What does it mean for ‘MT to be empirically meaningful? Roughly speaking, it means that a certain sssomption involving the magnitude M makes 2 difference for the prediction of an observable event. More specifically, there must be a cettain sentence Sy about M such that we éan infer with its help a sentence Sp in Lo, (The inference may be either deductive, as I shall take it to be in the following discussion, or, more generally, probabi- Tistie) Iti, of course, not required that So is derivable from Sy alone. cyRudolf Camap tis clear that we may wse in the deduction the postulates T' and the ‘ules C. If now Sy contains not only ‘M but also other terms of Vs, then the fact that S, is deducible does not prove that ‘M’ is meaning. ful, because ths fact may just be due to the occurrence of the other terms. Therefore I shall equite that Sy contain ‘M’ as the only term of Vz. Now it may be that any assumption involving only the magni- ‘tude M isin itself too weak to Tead to an observational consequence, and that we have to add a second assumption Sj containing other terms ‘of Vs but not ‘M, Let K be the clas ofthese other terms. For example, ‘Sy may say tha, at a certain spacetime point, M hs the value 5, and ‘Sr may say that, a the some space-time point or in its surroundings, certain other magnitudes have specified values. If Sp can be deduced from the four premises Sx, Sx, T, and C, while it cannot be deduced from Sx, T, and C alone, then the sentence Sy makes a difference for ‘the prediction of an observable event, and therefore has observational significance, Since ‘M' isthe only descriptive term in Sy, ‘M’ itself has ‘observational significance, However, this result must be qualified by a ‘proviso, Since we have used the second assumption Sx involving the terms of K, we have shown only that ‘M' is meaningful provided that the tems of K ate meaningful. For this reason the definition of the significance of 'M' most be made relative not only to T and C, but also to the clas K. ‘Mis shown by the indicated procedure tobe significant ‘provided the terms of K have been found by a previous examination to be significant. Therefore the terms of Vr must be examined in a serial order. The fist terms of Vr must be such that they can be shown to be significant without presupposing the significance of other descrip tive temas, This will be the case for certain terms of Vr which are directly connected by Ccrules with Lo. Other terms of Vs can then be shown to be significant by using the proved significance of the fst terms, and so on. The total Vz can be regarded as significant only if we can show for a certain sequence ofits terms that each term is sig. nilfcant relative to the class of the terms preceding it in the sequence. It is clear that the definition must be relative to T, because the ques- tion whether a certain term in Ly is significant cannot possibly be decided without taking into consideration the postulates by which it fs introduced, Perhaps the objection might be raised that, if significance is dependent upon T, then any observation of a new fact may compel 1s to take as nonsignificant a term so far regarded as significant or vice 50 ‘eu. However, it should be noted fst that the theory T which is here presupposed in the examination of the sigaifcence ofa term, contains only the postulates, that i, the fundamental laws of science, and not ‘other sclentfcally asserted sentences, og, those describing single facts Therefore the cas of the terms of Ly admitted as significant is not changed whenever new facts are discovered, This class will generally be changed only when a radical sewlution in the yrtem of science is made, especialy by the introduction of a new primitive theoretical term and the addition of postulates for tat term. And note father that the criterion here proposed is such that, although the whole of the theory T is presupposed inthe eiteron, the question of significance i stil ised for each term separately, not only forthe vocabulary Ve as a whole. (On the bass ofthe preceding considerations, I shall now give defini tions forthe concept of significance of descriptive tems inthe theoreti cal language. The definition D1 will define the ausiary concept of relative signicance, ie, the sgoicance of ‘M'relatve to a class K of other teems. Then the concept of significance itself will be defined in ‘D2. According to our previous considerations, the concept of sgnifcance must furthermore be relative to the theoretical language Ly, the observ tion language Lo, the set of postulates