* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download pdf
Intuitionistic logic wikipedia , lookup
Tractatus Logico-Philosophicus wikipedia , lookup
Meaning (philosophy of language) wikipedia , lookup
Truth-bearer wikipedia , lookup
Quantum logic wikipedia , lookup
Willard Van Orman Quine wikipedia , lookup
History of logic wikipedia , lookup
Naive set theory wikipedia , lookup
Laws of Form wikipedia , lookup
Law of thought wikipedia , lookup
Axiom of reducibility wikipedia , lookup
Interpretation (logic) wikipedia , lookup
Model theory wikipedia , lookup
List of first-order theories wikipedia , lookup
Falsifiability wikipedia , lookup
Mathematical logic wikipedia , lookup
Foundations of mathematics wikipedia , lookup
The Logical Study of Science Author(s): Johan Van Benthem Reviewed work(s): Source: Synthese, Vol. 51, No. 3 (Jun., 1982), pp. 431-472 Published by: Springer Stable URL: http://www.jstor.org/stable/20115753 . Accessed: 05/09/2012 19:47 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp . 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 [email protected]. . Springer is collaborating with JSTOR to digitize, preserve and extend access to Synthese. http://www.jstor.org VAN JOHAN BENTHEM STUDY THE LOGICAL OF SCIENCE* taken for often of science, between The relation logic and philosophy of usefulness of the criticisms fashionable current in fact is Although problematic. granted, should be taken seriously which there are indeed difficulties logic are usually mistaken, in the two mentalities" "scientific different other to do, amongst things, with having of logic is, or should be, a vital part of the theory 1). Nevertheless, (section disciplines a of notion to the devoted is this of bulk the this key make To paper science. clear, of this formal various in a logical perspective. explications First, "scientific theory" 3). (section logical theory is discussed notion are reviewed 2), then their further (section or in mathematics like those of Klein of grand In the absence programs inspiring one can here. do best is the this in metamathematics, Hubert ground-work preparatory of the and merits ends on a philosophical note, The paper discussing applicability ABSTRACT. formal of to the study approach science (section 4). CONTENTS 1 Introduction of Science 1.1 Logic and Philosophy of Science 1.2 Logicians and Philosophers 1.3 L?gica Magna 2 Formal Notions of 'Theory' from Hubert History David Hilbert 2.1 A Short 2.1.1 2.1.2 Frank 2.1.3 Marian Ramsey Przetecki 2.1.4 Joseph 2.2 A Systematic 2.2.1 Syntax Sneed Logical 2.2.2 Structures 2.2.3 Semantics 2.2.4 3 Formal to Sneed Perspective Pragmatics Questions 3.1 Properties 3.2 Relations concerning of Theories between Theories Theories Aftermath 4 Philosophical 4.1 What Is 'Application'? of Formalism 4.2 In Praise 5 Technical Appendix Notes References Synthese 51 (1982) 431^172. 0039-7857/82/0513-0431 $04.20 Copyright ? 1982 by D. Reidel Publishing Co., Dordrecht, Holland, and Boston, U.S.A. 432 JOHAN 1. 1.1 Logic VAN BENTHEM INTRODUCTION and Philosophy of Science an intimate exists connection between logic and the seem to of would be science obvious, given the fact that philosophy and Mill the two subjects were still one until quite recently. Bolzano even in in this their the twentieth work; but, century, main exemplify or currents like Hempel's view of science, 'hypothetico-deductive' an 'falsificationism' if often presuppose obvious, Popper's implicit, link with is true to an even greater extent of Carnap's logic. This That there of science, in con still echoing 'logical reconstruction' that of 1971. like Sneed what is the research So, temporary problem? contacts These rather superficial - going no are, on the whole, tradition is a deeper than elementary logic. The Carnap-Suppes-Sneed as there advanced of favourable but, well, applications exception; one encounters Padoa's occasionally, logic remain isolated examples: Theorem Method (1953) or Craig's Theorem (1953). (1901), Beth's or non like of modern Cohen's technique Highlights logic, forcing program of at all.1 Moreover, standard model theory, have found no applications the technical work which is being done often seems to lack contact with actual science. in an otherwise Are these problems just maturation promising or is there wrong calling for a marriage, something fundamentally is given by more and more people, rapid divorce? The latter diagnosis In the philosophy of Kuhn and, especially, Feyerabend. following the of view' is or, at science, they say, 'logical point inappropriate, extent the of The scientific best, inadequate (to reality being useless). or too 'com one should be concerned with is either too 'dynamic', plex' to be captured by formal tools.2 This type of criticism will not be discussed here. Either it amounts to such authors for different stating general personal preferences of science - indeed quite honour say history or sociology approaches, are adduced, able subjects - or, when these will specific complaints so to not be found illustrate much the inadequacy of logic invariably as that of the author's to be treated The problem logical maturity.3 here is rather that logicians and philosophers of good will have not to yet been able to get a successful enterprise going comparable foundational in mathematics. research THE LOGICAL STUDY OF SCIENCE 433 In the final analysis, there may be deep (and, no doubt, dark) a non-trivial reasons for this failure - prohibiting logic of science. And a struggle, to in fact, some logicians defeat without accept prefer the cover of manoeuvres like the following. under is the 'Logic or of mathematical deductive the of science, philosophy philosophy is for the philosophers.' natural science Or again: of 'Philosophy the of Science is, by definition, component (applied) logic'. pragmatic Both ways, the logician can remain at home. As I see it, however, it is never a at far too early for such conclusions. had the Logic good try - and this to is of science devoted the paper theory ground for clearing once said, the important such attempts. Like some bearded German the problematic situation, but to change it. thing is not to re-interpret 1.2 Logicians and Philosophers of Science It would be an interesting historical the adventures project to describe as compared of logicians in mathematics with those of formal in their fields of study. Subterranean of science grum philosophers has been quite hospitable to logic - even mathematics blings apart, like set theory, model absorbing whole logical sub-disciplines theory or recursion has No taken place, similar theory. development or biology. Far from physics experiencing common about should this worry making Schadenfreude, logicians cause with those philosophers or of science in foundational engaged not al has this research.4 methodological Why long ago happened causes may be advanced but there are also ready? Again, historical some serious methodological These should be mentioned obstacles. first, so as to look each other straight in the face. in 'mentality' between To begin with, there is a difference logicians however, in, e.g., of science. philosophers Briefly, logicians want seem content these often with philosophers more than usual This observation reveals the acade definitions. just between related disciplines. To see this, it mic animosity closely an (admirable) book like Reichenbach suffices to compare 1956 with and many theorems formal where 1973. Reichenbach discusses various formal issues concerning Suppes ever formulating solu Time without problems admitting of deductive tions in the form of elegant In the Suppes in theorems. volume, one to finds contrast, (and probably conceptual analyses leading formal results - say representation theorems guided by) beautiful 434 JOHAN VAN BENTHEM Robb's causal of Space-Time with the Special coupling analysis of Theory Relativity. This difference in mentality in goals: may well reflect a difference as an formal sense in versus aim in itself say, 'explications' Carnap's as a means formal definitions for obtaining desired theorems. Put another did not formulate their formal con way, Frege and Hubert of in to theories order called 'the logical ception papers publish structure of arithmetic', but in order to carry out their programs as laws of pure logic; proofs statements (derivation of all arithmetical of consistency). not does the of science know such But, philosophy as of that the In a say well, guiding programs 'Unity of Science'? a but is there subtle difference. The above sense, yes, mentioned and indeed they logical programs made claims which were falsifiable5; were falsified - witness, G?del's It is Theorems. e.g., Incompleteness so this characteristic which made them fruitful. the mainly (Compare case of the see In to it is hard contrast, analogous circle'.) 'squaring how programs like the Unity of Science, or even its implementations like 'fitting each scientific could be theory into the Sneed formalism'6 refuted at all. Who is going to unfold the really inspiring banner? To a certain extent, the preceding to a polite amounted paragraphs invitation to philosophers of science: invest more heavily in technical there looms the equally logical theory. But, on the other hand, obstacle of the self-imposed isolation of logic. Certainly, deplorable are first-rate out many nowadays logicians opening up to problems side the familiar circle of mathematics, of those the semantics notably of natural language. But the boundaries of logic should be set 'wider or so I will now try to argue. still and wider', 1.3 L?gica Magna both as 'mathematical Contemporary logic is a flourishing discipline, the handbook Barwise logic' (witness 1977) and as 'philosophical schemes like the one to be presented logic'. Therefore, organisational below might well be thought superfluous - being rather the symptom of a subject in its infancy. of logic (The most grandiose conceptions were drawn up in times of logical stagnation.) The only excuse for the is its modest to make realize aim, namely following people (or more or what could be. remember) logic is, THE LOGICAL STUDY OF 435 SCIENCE and however it Logic I take to be the study of reasoning, wherever occurs. Thus, an ideal logician is interested both in that in principle, and its products, both in its normative and its descriptive activity in both inductive and deductive aspects, argument.7 That, in all this, she is looking for stable patterns if one wishes) to study is ('forms', but innocent: the assumption of regularity underlies inevitable, any science. These patterns assume various shapes: the 'logical form' of a sentence of inference, the 'logical structure' of a book or theory, or in rules' discourse debate. 'logical Given any specific field of reasoning, the ideal logician chooses her Which level of complexity will be attacked: weapons. sentences, is theories?8 which perspective texts, books, inferences, Furthermore, most suitable; syntactic, semantic or pragmatic? which tools Finally, are to be used in the given perspective: which formal language, which type of theory of inference and of which strength? Thus she decides, e.g., to study certain ethical texts using a tensed deontic predicate - or a semantics "world course" logic with a Kripkean theory of a prag quantum mechanics using a propositional language receiving matic in terms of Even verification so, games. many interpretation aspects of the chosen field may remain untouched by such analyses, of course. dis then, there are various Fortunately, neighbouring to related interests be with consulted. ciplines On this view there is no occasion for border clashes, but rather for mutual trade with not just philosophy, mathematics and linguistics, but also, e.g., with psychology and law. These are not idle recom but important tasks. An enlightened like Beth, mendations, logician for instance, realized the danger of intellectual sterility in a standard of knowledge the genesis in advance from its gambit like separating to beyond the logical horizon, (thus removing psychology justification - witness Beth and Piaget 1966. Another by definition) type of project which should become is the among culturally respectable logicians of mathematical and juridical modes of systematic comparison itself (cf. Toulmin 1958). But, not even mathematical reasoning logic covers its chosen field in its entirety. A book like Lakatos 1976 makes - on the it clear how eminently view of logic, present logical subjects that is - have fallen out of fashion with the orthodox mathematico logical community. it may Finally, be noted that such cross-connections would also 436 JOHAN provide supposed in pure the dishes which, to flavour. form Only VAN BENTHEM after all, a spice like 'logical awareness' is the most insensitive tolerate palates spices .... 2. FORMAL NOTIONS OF 'THEORY' seem to be that of a in the study of science would concept in with the remarks made In accordance section 1.2, we will 'theory'.9 in mind in this section both its formal study keeping approach of results. First, a short historical and pleasant definitions sequence most than which be richer is given definitions may (2.1), already are aware of. Nevertheless, it should be emphasized that logicians a but tale for historical this is not a representative didactical account, a a rather special how is review there Then, purpose. showing an arsenal modern for the sys versatile and flexible logic provides The key tematic development of such definitions (2.2). Further logical theoriz of the to is the in order results, subsequent subject produce ing, enter intellectual interest of the the section (3). Hopefully, following as we will clear become advocated here proceed. prise 2.1 A Short History from Hilbert to Sneed 'the logical view' of a theory is that of a formal To many people, a a are set of axioms and formal whose components language, system, an apparatus It is one of theorems from these. of deduction deriving that they of the early modern achievements the amazing logicians so to unrealistic with this much notion. do extremely managed to for the the formalists fashionable blame it has become Nowadays, with actual this when of (that concept, practice compared 'poverty' never have denied this is conveniently But, forgotten). they would one them for their happy should rather their aims, congratulate given so often Like notion. in the but fruitful of this austere choice to it be of science, simple-minded. paid development aims may call for richer concepts. different in Nevertheless, E.g., cases one a needs the observation that additional many developing a in consists of and theory judicious interplay proof definition. Thus, a logical concern becomes of equal with definability importance The complications to be considered here are of a different derivability. - as will however whose from the kind, appear sequence, following THE LOGICAL STUDY to account for is how theme in theories natural science. empirical mathematics proper: main 2.1.1 David OF 437 SCIENCE of additional complexity our inside Still, story begins the Hilbert As is well-known, Hubert's of consistency Program proofs presup view of the above-mentioned mathematical theories, which had posed in the course of millennia of geometrical studies. But, developed a was as consis it view based of mathematics moreover, upon global a core more of 'finitistic' surrounded hull theories like abstract by ting or core set concrete The consists of theory. analysis simple manipu in some fragment of arithmetic. lations with numbers; say, encoded to become extended human comprehen beyond threatening are invented, theories infinite sion, 'higher' adding (possibly objects) to and indeed the of others, very process amongst speed up proofs arithmetical (Cf. Smorynski 1977.) discovery. one could a typical mathematical as a formalize Thus, theory a affair: 'concrete' translated two-stage part Tx (with language Lx) into some 'abstract superstructure' T2 (with language L2), or maybe in some mixed contained theory TX2 (with language Lx + L2). These These two set-ups are obviously discussed henceforth. related: for convenience, the latter will be the consistency of Tx was beyond doubt: but that of For Hilbert, - whence was the not10 the attempt to prove it by means within Tu was matter of Another of the out side Kreisel: Tx. range pointed by Hilbert assumed that such abstract extensions did not create new concrete to discover it easier for (they only make insights proofs this means that Tu is a conservative extension of Tx: them). Formally, if TXt2\-cp then Tx\-<p,for all Lx-sentences <p. this was refuted by G?del. and (Take for Tx: Peano Arithmetic, for Tu, say, Zermelo-Fraenkel Set Theory - <pbeing the relevant Liar the above notion of 'theory' remains inter Nevertheless, Sentence.) in view of later esting, especially developments. Before views is to be turning to these, one more aspect of Hilbert's noted. It would be rash to assume that the underlying logic of Tx will one one be predicate For want to have a more logic. thing, might - but this is not the issue here. It is at this constructive level11 logic All 438 JOHAN VAN BENTHEM that the complexity of Li-statements is important. Clearly, like 7 + 5= 13 are 'concrete' - but what about statements, seems to have allowed Hilbert universal statements, quantification? but combinations like Vxyx + y <x like Vx not, e.g., quantifier y; - unless reformulated these could be That 3y<p(x, y) constructively. are rather 'abstract' is brought out by the Skolem such combinations theoretical functions: equivalence introducing rather individual Vx3ycp(x, y)^3/Vx<p(x,/x). say Vx3yx < y as Only when such an / may be given constructively Vxx < Sx is this permitted. Thus, one uses only a fragment of predicate Ti.1213 logic in formulating 2.2.2 Frank Hilbert's Ramsey view of theories was One could inspired by mathematics. con 'finistic' introduces say, perhaps, part empirical concrete siderations of but (actual manipulation symbols), only in a sense. a A in wrote weak little short very later, 1929, Ramsey paper called for a long time. (Cf. Ramsey 'Theories', which lay dormant also and notably the role of theories, 1978). In it, he describes as follows. there is the so-called theories, First, 'primary empirical the prima facie descriptions of our various language' containing These descriptions may be observed (laws) regularities experiences. or reports of individual observations Then, in order to (consequences). a so-called this complex field, one introduces 'secondary systematize a of A will trans with level abstraction. language', higher dictionary that the into 'secondary' ones, and the latter will be concepts some means set of of axioms, from which the simple organized by are laws and derivable. consequences original for there are similarities with the Hilbert view. Notice, Evidently, to statements. 