T, and the correspondence ules (C. We presuppose that the specications of the languages Ly and Lg contain also a specification of the clues of descriptive tems, that i, Veand Vo, respectively. DI. A term ‘Mis significant relative to the class K of teams, wi respect to Ly, Lo, T, and C=», the terms of K belong to V>,‘M belongs to Vs but otto K, and there are thre sentence, Sy and Se in Leand So in Lo, sich thatthe following conditions ate fated: (a) Su contains ‘Mf as the only desrptive term. (b) The descriptive terms in Sg belong to K. (9) The conjunction Su TG coat (i, pol elf) (@) So is logiealy implied by the conjunction SySe-. (©) 5 isnot logealy implied by Su-7-C. ‘The condition (c} 1s only added to asore that the situation deseibed in Sy and Sx is possible, ie, not excluded by the postlates T and the ‘Grules; otherwise the condition (d) would be trvally fulfilled. ‘D2. term 'M,’ is significant with respect to Ly, Lo, T and C= oy Sage a ean of es Mi, «Me of Vn uch at ey fom ‘Mi; ('=I,..-» 1) is significant relative to the class of those terms which precede iin the soquence, with respect to Ly, La, and C. 1Rudolf Camap ‘The sequence of tenms referred to in D2 must obviously be such that ‘the fist term ‘M/ can be shown to be significant without the help of ‘other terms of Vz, In this case ‘M;’ satisfies D1; the class K is the null class; the sentence Sj. contains no descriptive terms; it i logically true and can therefore be omitted. In the simplest case of this kind, ‘M,’ foceurs in a Cerule, ike “mass” end “temperature” in our previous ‘csamples. Suppose that the fst three terms of our sequence are of the kind described, Then for the fourth term, the sentence Sx may contain any one or all thee of these terms. In this way we may proceed, step ‘by step, to other terms, which may be more and more remote from ditect observation {A slightly stronger criterion might be taken into consideration, ob- tained by the following modification of DI. In addition to the ventence Sk, another sentence S'y is used, which contains likewise ‘M’ as the ‘only deseeiptive team. ‘Then the analogue to condition (c) for S'x is added, and furthermore the analogue to condition (d) with S'y taking ‘the place of Sy and the negation ofS, taking the place of S."Ths here ‘the assumption Sy leads to an observable consequence, as in DI, but another assumption S'y about M, incompatible with Sy, leads to an- ‘other observable consequence. However, the simpler criterion stated in ‘DI seems suicient as ¢ minimum requirement for significance.) In the informal discussion at the beginning of this section, 1 have referred to the deduction of Sp frm certain premises. Correspondingly, ‘Di(a) requires that Sp is logically implied by the premises. However, this simple situation holds only if the C-postaates have universal form, as ‘we mostly assume in our discussions. In the more general case that also statistical laws are admitted as C-postulates (see the remark at the end ‘of Section V) and perhaps ako as postulates of then the result is a probability connection between Sy.Sz on the one hand, and So on the ‘other, in this case, the conditions (2) and (¢) in DI are to be replaced by the condition that the probability of Sp relative to Sw Sx, presuppos ing T und C, is different from the probability of Sp relative to Sx alone. VII. The Adequacy of the Criterion of Significance “The erteron here proposed is admitedly very weak. But this isa rest of the development of empiricism in these last decades, The ‘rginal formulations of the criterion were found to be too strong and too narrow, Therefore, step by step, more libeml formulations were 32 ‘THEORETICAL CONCEPTS introduced. Hempel has given in his article (15) «clear survey ofthis development. One change was the replacement of the principle of veri fablty by the weaker requirement of confrmability or testability, as formulated in my paper (5). At the time ofthat paper, I still believed that al scientific terms could be introduced as disposition terms on the basis of obsertion terms ether by explicit definitions or by so~alled reduction sentences, which constitute a kind of conditional definition (ce Section X). Today I think, in agreement with most empiricist, that the connection between the observation terms and the terms of theo retical science is much more inditect and weak than it was conceived cither in my earlier formulations o in those of operationism. Therefore a criterion of signifcance for Lr must likewise be very weak. Tn discusions of the requirement of