'laws' the above universal how instance, correspond than in this restriction is even more reasonable For empirical theories, at context. E.g., one may observe the original mathematical particles or state universal from like 'in certain positions, these, extrapolations at most is occupied this trajectory each position thrice'. But, a late 'primary' at each is somewhere localization like 'each particle principle to the theoretical amounts that a position claim function uses of various conditions lies in the possible Another analogy instant' exists. on the THE LOGICAL OF STUDY 439 SCIENCE In the Hilbert relevant Skolem functions. like these case, one would to be 'constructive'. In the empirical case, one would like them to be or continuous of constructive (and hence by means approximable measurable functions).14 In order to bring out the postulated character of the secondary a as an that stressed could be language, Ramsey theory regarded second-order viz. & sentence, existentially quantified 3Y(A(Y) Y A is the where the and D(Y,X))secondary vocabulary, 'axioms', D the 'dictionary' coupling Y to the primary vocabulary X. Thus, not too much ontological to theoretical entities: reality should be ascribed a methodological turn which is quite typical for Ramsey's philosophy in general. simple Notice that the whole is based point upon the following logical observation: then 3Y(A(Y) & D(Y,X))hS(X); ifA(Y),D(Y,X)\-S(X), where S is any primary statement. There is a connection here with Hubert's if have you (cf. Smorynski 1977): Program 'primary' reasons for believing in 3Y(A(Y) & D(Y,X)) then you have (consistency) reasons for in the truth of any primary consequence primary believing S(X) of the whole theory (conservation). secondary This syntactic point of view is quite true to the spirit of Ramsey's in the pre-semantical work, which appeared stage of modern logic. Its full impact only became as we clear in a semantical setting, however, shall see later on. In authoritative text books on the philosophy of science, one like Nagel finds the so-called 'statement of scientific view' 1961, an observational theories. There, as such a theory is taken to possess well as a theoretical The first is actual vocabulary. interpreted by Intermezzo. inspection, interpretation respondence measurement, derived and from so on the - the second receives a 'partial' first so-called 'cor through in for (taken Ramsey's granted to implement precisely. E.g., This division principles'. turns out to be difficult approach) like explicit of original strong proposals, definability ('reduction') to un theoretical out to observational turned be terms, vocabulary tenable. Nagel is not unaware of relevant logical work. He discusses mean to which be taken may Craig's Theorem, that, if one can find a recursive for some mixed axiomatization 'observational/theoretical' for its obser theory, then one can also find such an axiomatization 440 JOHAN VAN BENTHEM form of reduction.15 Although there are part alone: a weaker in this picture, 'semantic' elements its main emphasis is still syntactical. to find that Nagel does refer to Ramsey's It is interesting views, be it in passing, in his discussion of instrumentalist views of theories versus sentence was desig realist ones. On his account, the Ramsey statement ned to turn an otherwise undetermined theoretical form vational certain as a 'convenient short-hand' for some instrumentalistically a set statement. into of determinate observation complex reports) too turned things the other way around: an otherwise (Our account was a statement to theoretical about entities weakened strong specific statement. in existential is Nevertheless, merely Nagel right classify on the instrumentalist side - and it is ironical that the ing Ramsey (meant revived 'semantical' Ramsey has become popular with latter day realists.) In order to pick up the main thread of our story, we go back to a 'semantic conception when E. W. Beth proposed of theories' 1947, still in of Beth alive the work authors like F. (cf. 1947), Suppe or B. C. van Fraassen. as describing to Beth view theories proposes Briefly, or certain structures what you like), on 'histories', systems, (physical the pattern of Tarski semantics. Thus, e.g., laws of the theory may be on the trajectories viewed as restrictions open to the system in phase - and that before the first were These ideas space. prophetic flowering of logical model theory. A more influential author propagating similar views is P. Suppes, was to whose theories 'set-theoretic formulated approach predicate' and applied in the fifties already. 1957, Suppes (Cf. Suppes I960.)16 would be defined as a class of E.g., Newtonian particle mechanics structures certain set-theoretic systems) (time-dependent satisfying in set-theoretic Newtonian conditions formulated These terminology. - set axioms of mechanics may be taken as the (syntactic) theory then as of but this is not the apparatus deduction; serving underlying - let alone a full first-order made predicate-logical explicit usually We will have more to say about such a ianguage-free' axiomatization. (as it is sometimes called) below. For the moment, just note approach in which the historical of Klein's precedent 'Erlanger Program' was the geometry approached purely through study of 'structurally' on of certain transformations (invariants of) groups spaces. geometric Suppes did use model-theoretic tools occasionally e.g., Padoa's THE LOGICAL STUDY OF SCIENCE 44? the of Mach in his discussion method (cf. Suppes 1957). Nevertheless, not fit in the does with set-theoretic smoothly predicate approach on structural the model biased of theory being perspective logical our next author is more in line with the side, so to speak. Therefore, present exposition. 2.1.3 Marian Przelecki in formal semantics tradition of scientific is a strong Polish an elegant 1969 little the book and theories, provides Przet?cki this. is derived from earlier work by of (Its inspiration partly example a The author model Cf. 1978.) presents Ajdukiewicz. Giedymin account in the following theories theoretic of empirical steps. To an whose there is observational language L0 predicates begin with, There in a certain concrete of osten domain U0 by means interpreted means cases and of (More precisely, by extrapola paradigmatic tion from these.) This domain l/0 is gradually enlarged by us to an and galax (Thus, electrons empirical domain U of 'physical objects'. will be all those structures ies join dishes and babies.) L0-structures as a with domain U which contain l/0 (with its fixed interpretation) a substructure. is theoretical Lt Next, added, calling for vocabulary of the L0-structures. The relevant model-theoretic suitable enrichment of is that of expansion, concept consisting yielding L0+ Lt-structures some Lo-structure to which have been added (without interpretations Yet not all such expansions changing the domain U) for the Lrterms. are necessary will do: in terms of the Li 'meaning postulates' is a set T of the of the part vocabulary. Finally, syntactic theory some class of first-order L0+ Lrstruc L0+ Lrsentences defining are si?n. tures. rise to various e.g. logical questions, in terms of of definability of Lt-vocabulary a wealth and of notions the book contains Indeed, L0-vocabulary. results which cannot be reviewed here. (Recall that this is not meant to be a comprehensive the most historical survey.) Probably striking an 'analytical' a T into one is Przet?cki's to theory attempt separate This framework different concerning gives kinds without part A, containing meaning import, and a postulates empirical claims the real work of S 'synthetical' part making empirical doing of Poland certain The L0-models). (by excluding People's Republic = has invested many years of research into this elusive equation T 442 JOHAN VAN BENTHEM A + S, without answers. in this finding satisfactory Nevertheless, some have been raised. E.g., A connection, questions interesting should clearly be 'semantically in the sense that it non-creative'; should not exclude L0-structures: each L0-structure must be expand able to an L0 +Li-structure is a model which for A. A true model wants to find the side of this coin. Is it theorist, Przet?cki syntactic that A is 'syntactically in the sense that the only non-creative', derivable from it are the universally valid ones?17 This Lo-sentences is the precursor of a more general one (treated by type of question in related to which we now turn. Przet?cki publications) The L0 + Li-theory T has a class of models MOD(T). Accordingly be said it may structures which to describe the can be expanded ?L0 class MOD(T) to L0+Lt-structures of those L0 in which T holds. Notice the way in which Ramsey's idea is implemented: this a 'second-order class of Lo-structures receives existential' description in this way. Now, let us return to the original of version syntactic form. We have an L0-theory ideas, in the following Ramsey's T0 which is contained in the L0+ L?-theory T; i.e., (1) if T0\-(p then T\-<p, for each L0-sentence <p.(Extension) what about the connection between Then, MOD(T0) (those Lo are models structures which of T0) and the class MOD(T) ?L0 des at least, it should hold that T does not exclude cribed by T? Clearly, models of T0: (2) MOD(To) ?5contained inMOD(T) \ L0. (Ramsey Extension) But, the above syntactic T has Lo-consequences Hilbert (3) requirement does not guarantee requirement outside of T0. Now, there was of conservative extension, reading then T0hp, if Thp Extension) for each L^sentence indeed, (2) implies (3), as is easily imply (1).) Do the two conditions match answer is negative. And, seen. <p. (Conservative (Notice precisely? this: maybe also the old that (2) does Unfortunately, not the Let T0 be the complete first-order Counter-example: theory stating that Time has a beginning from there via a 1-1 0, and proceeds successor never making S ('to-morrow') operation loops. (Infinitely THE LOGICAL STUDY OF SCIENCE 443 to secure this.) The latter phenomenon is axioms are needed a T in the transitive theory having (finitely axiomatizable) explained such that Vxy(Sx = y -> Bxy), as well as a pro relation B ('before') & lEY), such that EO, Vx(Ex-^ESx), perty E ('early') Vx3y(Bxy to from 'late', never to (Time goes Vx(~IEx-?Vy(Bxy 'early' ->~lEy)). a form model for T0 numbers the natural (N, 0, S) return.) Now, to a model cannot be expanded for T.18 (More which 'empirical' many in full here.) take too much space to present would counter-examples a to this establish model-theoretic Thus, attempt duality has typical failed. Still, there is a result which one can prove, viz. the equivalence version of Conservative with the following weaker of Extension Extension: Ramsey (4) each L0-structure in MOD(T0) to some L0+ Lt-structure extension an L0-elementary has in MOD(T). the condition that the original situations (Lo-struc empirical be to L0+ Li with 'enrichable' theoretical entities suitable tures) structures which are models for T, is now related to their being thus 'extendable' - addition of individuals becoming allowed as well. Given the fact that this is a not unknown in procedure physics (postulating new particles, or whole planets) model-theoretic curiosity has led to a Thus, not unreasonable amendment on Ramsey's views. will have given an impression of Przet?cki's considerations These we model-theoretic turn to the most finally, maybe spirit. Now, influential author on formal philosophy of science in recent years: 2.1.4 Joseph Sneed In his book Sneed 1971, this author gave an analysis of classical particle mechanics which transcends the above in using a formal machinery various Since this work has become well-known respects. interesting cf. Stegm?ller (largely through allied work; 1979), it will suffice to are mention to those ideas which relevant the present discussion. only In the above picture of a theory, with ingredients T0, MOD(T0), T, it is implicitly assumed that a division into non-theoretical MOD(T), if you wish) vocabulary vocabu ('observational', (L0) and theoretical has been In effected this is lary (Lt) satisfactorily. practice, however, a difficult problem, as we have seen - and Sneed gives an ingenious us here). solution of which need not concern (the exact nature 444 JOHAN VAN BENTHEM on of his work Another fruitful practical perspective is the emphasis to apply a theory to a given empirical what it means situation. On the to show this becomes: how view, Ramsey by Sneed, adopted theoretical be introduced may concepts (forces, ('super-natural') and so on) turning the situation into a model for the whole wishes, theory T. will any Will such an addition always be possible? More precisely, model of T tL0 = {(pE L0|Th<p} (clearly, the most'fitting'choice for the a answer to is be model for T? The part T0) expandable empirical a we this T extension of and is conservative T0, negative: evidently, are asking if it is also a Ramsey remains extension. Thus, the problem like in the preceding In Sneed's sub-section. then, the terminology, not need be eliminable' from 'Ramsey vocabulary always to If one wishes it will have be this, postulated.19 theoretical the theory. we Next, most'. turn from Will questions about 'at least' to questions do? theoretical 'at about that Przet?cki (Recall just any expansion at this point.) This seems implausible. 'meaning postulates' are to be formulated. Sneed considers the suitable restrictions Thus, a there be for model T0, expan that, given any unique requirement sion of it to a model for T. Model Theory gives one the cash value of this proposal for a large class of languages. By Beth's Definability amounts to this of Theorem, explicit definability 'implicit definability' introduced reduc Lt in terms of L0 (on the basis of T). Thus, old-fashioned tionism would re-emerge! as Sneed This line of thought should not be pursued, realized, one wants because it is too 'local'. The kind of restriction is a more one, 'global' cross-connections concerning between expansions of What we have in reality is a class of empirical are in such a which situations, simultaneously empirical expanded are obeyed. of that certain 'constraints' Two typical examples way are the following such constraints different situations. in different particles the same mass values (Cl) ('Mass' Thus, Milky the earth Way. is a theoretical gets the empirical situations should receive through different expansions. function in Sneed's analysis.) same mass in the Solar System as in the THE LOGICAL the physical (C2) mass separate STUDY OF join of two particles 445 SCIENCE receives the sum of their values (whether present situation considered, - at some or not). earlier or later time - in the the picture of an empirical say, (T0, theory now becomes, - where C is the set of constraints. T, MOD(T), C) MOD(T0), Applying such a theory to a class E of empirical situations means finding a E of C simultaneous such that the expansion satisfying resulting class is contained at in MOD(T). E will have to be a least, (Obviously, 'suitable' candidate, in the sense that it is contained in MOD(T0).) Thus, In later publications is (cf. Balzer and Sneed 1977/8), this picture a of its set-theoretic 'lan stripped purely linguistic content, leaving tradition. E.g., MOD(T) in the Suppes is guage free' formulation an a X class of C becomes while Lt-structures replaced by arbitrary con class of subclasses of the L0+ Lt-universe certain satisfying ditions. inclusion is one of these.) This (Closure under set-theoretic move will be discussed in section 2.2. of 'constraint' the notion which complicates a model-theoretic of view. it is Thus, point a about In little alternatives. fact, possible speculate be several promising escape routes, open to further The first is to have one single domain for the language L0 exploration. the 'empirical in 1969), considering (as already proposed Przet?cki as its subdomains. situations' would then apply to Ramsey expansion it Clearly, above picture to worthwhile there seem to the whole is the from 'universal' L0-structure: immediately guaranteeing con straints like (Cl) above.20 Possible be that this objections might too global, and also that it is not obvious how all constraints becomes further complications. As for the 'globality', may be captured without notice that one need not think of a literally domain: it 'universal' a class of L0-structures would suffice to consider directed by model their direct union for convenience theoretic inclusion - introducing a This is conceptually observation second route which opens up only. as well. attractive a Up to now, we have followed theory in assigning ordinary model class MOD(To) in mathematical classes of 'isolated' L0-structures practice, one is normally of such models, with relations to a theory T0. But, already confronted with 'structured' and operations connecting 446 JOHAN VAN BENTHEM be natural, it would e.g., to replace MOD(T0) by the mor of their natural T0-models category CAT(T0) plus connecting of then choice will be The dictated phisms. particular morphisms by of the theory considered. think of the necessities E.g., one might On this view, constraints (as substructures). isomorphic embeddings come to expand a class of L0-structures in as follows. One wants structure: class should i.e., the expanded together with its categorial - the be an Lt-subcategory original morphisms remaining morphisms.21 them. Thus, Put the forgetful should yield CAT(T0) when fashionably, functor to Some reflection shows that this takes care of the CAT(T). applied as As it the will even be constraint well.22 stands, (Cl) requirement too strong in many cases. E.g., it seems reasonable that isomorphic trans situations up to a co-ordinate empirical (being equivalent but not each single say) receive isomorphic expansions; an remain need Lo-isorphism L0-\-Lt-isomorphism! (E.g., not each a metrical of remain need purely topological automorphism Space in with when is metricized the accordance Space isomorphism, This second well have considerable model may possibility topology.) or it accommodates all constraints interest in itself, whether theoretic formation, not. section 5.) (An example will be found in the appendix, our historical A This concludes very rich and promising survey. formal notion of 'theory' has emerged - whose logical study is still in the fate of the usual properties of and its infancy. For example, terms defined in of relations between such theories (normally formal and dis is still to be explored. Some relevant systems) suggestions it should be remarked that cussions will be found in section 3. Finally, of contemporary this survey is by no means logical representative a of theories, model-theoretic views standpoint. being inspired by our scope outside tracks have remained different Thus, interesting are new Wessel That there frontiers for e.g., (cf., 1977). logicians will was our have become and that clear, however; abundantly goal. 2.2 A Systematic Logical Perspective in that it A phrase like 'the logical structure' of a theory is misleading one one to the suggests study of single (or preferred) logical approach but there is one single logical research mentality, science. Certainly, an aspect of that is precisely the plurality of available logical ap on one's Thus, depending specific aims, there are many proaches. THE LOGICAL STUDY choices available 1.3). For (cf. section in particular, the following perspectives 2.2.1 OF SCIENCE 447 the study of scientific theories, should be kept in mind. Syntax in 2.1 already how the syntactic notion of a formal remarked a to out be for many enquiries. It was system turned happy choice a arose from also recalled that this notion of long development - in which the axiomatic of deductive Euclidean organisation geometry its value as a means for efficient organisation and a knowledge proved stimulus for rapid development of deductive Even so, one knowledge. has a variety of syntactic in the actual computer approaches. E.g., in the Eindhoven formalization of mathematics AUTOMATH project a syntactic text logic (cf. Van Benthem 1979), one needed Jutting lambda-calculus in a natural of a version of the typed consisting a from providing of truths, deduction Apart catalogue presentation. It was a model such a system also provides for actual mathematical prose. are a natural and interesting (Indeed, modern computer languages at the text level of the syntactic approach extension in logic.) Similar no doubt for the natural sciences would projects inject invaluable a priori debate into the now often practical experience purely concerning 2.2.2 the possibilities of their formalization. Structures was made no distinction in geometry be the advent of and spatial intuition. After led to the concept of this separation non-Euclidean geometry, structures certain syntactic ('spaces') modelling purely geometrical sentences. is exemplarily sets of geometrical The resulting interplay 1959. But, purely 'structural' views of illustrated in, e.g., Tarski arose very early too: witness Klein's theories 'Erlanger Program' to in section 2.1, or Poincar?'s mentioned approach group-theoretic Before tween the 19th century, derivation syntactic and mechanics. geometry In the natural sciences, lip service was paid for a long time to in structural views lie even more at axiomatic ideals; but, practice, hand there. E.g., Newtonian mechanics be identified with might easily a certain class of real 'mechanical Which brings one to the systems'. are use in the logic most for 'structures' suitable which question just of natural science. 