confimmabity (or, in caricr times, veriabilty) the question was sometimes raised whether the possiblity of the event which constitutes the confirming evidence was to be understood as logical possiblity or as ausal possibility (i com patiblty withthe laws of nate oc the ws ofa given theory}. Accord ing to Shlck’s conception (22, p. 153) the possiblity should be under stood in the widest sense, a8 logical possibility. His main argument was the uncertainty about posibility in an empisal sense. He pointed out that the obsewer does not know whether certain operations ate empiri cally possible for him or not: For example, he does not know whether he is able to lift this table; he is quite certain that he cannot lift an automobile; but both events are stil conceivable and should thetefore be regarded at posible evidence. Schlek’s point was that a question of signifcance should never be dependent upon contingent facts On the other hand, Reichenbech and I (5, p. 423) maintained the view that logical possibility isnot sufficient, bat that physical (or, more generally, causal) possibilty is required. The question whether a given sentence of Lr is confimable most be taken as relative to a theory T. In examining such @ question, a proposed evidence or a proposed test procedure cosld certainly not be accepted if they were incompatible with T. For example, on the basis of modem physics, which takes the velocity of light as the maximum signal velocity, any proposed test or dence involving «signal with a higher velocity eould not be acepted a proof of signfcance, The defnition DI is based on this conception. ‘The conjunction Sy Se-T-C is required to be consistent by condi tion (c). Since Se i logically implied by this conjunction, Sy 3Rudolf Camap compatible with T and C and this causally posible, However, itis to be noted that cans possibility as here understood is much weaker ‘han the kind of empirical possibility which Schlick had seemed to have in mind. In Schlick’s eample, neither the lifting of the table nor that of the antomobile is excnded by our criterion, because these ‘vents are not incompatible with the T (and C); T contains only the fundamental laws of science, while those events are merely excluded by our empirical nowledge ofthe observer's abit to Tift things. [shall now examine the question of the adequacy of our criterion in more specific terms. Let us consider the case that the vocabulary Ve Consists of two patts, Vs and Vj , such that the terms of V, are empiri: cally meaningful, while those of Vs are entitely devoid of any empiticl meaning, To make this presupposition about Vs and Ve more specifi swe assume the following: (1) IES, and Sy are any sentences of L suc that all descriptive terms ‘of S belong to'V; or to the ebservational voeabulary Vo and those of ‘S10 Vz, then nether of the two sentences logiealy implies the other, unless the implying sentence is logically false or the implied sentence is logically true ‘Now a proposed criterion for the significance of terms of Vr should ‘be regarded as too narrow if it excluded aterm of Vand as too broad itt admitted a term of Vs. It would be adequate only if it were neither ‘00 narrow nor too broad. For example, we might think of V, 95 containing terms of physics, and of Vz 25 containing meaningless terms of speculative metaphysics such that the suppestion (1) holds. First lt us considera postulate system 1 consisting of two parts, Ts and Te, Ts containing only terms of V,, and T's only terms of Vi. T, may, for example, consist of fundamental laws of physics, and T's, ‘of metaphysical principles. criterion of significance which is adequate inthis special case can easly be given. We calla postulate of a sytem T an isolated postulate if its omission feom T does not diminish the ass of sentences in Lo which are deducibe from T with the help of the Crules, ‘Then we take a term of Vr as signcant if it occurs in a CCrle or in a nonisolated postulate of T. In the case of the above sgstem TY, according to (1), all postulates of T's and no others are colted; therefore all terms of V; and no others full the eriterion (of significance just mentioned, “TMBORETCAL, CONCEPTS ‘This criterion is, however, not generally adequate. It would, for cxample, not work for a theory T” logially equivalent to 1" but soch that no postulate of Tis isolated. Those who are sceptical about the possibilty ofa extein of significance for Lx have probably a situation of this kind in mind. (Hempel discusses a siular example.) They be- lieve that itis not posible to