448 JOHAN VAN BENTHEM a start, let us quickly of dispose is abstract 'Mathematics about objection. This is true; about non-formal Reality'. or deal with are models natural scientists For a 'philosophical' pseudo natural science structures, it is too true. What only of reality, representations models? which these go into the logical study of science. So, seems of 'structure' rich model-theoretic notion the True, ordinary to kind of in accommodate any system (cf. Suppes enough principle - but it would also be natural to 1974) try and work 1960, Montague of their in the sense of general system theory, because with systems Padulo and Arbib 1974.) (Cf. time-dependent presentation. explicit and E.g., representation system-theoretical theory contains many 'logical' lessons.23 it may be of interest to observe that a classical In this connection, its main formulates of in the mechanics paper quantum logical study a struc in with result on 'hidden variables' (a topic syntactical ring) as a and than result about classical rather tural Hilbert space terms, Kochen and axiomatic theories. (Cf. Specker quantum-mechanical 1967.) form can be not all this mean that the ladder of linguistic Does thrown away, now that one has arrived at the structural 'reality' seem to be the 'set-theoretic theories? This would scientific behind indeed for many problems this is a earlier on. And view' discussed - witness and work Sneed to much sensible do by perfectly thing of which the obvious systems problem E.g., presentations Suppes. the same 'real' system describe co-ordinates, (in different say) is a about classes. suitable structural Or, when isomorphism problem finite machines relations of between talks about (cf. analogy Ashby are machines the that 1976), mutually homomorphic Ashby insight owes theories. to underlying Other syntactic nothing full the per may structural-linguistic require questions interesting as will be illustrated below. however: spective, be observes the friction one sometimes If this be so, then why and the 'model-theoretic' tween the 'set-theoretic' (Cf. approach? in such cases, there are priority disputes 1974.) Inevitably, Przet?cki is 'better'V* E.g., on the the atmosphere: whose perspective poisoning one loses that it is claimed set-theoretic side, except maybe nothing models for mathematical troubles theories, which (with non-standard one did not want anyway). however. Why This seems short-sighted, isomorphic not leave the door open, e.g., to 'non-standard mechanics', when THE LOGICAL STUDY OF SCIENCE 449 are just discovering the delights of non-standard analysis? ladder forever means away the linguistic Throwing cutting oneself off from the original motivation of many structural A concepts.25 similar warning was given by Dieudonn? to mathematicians wanting to apply logical gadgets like ultraproducts without being bored with mathematicians stories about their logico-linguistic of course, origin. This is possible, but one deprives oneself from understanding as their full importance, well as the heuristic fuel for further discoveries in the same vein. 2.2.3 Semantics Since logical model field (cf. Chang and theory is such a well-known Keisler to just a few 1973), it will suffice here to draw attention relevant To begin with, here is a short list of questions points. are meaningful theories which per concerning only in a semantical spective. as a structural notion - that is, 'determinism' (1) Does mechanical is possible into the future given ? past history, only one continuation as a linguistic (in the relevant set of structures) imply 'determinism' at time t + 1 are explicitly state variables notion: i.e., the relevant definable in terms of those at time t - on the basis of the mechanical axioms? 1976: 'Deterministic Theories'.) (Cf. Montague "a point relation is said to be objective if it is (2) In geometry, invariant with respect to every automorphism. In this sense the basic are objective, relations and so is any relation in defined logically terms of these. relation may be so [...] Whether every objective defined raises a question of logical completeness [...]' (Weyl 1963, p. to be sure. (Cf. the appendix 73): a model-theoretic (section 5) question, for some relevant results.) Finally, the direction may also be reversed, as in - viz. that like Einstein's (3) Do principles' 'relativity linguistic transformation be invariant under the Lorentz 'physical equations' admit of a structural characterization? One could ask scores of similar questions, say about the relation between 'linear' functions and 'linear' polynomial forms, all of them to look for a the same peculiarity of model illustrating theory: systematic Another and structural duality between syntactic constant attention point which deserves points of view. is the following. 450 JOHAN VAN BENTHEM one semantics should not make of plain Tarski great success link structures in between and is any language forget that, principle, interest. Thus, e.g., non-classical semantics of model-theoretic of the or the 'vague' semantics of Fine 1975 are equally forcing variety, a In the choice of certain class of then, respectable. particular, structures does not commit one yet to a fixed underlying apparatus of The or otherwise.26 This observation, deduction: classical, intuitionist, by one with an excellent to make more the way, provides opportunity exact sense of the famous in Quine Thesis' 1951.27 'Revisability 2.2.4 Pragmatics them, syntax and structures (and hence: Model Theory) seem to exhaust all possible upon theories. logical perspectives our actual hand themes concerning But, moreover, many pragmatic can notions of above be studied by logical means. the Thus, ling as activities rather than products theories scientific of such activities are not irrevocably outside the scope of logic. Just a few examples Between would have to provide the backing for this claim here. In the purely syntactic there is already the matter of perspective, of proofs. the heuristic Lakatos in mentioned section 1976, 1.3, (Cf. or Hintikka and Remes the logical study of the 1974.) Moreover, or attack of sentences in dialogues actual defense has been initiated in works like Lorenzen and Lorenz 1978. In the semantic stu too, there is room for pragmatic perspective will that successful dies. E.g., model has theory presupposes interpretation seman taken place already. How? Here is where 'game-theoretical tics' in the style of Hintikka useful may become (cf. Saarinen 1979). There is also a connection here with actual measurement (cf. note 15), as well as with to physical Giles' theories (cf. Giles 1979). approach an 1979 volume Hooker the Indeed, signals interesting 'logico-prag turn in the study of physical it will be matic' theories. Nevertheless, are only the first landmarks clear that these references in a hopefully fruitful new area of logic. an appropriate the pragmatic perspective occasion Finally, provides for stating a concluding remark about the notions here. considered Exclusive concentration reveal a lot about the upon 'edifice as theories of science', intellectual but it may products may also generate THE LOGICAL STUDY OF 451 SCIENCE in jeopardy when of science pseudo-problems. E.g., is the rationality and two of its beams do not fit exactly - say, Newtonian mechanics be Will 'continuous Einsteinian mechanics? Of growth' endangered? course is guaranteed not: the continuity of science the (re-) by worker's know-how of common intellectual construction rules, stra tegies, whole and the theories: 3. like. Take away specific axioms, languages, this cat will still retain its rational grin. FORMAL QUESTIONS CONCERNING or even THEORIES as the illustrates how logic may be viewed section preceding a non-trivial scientific More of whole theories.28 study specifically and tools was A theory reviewed. range of logical considerations should always be studied with specific questions in mind and these This brings us back to call for a suitable choice of logical perspective. The one made in section 1.2 about logical research the observations on the the lure of theorems horizon. Here a related being guided by will treated. What becomes of the be question logician's pet theorems when her subject is used in the present context? To take just one is rather dear to her: does Lindstr?m's curious weed which Theorem a in this more have any significance In area? sense, yes, general insofar as this theorem in general) is (and abstract model theory of concerned semantics. with the general of logical properties languages plus It has been argued that scientific theories may be based upon quite different logics; and abstract model theory may help reasons meta for choices, by telling one which provide pleasant theorems will be gained or lost in each case. (Cf. Pearce 1980A for an of abstract model application theory to the Sneed framework.) some More favourite systematically, topics and results of logical now (be it very sketchily) with respect research will be reviewed to to the study of scientific their possible relevance theories This will serve as a first measure of the required angle in general. of logical re-orientation. To be sure, there exist some important logical topics already in the or analyticity. of empirical theories, like causal explanation metascience The point is that we want more. To begin with, classical research into the foundations of mathematics area. has produced mainly results in the following 452 JOHAN VAN BENTHEM 3.1 Properties of Theories In the syntactic of notions with a logical tradition, there is a multitude us text book Let follow the traditional list: special just pedigree. then the familiar triad of axioms, derivability, definability, forms and, finally, decidability independence, consistency, (or completeness, of the well-known standard of the theorem set). Many complexity results of logic are in this area: Craig's Theorem, G?del's Theorems, Church's Theorem. Some of the original these problems inspiring seem to become less urgent in the case of empirical would theories. E.g., consistency will be a minor concern - the much-praised 'physical one might want intuition' being there to keep us straight.29 Similarly, so to 'neutralize' to G?del's rather for results, speak, searching our the of non-mathematical theories. for parts 'partial completeness' are ones in in the structural tradition often mathematical Questions even their own right, having become in science respectable empirical a before the adolescence of modern One of logic. example pertinent result which be called in spirit is the theorem in might 'logical' 1964 to the effect the that the only mappings between are those reference frames preserving causal connectibility physical of the Lorentz group. Some relevant model-theoretic in topics were mentioned already section 2.2.3. A more systematic list follows the semantic duals of the of axioms above of (more generally: topics.30 Forms complexity are so with connected the to theorems dear definition), preservation con This the of relation between structural closure topic logicians. seems of wide ditions and explicit In this con interest. definability Zeeman like Keisler's characterization of elementary out first-order of structural data in definitions classes, up conjuring terms of closure under isomorphisms and ultraproducts, should have value. Next, theorems great inspirational completeness linking syn tactic derivability with structural retain an obvious consequence is the finding in Giles interest. (A nice example 1979 that the logic of a nection, key results Lukasiewicz's certain system Loo.) 'physical game theory' is precisely and Beth's Theorem Padoa's Method tying up explicit definability with structural determination have already been applied repeatedly in the theory of empirical science 1960, Montague 1974, (cf. Suppes Rantala retreats into 1977). Of the above-mentioned triad, consistency - the for theories being presup the back-ground of models possession THE LOGICAL STUDY OF SCIENCE 453 has fallen out of Moreover, independence posed model-theoretically. themselves favour with the logicians already, being of purely acade with the important is connected mic interest. Completeness, however, im of categoricity, which may have wider structural phenomenon portance. clear from this quick look is not that all these logical becomes as they stand - only that this type of result remain interesting and proven in the interest if it can be formulated remains of potential as more complex in section theories of developed empirical setting Take Tennenbaum's chosen. 2.1. Here is one example, arbitrarily are the theorem beautiful stating that the standard natural numbers What results with recursive operations of Peano Arithmetic only countable model Here is a more of addition and multiplication. 'empirical' one in the same spirit. The only countable models for linearly ordered Time the which are homogeneous (in some natural sense) are the rationals, van that and their order). (Cf. (in product integers lexicographic Benthem about a result in this spirit for mechanics? 1980.) What or the succession of scientific Interest in the edifice of science, soon one will lead if into the theories wishes) ('scientific progress', area. less settled following 3.2 Relations between Theories in have been discussed theories relations between Some syntactic extension. conservative extension and 2.1 already, section notably Important relevant results are, e.g., the theorem of Craig and Vaught 1958 stating that each recursively axiomatized theory ( in a language or additional with identity) is finitely axiomatizable using predicates; trans well be results. Such questions G?del's non-conservation may a con is classical mechanics sciences. ferred to the empirical E.g., would this be an of classical servative extension not, (If analysis? to applied theory.) from application interesting example of 'feed-back' These notions of extension may be regarded as special cases of the relations of interpretability and embeddability whose (respectively) are as follows. Tx (in Lx) is interpretable formal definitions in T2 (in L2) if U with one free variable (the "universe") there exists some L2-formula as well as some effective translation t from the non-logical constants in to for such that Lx fitting (possibly complex) L2-expressions, T2\-(r(a))u each axiom a of Tx. (Here r(a) is the result of replacing each non-logical 454 JOHAN VAN BENTHEM "U" indicates while the superscript in a by its r-counterpart; to in of all relativisation U.) For technical r(a) quantifiers subsequent that (r((p))u follows It one then that also reasons, T2|-3xU(x). requires case this In in T2 for each theorem of derivable becomes Ti. <p are can be also i.e., only Tptheorems reversed, provable implication in T2. These arose already in (through t, U) in T2, Tx has been embedded constant versus non-Euclidean the early history of Euclidean Notice, geometries. an inter establishes 'circle interior' model for example, that Klein's of into a conservative extension from hyperbolic geometry pretation a an constant for individual Euclidean (obtained by adding geometry a a an kind of here is As of relevant circle). logical result, example theorem for inner models' 1980a): "If Tx (cf. van Benthem 'compactness are that each theories such (in Li), T2 (in L2) finite part of Tx is D translated in in Lx L2 T2 (constants being identically), interpretable extension in some conservative then the whole theory Tx is interpretable these of T2".31 In general, there is a scarcity of results concerning like notions. Lately, however, logicians have been studying questions of Peano Arithmetic, the above for special cases. E.g., for extensions one has the stronger 'Orey Compactness Theorem' 1979). (cf. Lindstr?m is replete with science 'reduction structurally, empirical Speaking versus sta for thermodynamics calling logical analysis: phenomena' versus relativistic or quantum classical mechanics tistical mechanics, etc. Very often no uniform mechanics, satisfactory logical notions have been forged yet to get a good grip on these. (Cf. Pearce 1981, however.) illustrate Time. Sup this, here is a small example concerning - on the one has worked with dense time pose pattern of the always at it that is discrete rationals Now becomes clear time say. <Q, <), some deeper level, on the pattern on the integers (Z, <). The resulting structure of a could be the lexicographic product Q x Z consisting To unbounded linear sequence of copies of the integers. Notice is that of discrete that the theory of this structure time! Should one new Q is to the Q x Z - and, if that the reducible 'richer' say original to choose from. E.g., sense? There are various relations so, in which Q is clearly embeddable into QxZ. the isomorphic ally Conversely, ^ a is obvious contraction QxZ onto from mapping -homomorphism dense in its backward the 'p-morphism' clause of direction, Q, satisfying, that the is not a modal contraction 1971). (Notice logic (cf. Segerberg < -homomorphism.) seem to be would The correct view of the situation THE LOGICAL STUDY OF SCIENCE 455 the following, however: Q is reducible to a certain 'level' of Q x Z, in a to the sense of being quotient isomorphic of the latter structure some suitable relation. under that the above (Notice equivalence was & not in the model-theoretic congruence sense.) This equivalence seems of potential wide application.32 notion of reduction two Third comes the model-theoretic upon the previous perspective sets of notions. relations between 'conservative Duality syntactic were extension' and structural extension' 'Ramsey already treated in an interesting duality result will be section 2.1. As for interpretability, found in Pearce to 1980B, which also contains many useful references literature. (Even so, powerful criteria are still lacking to the technical or embeddability in given cases.) One partial disprove interpretability is result Theorem: the following of the Compactness consequence easy in T2 up to 'A finitely axiomatized is theory Tx interpretable a for contains and if model if each substructure T2 only disjunction33 is a model for TV' with L2-definable domain and predicates which to is 'local' structural Thus, equivalent 'global' definability definability A simple syntactic trick enables one to get rid of up to disjunction. the latter phrase: a 'single' one - any 'disjunctive' may be replaced by interpretation result for inter and so we have a second duality its of a V3-form 'structural' is clause pretability. Unfortunately, which does not make for easy counter-examples (unlike in the case of or Beth's Cf. G?del's Theorem Theorem). Definability Completeness (section 5) for more related results. this unsatisfactory there is an state, interesting a to be observed 'reversal of direction'. here, phenomenon namely is a common event in semantics: the experienced This reader will account like Sneed's of how to have pondered already about cases content and applicability threaten to become 'apply' a theory, where In the present case, the relevant observation is inversely proportional. to structural from Tx to T2 amounts that syntactic 'reducibility' from models of T2 to models of Ti.34 Thus, e.g., struc 'reducibility' the appendix Even in of orthomodular relations like the nonembeddability Hilbert and do Kochen space lattices into Boolean 1967) (cf. Specker algebras not correspond to of the quantum smoothly non-interpretability