give aertetion for postulate systems like ‘T*, However, I think that the cteion for tens proposed in Section VI inadequate for cses ofthis kind. Consider forthe postulate system T” the sequence of terms which is requited in D2, This sequence must necessarily begin with physical terms of V,, besauss, according to our assumption (1), there are no Cres for any ofthe metaphysical terms ‘of Vp. Thea the sequence may go on to fate physical terms, which are connected with Le not dicetly by Cues, but indzeetly by other physical terms. Now we sll see that the sequence cannot reich any tecm of Vii thos our citeion is not too broad for systems like T”. ‘We will show this by an indiect proof. We assume thatthe sequence reaches terms of Vs; let ‘M' be the fst term of V, in the sequence; hance the preceding temas bong to V,, and thus are meaningfl. M" is significant relative tothe class K ofthe preceding terms, with sespect to Ly, Lo,T, and C, in the sense of DI Intuitively speaking, *M? must then be meaningful, in contadiction to our preposition about Vi. Our task i, to deve formally @ contradiction with the presuppos tion (1. ‘According to DI(} @) SySe-T".C > Sp is logically tue. Now "is logically equivalent to 1" end ths to T’,',. Hence we ‘obtain from (2) with a simple transformation: ) SaP'y 3 U is logically tue, where U is Se-T'.C 2 So. Hence: (4) Su-Ts logialy implies U. Now all descriptive terms in S/T belong to Vy, and those in U. belong to V, or Vo. Thus (4) is in contradiction to (1), because (6) Sy.T' is not logically false (by DI(c)), and (6) U is not logically true (by D1(e) ). “This shows that the sequence cannot reach the terms of Vs. 5Rudolf Carmap We have shown that our criterion isnot too broad ifthe given set (of postulates T” is logically equivalent to a set T” which consists of two pats, one containing only meeningfl terms of V,, the other only meaningless terms of Vs. The situation would be diferent fora theory TT that did not ful6N this condition. In this case, T mast include a postulate A such that A contains terms from both V, and Vs, but A is not logically equivalent to @ conjanction A:.Ae in which Ay contains only terms of Vs, and Ay only terms of Ve. But such a postulate A would express @ genuine connection between the occurring tems of V, and those of V,. Therefore these terms of V would not be entirely devoid of empirical meaning, against our assumption “The resi that ou criterion of significance isnot too broad depends cesentially on the following featare of our definitions. We refer in D2 to-a sequence of terms, and we require in effect for the significance of ‘term "M'of the sequence that ‘MT is significant (in the sense of D1) relative to the class K of the terms which precede ‘M' in the sequence and which therefore have already been found to be significant. We can easily se that the criterion would become too broad if we were to change D2 so as to give up the requirement just mentioned. More specifically, we can show the following. A meaningless term ‘My’ of Vs can, according to DI, be significant relative toa class K which contains, in addition to terms of Vi, also a meaningless term of Vs diffrent from “My, say ‘MY,’ We shall show this frst informally. The decisive point is that now, in distinction to our actual definition D2, we can have as the additional assumption Sx & sentence connecting the meaningless teem ‘My’ with a meaningful (physical) term of V,, say ‘My’ Ne ‘here may be a (metaphysical) postulate Ay of T which connects My with My. With the help of this postulate, we cam derive from the assumption Sy about Mp alone a sentence about M's; from this with the sentence Sx mentioned above a physical sentence about My, and from this with a suitable Crule an observation sentence. ‘The formal derivation is as follows. We take as @ postulate of T: (As) For every spacetime point, the value of M', higher than that (of My by one, ‘We take as an instance of a C-le: (Cy) Mile) 56 ‘THEORETICAL CONCEPTS where a is the set of coordinates corresponding to the location a re- ferced to in Sp. Finally we take Sy and Sy as follows: (Se) Mi) = M" ‘S.) Mae) = 4. Now we can detive from Su with Ax: 0 Mile) Inve with Se: (ii) Ms and hence with Cy cy Se: “Thus the condition (4) in DI is fullled. Therefore, ‘My’ is sgnif- cant relative tothe lass K of the terms ‘Mand ‘My ‘We have just scen that, inthe definition ofthe sigicance of MP relative to K, we mast not admit a meaningless term in K and thereby in the additional assumption Sp, because otherwise an observation sen- tence