into that of classical mechanics! mechanical axiomatic theory in 1960 thesis that the of the structural formulation Similarly, Suppes to in be the terms: 'for any reduced following biology may physics' an to construct it model of a biological [is] possible theory isomorphic tural 456 JOHAN VAN BENTHEM within physical theory' is very hard to fathom in syntactic terms. at such an elementary these uncertainties level, we will the far more involved from formulating (and numerous) of 'theory' of 'reducibility' arising from the richer concept ones out in 2. the fruitful section Nevertheless, picking developed model Given refrain notions and Sneed the many Sneedian among (cf. Balzer proposals one tasks. will first be of the 1977/8) logicians' to the above let us return once more To conclude, temporal Q and QxZ illustration. The structural reduction relation between was that of division by the equivalence relation of 'being only finitely a corresponding What about discrete many steps apart'. syntactic seems the two syntactical At first this formulation? sight, hopeless, a versus discreteness. inconsistent: theories So, density being be a spectacular, That would but evidently revolution? scientific from A better description is the following. From rather shallow conclusion. dense linear Time the theory of unbounded (Tx, language Lx) one a on to two-sorted passes theory T2 retaining Tx at one level and linear Time at the other discrete unbounded the theory of having the two sorts. between 'bridge principles' together with some obvious concern matter the Further model-theoretic just questions might in terms of the 'lower definable when the 'upper sort' is explicitly too aristocratic for that.) it is clearly sort'. (In the present example, 4. PHILOSOPHICAL AFTERMATH not only in its background This paper has been extremely general, but in its technical considerations also (section 1), development as as to follow). the well 2, 3; (sections Nevertheless, appendix of more is not advocated here - to the production logical generalities author has taken a vow that this paper has (The present contrary. the observation Recall been his last sin against his own precepts.) made in section one wants 1.2: should be guided by specific logical research to obtain. Now, it has often been noted that both from its the internal dynamic of a theoretical subject and the demands source set of inspiration such goals. The latter may applications at the present deserves the main also emphasis juncture: witness in Suppes editorial exhortation 1973 to carry the banners Suppes' to the logical study of Space, Time and Kinema from mathematics route is not the same as a Cause, certainly; A of march tics. but, it is results THE STUDY LOGICAL 457 SCIENCE of a general Program. The remainder to some feelings about the scope of all we have to propose by way of this section will be devoted such an undertaking. 4.1 What OF Is 'Application'? of the a priori 1.1 it was argued, as against proponents out to far logic will how find the that of only way logic, inadequacy to avoid us in order to and is of science in the take go try. But, study some meant is what about by unnecessary disillusions, prior thought its frequent of logic seems appropriate. Quite generally, 'application' at the academic use in scientific talk - and its usefulness money - should not hide the fact that is a very diffuse term, in tap 'application' For the latter purpose, indeed, a separate urgent need of clarification. In section paper (or book) will be made. would be required: here only a few relevant points scientific theory means logic' in the study of a certain 'Applying a statement. un-informative but For, these neat, using logical tools: mere to notions and theorems from methods 'tools' may be anything even esoteric or notations. 'tool' would be that the Sometimes, only else everything (though real) quality called 'logical sophistication' not in is This situation different work. hard prin being supplied by are some engineers ciple from that of any formal discipline. Maybe - but as a wand kind of to to able magic problems 'apply' mathematics on the of new mathematics most require the creation applications spot. to imply that should not be misunderstood The preceding paragraph if matters. For some vague is of all that view' instance, 'logical point one merely then what in the study of science, used logical notions sauce would be the difference say, the information-theoretic with, so infant has which many 'receiver', 'channel') poisoned ('sender', labels for analyses? in the cradle by substituting (Cf. the note a to in renascent referred Aristoteleanism of 6.) It was danger of 1.2 1.1 that and the of sections contention application precisely not notions here: should of science in the stop logic philosophy or theorems should be tied up with existing, (already 'regulative' created especially for the purpose). And - an esoteric logical sophisti these requires a very down-to-earth cation able to produce training in sciences technical Thus, logic.... the question arises what it means to apply a logical theorem. 458 JOHAN VAN BENTHEM traffic lights'. Very often, these will serve as no more than 'methodical us are tells too weak to recursion that certain grammars E.g., theory are whereas others worth the generate languages, given trying. (Cf. famous in Chomsky Beth's 1957.) Similarly, ascending hierarchy us that a certain Theorem warned formulation in of constraints to Reductionism. section 3.1 paved the way Notice the familiar here: failure is the only definitive form of Popperian point exemplified Positive results may be less informative, because the success.35 reason for their success remains note unclear. 27 about the (Cf. success of 'classical' calculi of deduction.) It may be of interest to observe how critics of modern logic fail to this the influential 1976 Perelman acknowledge point. E.g., polemic the fundamental establishes of formal logic in the study inadequacy of juridical reasoning by means of examples like the following. In the were French tradition after the Code legal Napol?on judges required on the presuppositions to give their verdicts that the Law contained neither 'conflicts' nor 'gaps', while a decision could be reached in a simple methodical objective was behind this strictness motive way. (By the way, the honourable to minimize thus juridical arbitrariness; out this ideal turned untenable the citizens. Nevertheless, protecting in practice: there remained an irreducible component of interpretation as for the methodological on the part of judges.) Now, of background this ideal, Perelman remarks (correctly) to requiring that it amounts to be a consistent, the Law and decidable complete theory: a rare species even in mathematics, and so... one more example of the of logic. In the light of the preceding inadequacy paragraph, however, an application of logic has been given, demonstrating the inherent room for of certain limitations of thus Law; conceptions creating more or con methodical sophisticated conceptions, supplementary of content (rather than mere form). reason why results may be of doubtful ap positive their nature. is plicability (necessarily) general E.g., having a general from practice - which results recursion theory requires some distance in paradigms of 'constructivity' like primitive recursive functions siderations Another easily becoming Many physically non-computable. logical existence as theorems are proven by means of quite un-realistic 'constructions', was noted already in connection note with Craig's Theorem (cf. 15). in such proofs) may be largely Thus, (as embodied logical methods inspirational. THE LOGICAL OF STUDY 459 SCIENCE an optimistic search In the present author's however, experience, in will usually be a successful instances for manageable strategy cases. Reality may be harsh, but she is not mean. concrete 4.2 In Praise of Formalism will be converted devoid of 'logical sensitivity' by apologies Nobody never make one them to un should for formal methods. Thus, so to do to one in order has audiences. (Unless sympathetic make a living, of course.) As one need not make them to sympathetic audiences, a rather silent conclusion seems to follow .... a few virtues of 'the formal it may concern, Still, to whomever are will which be extolled here, praised too seldom in the approach' or defenders). I do cultural climate (whether by opponents prevailing to be a road to instant not take 'formal' or 'exact' philosophy a means for It is, if anything, and rock-bottom insights. rationality afloat on Neurath's ocean), gauging the re-fitting our raft (precariously our the underneath. of intellectual Accordingly, ignorance depth of work choice for this type ?f (instead all-embracing philosophies) rather than has always seemed to me a matter of intellectual honesty poverty. is a virtue, of course, but not a particularly exciting one. Honesty Let me, therefore, be more explicit about the merits of being formal and precise - or, better, about making things a little more formal and us one to become For clearer about cherished it forces thing, precise. intuitions - teaching us the invaluable art of being wrong. (Saying that means and the world is an 'Organic Whole' learning, nothing: risking, is a heroic and instructive saying that it is a Finite State Machine as we unravel our concepts, is their real wealth mistake.) Moreover, are Hence and hitherto unveiled, opened up. possibilities unexplored - as has been to creative is a stimulus formal precision phantasy of 'creative stressed in Piaget 1973. The fashionable opposition freedom' and 'logical armour' does not do justice to logic (nor, one fears, to creativity). I take to be of a task of logic emerges which is also where Here a to the philosophy It should provide of science. vital importance are out under tested ideal cir where ideas 'conceptual laboratory' cumstances. too, there should be a 'conceptual Very importantly sanctuary' (or 'mental asylum'?), where old