could be dived, leading to a deceptive appearance of significance. ‘This is indeed excluded by D2, However, DI allows ather premises for the derivation which contain meaningless terms, vie, postlates of T. Not only the pastultes which contin the meaningfel terns of V, and the term ‘M" in question are allowed but also postulates containing any terms of V,. Could this not led to the same false appearance of sg- niffanee for an actully meaningless tenn “MT as the use of meaning- Jess terms in Si would do? In the above example Sx connected a meat ingless tec ‘MY’ ith a meaningful term 'M;, and this fact led to the undesired reslt’ Now the we of T would lea to the same result if « postulate of T were to make a connection between those terms. For example, postolate might yield as an instance the sentence “M,(2?)= Mie)" which was used a5 Sx in the extlir example, Thus the same ‘bservation sentence So could be derived from Sy even without the use af any second assumption Sr. As an alterative, a postulate might state 2 connection between ‘Mand ‘M,’ in a conditional form, whic, ‘hough weaker, would likewise make possible derivation ofan observa tion sentence. Does then the fact that D1 permits the use of all postu- lates T make this definition inadequate? It doesnot, because the occur rence of a postulate making 2 genuine connection between a team of Vs and ane of Vs is excluded by oar presupposition that the tems of 'V, ate meaningful and those of Vz meanings. By vitue of such a 37Rudolf Camap postulate, the term of Vs (in the example, ‘M'’) would obtain some measure of empirical meaning, as we observed earlier in this section with reference tothe postulate A. The essential diference between the two cases is the following, Ifa sentence connecting a meaningful term with another term in an insepanble way (eg, by an equation, 2 con tional, a disjunction oF the like, in distinction to conjunction, which can be separated into its components) is a postulate or provable on the bass of postulates, then it i slated as holding with physical neces- sity; therefore it conveys some empirical meaning on the second term, On the other hand, ifthe same sentence is not provable but is mexely used as the additional asumption Sx in DI, then it has no such eect; it need not even be tre. ‘The preceding considerations have shown that our criterion of signif- ‘ance, formulated in DI and D2, i not too Tbeal. Tt does not admit 4 term enticely devoid of empirical meaning. Now we shall consider the question whether the criterion might be too narrow. Suppose that the term ‘M" has some empirical meaning. Then it wil be possible to derive an observation sentence from a suitable asumption § involving “Mt and otner tems. Could it then till happen that our erteion would exclude ‘M? The definitions D1 and D2, while permitting the inclusion ‘ofall postulates T and C among the premises for the derivation of the ‘observation sentence, allow in addition only the two sentences Sy and Su for which speci restrictions ae stated, especially the following: (1) Sx may contain only terms of Vs which are diferent from ‘MT and Ihave to be significant; hence the following terms are not allowed in Sr: (@) terms of Vs, (6) terms of Vo, (© The term “Mt. (2) Sw contains ‘M? as the only descriptive tem, ‘We will now examine whether these restrictions are natrower than is necessary and thus might led to the exclusion of a meaningful tem ah 1a, We found earlier that it is necessary to exclude the terms of Vs from Sx, because otherwise the criterion would become too broad. 1b. Ts it necesary to exclude the observational terms Vo from the premises? Could it not be that, for the derivation of an observational 58 conclusion $ from Sy, we need, in addition to T and C and the assump- tion Sx in theoretical terms, some assumption in observation terms, say S'0? This might well happen. But then the conditional sentence S'0 > So is desivable from the premises specified in DI, and this is a sentence in Lo. Thus ‘M’ would ful6ll Di, with the conditional sen- ‘tence taking the place of S. 1c and 2, The condition (a) in DI requires that Sy contain “M' as the only descriptive term. The question might be raised whether this ‘requirement is not too strong, Could not the following situation occu? “Mand the texms of K are meaningful, and So can ineleed be derived with the help of T and C from an assumption $ containing no other descriptive terms than ‘M' and the terms of K, but S cannot be split up into two sentences Sy snd Sx such that Sy contains only ‘M’ and Sr does not contain ‘M.’ Let us assume that the sentence S refers to spacetime points of a certain spatiotemporal region a. Then we can form sentences Sy and Se which fol6ll the requirements of DI in the following way. Since $ is supposed to be compatible with T and C, there ‘must be a posible distribution of values of M for the space-time points of the region e, which is compatible with T, C, and S. Let F be a Jogeal constant, designating a mathematical function Which represents such a value distribution, Then we take the following sentence as Sy: “Por every space-time point ina, the valve of M is equal to that of F." ‘This sentence Sy is compatible with T.CS. Then we take as Sx the sentence formed from $ by replacing the descriptive term ‘M’ by the logical constant ‘F. Then Sy contains ‘MT as the only descriptive term and Se contains only terms of K. Furthermore, $i logically implied by Si and Se. So is logically implied by $:T.C., according to our assump- tion, and hence also by Sy.Sx-TC. Therefore ‘M’ fulfills the defini tion DI "Thus we have not found a point in which our exterion is too narrow. VIII. A Criterion of Significance for Theoretical Sentences "The following two problems are closely connected with each other: fist, the problem of a eriterion of significance for descriptive constants, and second, the problem of the logical forms to be admitted for sen. tences. For the theoretical language, the connection between these problems is stil loser than for the obseration language. In the latter, 2Rudolf Carnap we may decide to have primitive predicates like “blue,” “col” “warmer than,” and the like, while we are still undecided as to the forms of sentences, especially of general sentences, and the structure of the logic to be built into the language. On the other hand, if we wish to have terms lke “temperature,” “electromagnetic field,” ete. as primitives in Lx then we need alo the accepted postulates for them, and thus we Ihave to admit real number expressions, general sentences with seal ‘number variables, ete Tt seems to me that the best approach to the problem of a criterion of significance for sentences isthe following. We look fist for sclutions to the two problems mentioned above; and then we take the most lib- cral criterion of significance for sentences which is compatible with those solutions. That isto say, we then accept as significant sentence any expression that has any of the admitted logical forms and contains only descriptive constants which are significant. (I have used a similar approach for Ly in (5).) I propose to apply this procedure now to Ly. ‘A criterion of significance for descriptive terms was given in Section V1. Some of the questions conceming the logical forms of sentences were discussed in Section IV, expecially the question of the kinds of variables to be admitted in universal and existential quantifiers. We decided to admit at last those kinds of variables and forms of sen- tences which are essential for classical mathematics. Without actually specifying here the details of the rules, we shall now assume that the logical forms of sentences have been chosen on the basis of the con- siderations in Section TV, and that the rules of formation for Ly have been laid down in accordance with this choice, Then, applying the procedure proposed above, we define as follows: D3. An expression A of Le is significant sentence of Le = oy (a) A satises the rules of formation of Lz, {b) every descriptive constant in A is significant team (in the sense of D2), ‘The procedure used in this definition might pethaps appeet as obvi- ‘ons. However, a closer examination shows that this is not the case. In fact, this form of the definition {aside from the question of its content, i, the choice of the particular rules of formation and of the particular significance criterion for terms) is not in agreement with certain very ‘arrow criteria of significance which were sometimes proposed. For o ‘THEORETICAL, CONCEPTS example, verifability as a condition for the significance of a sentence ‘was sometimes understood in the strict sense of the actual possibility of catrying out a procedure which would lead either to 2 verification ‘or a falsification of the sentence. According to this citeion, in con- trast to D3, the significance of a sentence is not only dependent wpon its logical form and the nature of the descriptive constants occuring in it, but alo upon the location in space and time refeced to and the development of technology. For example, an empiricist applying this narrow criterion would regard as significant 2 sentence asecibing an observable property P to a body in his laboratory, while he would reject as nonsignificant another sentence which ascribes the samme property to 1 body not accessible to him or not accesible to any human being, eg, Decause of technical dificalties or remoteness in space or time. However, even at the time of the Vienna Circle, we did not interpret the principle of verifabilty in this narrow sense, We emphasized that the principle required, not the actual possibilty of determination as ‘rue or false, but only the possibilty in principle. By this qualification we intended to admit cases in which the determination was prevented ‘aly by technical limitations or by remoteness in space or time. We sccepted, for example, a sentence about a mountain on the other side of the moon as meaningful. We stated the genecal rule that, if a de- scription of an event in our neighborhood is regarded as meaningful, then an analogous description of an event in prehistoric times, or an event on the earth before there were human beings, or before there ‘were any organisms, or at a future time when human beings will not ‘exist any mot, should likewise be accepted as mezningful. On the bass of this conception, the spacetime location referred to in a sentence was regarded as irclevant for the question of meaningfuliness this is in accord with D3, TF D3 is accepted and, in line with our earlier considerations in Sec- tion IV, all constants, variables and forms of sentences of classical rathematics are admitted in Lr, then the class of significant sentences of Le is very comprehensive. We must realize that it includes certain sentences for which no observational evidence can ever be relevant, eg, the sentence: “The value of the magnitude M at a ceetain space-time point i rational number,” where ‘M' is significant, But every physicist ‘would reject a language of physics so restricted that sentences of this and similar kinds were excluded. He would regard their inclusion as a aRudolf Carnap negligible price to be paid for the great convenience of using the whole of classical mathematics. It seems to me that no serious objections can bbe raised agunst these sentences, since it is in any case not posible to ‘sive an observational interpretation for more than a small part of the sentences of Ly. We should require no more than that for such a mag. nitade there are certain sentences which have an influence on the pre- diction of observable events and thus the magnitude itself has some amount of observational meaning. T wish to emphasize that the proposed criterion for the significance of sentences is not meant to guarantee the frutfulness of If all texms ‘of Ve ful6ll D2 and the postulates TT are in accord with the rules of formation, then these postulates are indeed regarded as significant. But this should by ao means be understood as implying that T must then be & scientifally satisfactory theory. T may stil contain postulates Which are of very litee use ftom a scientific point of view. But the ‘question of scientisc frutfulness of sentences and of a theory should bbe cleatly distinguished from the question of empirical significance. ‘There is no sharp boundary line between fruitful and useless hypotheses or theories: this is rather a matter of degre. It seems even doubtful ‘whether itis posible to formulate in a completely general way a defini- tion of a quantitative degree of fruitfulness of a scientific theory. It should be noted that the significance criterion for Ly cannot be simply absorbed into the rules of formation. These rules determine only the forms of sentences, not the choice of primitive descriptive terms. ‘The significance of these terms depends on other rules of Ly viz the list of postulates T and of C-postulates and the rules of logical dedue tion, asa glance at the essential condition (4) in DI shows. (The rules of deduction may be given either in a syntactical form, as rules of deriva tion ina calculus or in a semantical form, in terms of logical implication, have used in Dl the latter form because it is more comprehensive; it presupposes rules specifying models and ranges, not given in this article) IX. Disposition Concepts ‘Among the descriptive terms which do not belong to the observa- tion language Lo there are two different kinds, which today, in distinc tion to my previous conception, T should like to regard as essentially diferent, One kind is that of the theoretical terms, which we have a "THEORETICAL. CONCEPTS