discarded scientific ideas 460 JOHAN VAN BENTHEM are kept alive36 - like strains in a pollen bank for grain, which may be needed again at some future time. Hopefully, in this way, logic could a in the the worlds deal between gap great closing self-imposed help sense and science from which our culture is suffering. of common 5. TECHNICAL APPENDIX text of this paper is rather sketchy, main detailed technical In order to establish at least some arguments having been suppressed. here are some samples of more technical work logical credentials, around our general theme. The I. Invariance From a Model-Theoretic A typical model-theoretic a certain starts studying Perspective line of research might be the following. One the discrete two-dimensional domain, say a certain lattice ZxZ; endowed with structure, say the ternary comes a relation 'further than'. First structural (from) question. Which of the lattice are automorphisms with respect to this bijections structure? the group of (The answer would be, in this case, precisely rotations and These translations, reflections.) automorphisms give rise on to invariants: are relations the domain onto which n-ary mapped themselves all in the remarks group. (Cf. previous by automorphisms about Klein's These invariants form a very Program.) interesting or closed under like intersection class, being operations complement, Could, then, all invariants be characterized projection. linguistically some at least, the original relation is suitable using language? Clearly, an invariant, are predicate-logically together with all relations which definable from it. (Cf. the earlier quotation from Weyl the 1949.) Does converse is a difficult question, hold too? This for 'internal' calling to some mother results (relativized rather than definability structure), 'external' ones like Keisler's of elementary characterization classes. Here come a few pertinent results. (Cf. also Rantala 1977a.) 1. In finite structures, PROPOSITION invariance for automorphisms first-order implies definability. some invariant. Proof: Let 3) be some finite structure, and AQDn a A direct combinatorial first-order definition. argument will produce Here we just observe, however, that this also follows from pro THE position 3 to be proven (L-I-A)-saturated. STUDY LOGICAL if we below, Thus, let x be any OF 461 SCIENCE can only show of, sequence say, that (3, A) is s variables, some set of (L + A)-formulas and 2 = 2(x) finitely many involving in 3. Saturation satisfiable D is finitely in which requires parameters it were not. Then, for satisfiable. that 2 be simultaneously Suppose each s-tuple d in Ds, there exists some formula cr? in 2 such that o\j is it would such s-tuples, false at d. But, there being only finitely many in 3 : contradic follow that some finite subset of 2 is non-satisfiable on 2. Q.E.D. ting the original assumption 1 may Counter fail, however. proposition infinite structures, natural of the structure numbers In the consisting IN0Z example: IN is a in the usual of the followed ordering, copy by integers, in definable first-order without invariant for automorphisms, being terms of the ordering relation. results arise only when single structures are Often, model-theoretic let T be some first-order theory in a theories. their Thus, replaced by some A is where Indeed, a n-ary relation symbol. language L-l-A, In duality between invariance and definability 2. A2 is an L-automorphic PROPOSITION definable for T if and only if A is explicitly if to left, clearly, Proof: From right ThVx(Ax~?i((x)) is now forthcoming: invariant in each model in T up to disjunction. v ... v Vx(Ax^?m(x)) where 8\,..., then, in each model for T, A will be 8m are L-formulas some it will be invariant in the above first-order defined by 6?.Hence, sense. Theorem may be applied Svenonius' for the converse, Secondly, Let 3 be an L-structure theorem and Keisler 1973; 5.3.3). (Chang for T. with (L + A)-isomorphic (3, A), (3), A') to models expansions = + A' A of then being A, ((L A)-automorphisms By the invariance this implies theorem By the above-mentioned L-automorphisms). explicit definability up to disjunction. Q.E.D. like the results to 'internal' problems such changed to be shown have one, it would that, e.g., any original geometrical = Th((Z x Z, A)) in given invariant A in Z x Z gives rise to a theory T In order to apply 462 JOHAN BENTHEM VAN the interpretation of A is an invariant. We do each of whose models not solve this problem here. invariance inside which there does exist a kind of structure Finally, - an observation further ado which without implies explicit definability will below: be used invariant in an (L + A) 3. If A is an L-automorphic PROPOSITION in 3. saturated structure 3, then it is first-order L-definable an relation. A be Consider this Let any n-ary n-ary Proof: a of objects of a in 3 cannot be the L-type in A. Now, sequence of the satisfied outside of A. Otherwise, (a consequence L-homogeneity not of 3 an would saturation L-automorphism provide property) given + the of the Put onto A itself. (L A)-set consisting differently, mapping L-type of a together with the formula "~1Ax is not satisfiable in 3. Hence - 3 some finite subset of it will not be satisfiable (L + A) being saturated. Say ...,ts(x), {ti(x), is not (1) where in 3 satisfiable ~1Ax} ;where ... & ti & ts. since each a in A will Moreover, (L + A)-set ,ts are true of a. Equivalently, is true in 3; Vx(T?(x)-?Ax) Ta=def tx, ... satisfy some such L-formula r?, the {-\Ta\aG A}U{Ax} is not finitely many t = either. ax,..., sequences -> t(x)) Vx(Ax (2) where in 3 satisfiable def t?, In combination v . .. v with Vx(t(x)<-?Ax) A has been L-defined. II. The Categorial Mode there Hence, again by saturation, at in A such that is true in 3 exist ; r?t. (1), it then follows that is true in 3: Q.E.D. of Thinking in connection mentioned The categorial perspective a to characterization constraints may be used provide with Sneed's of first-order THE LOGICAL STUDY OF SCIENCE 463 inside a structure. To see this, let 3 be some L-structure, definability A some n-ary invariant in it. Now, 3 belongs to the category CAT all of with (L) L-structures, (not neces L-isomorphic embeddings its for and the construction sarily surjective) morphisms, ultraproduct as a characteristic I do not know of any operation. (Unfortunately, categorial definition behaviour.) of ultraproducts in terms of their 'morphological' 4. A is first-order PROPOSITION in 3 if and only if L-definable is simultaneously a to of CAT expandable sub-category such that all and of (3, A) containing morphisms operations same + the in remain CAT(L) CAT(L A). CAT(L) (L + A) = is first-order L-definable, say A <p?, then the obvious of each L-structure 3' to (3', ??') will do. For, L-isomor expansion remain (L + A)-isomorphisms phisms preser by their fundamental vation and Los' Theorem the same for property, guarantees Proof: If A ultraproducts. that CAT(L) has been expanded as indicated. suppose Conversely, Let U be any countably over some ultrafilter incomplete a-good = = a index set I; where the |i| max(^V0, |D|) and |I|+. Consider - which 11^ 3 in CAT(L + A) expanded ultrapower equals Uv(3, A), Notice that this ultraproduct is (L + A)-saturated; by assumption. by AUu? is invariant in 1973, theorem 6.1.8. Moreover, Chang and Keisler the ultraproduct: of this structure being automa L-automorphisms con The promised (L -I-A)-automorphisms, tically by assumption. clusion then follows at once from proposition 3. Q.E.D. this line of thought will yield more Hopefully, tion in the first draft of this paper was verified III. Varieties The results. (This specula in Pearce 1980B.) of Reduction of 'reduction' has been shown to admit of various logical in section 3.2. First, a syntactic result of that section will explications be proven here. Let Tx (T2) be a first-order t theory in language Lx (L2). Translations from Lx to L2 will assign, possibly to complex, Lr L2-predicates - those in Lx n L2 being mapped primitives (Thus, one identically. the idea that a partial correspondence is given in ad implements notion 464 JOHAN VAN BENTHEM of Tx in T2 means that some translation vanee.) 'Interpretation' r(Tx) be in to a unary L2-predicate relativized may proven T2, possibly some sub-domain. a Since Tx may be infinitely axiomatized, denoting theorem' would be was here. in fact, ?s 'compactness But, helpful stated in note 31, no such result holds in general. (Cf. Lindstr?m 1979, however.) What we do have is the following: PROPOSITION 5. If each finite subset of Tx is interpretable in T2, then the whole in some conservative exten theory Tx is interpretable sion of T2. in either Proof: Take a new unary predicate constant U not occurring consider T2 U T V (where all quantifiers in Tx Li or L2. Now, occurring are relativized to U). Tx is trivially interpretable in this new theory. Thus, to show that the latter is a conservative it only matters extension of T2. To see this, let <pbe any L2-sentence such that T2U T^Vy. It follows that, for ax,..., many finitely ak E T2 U Tx, {aV, , ?u }h<p. of these formulas, taking a to be the conjunction nor since neither U the of a Then, (Lx L2)-vocabulary T2\-au^xp. occur in T2, it follows for the universal second-order closure u -> that T2hV(a u-?(p). V(a <p) taken with respect to U and Lx-L2on Tx and T2, there exists some translation Next, by the assumption t of Lx into L2, as well as some unary L2-predicate ? such that From the previous that follows it then T2\-(r(a))?. paragraph, Equivalently, -> T2\-(j(a)? <p, and hence T2h<p. Q.E.D. In section 3.2 the following model-theoretic theories Tx: stated, for finitely axiomatized interpretation result was 6. Tx is interpretable in T2 up to disjunction if and an L2-definable for T2 contains sub-domain which is a model for Tx. together with L2-definable predicates to From left if Tx is not Proof: right, this is obvious. Conversely, as indicated, then T2 united with the set of all negations interpretable of L2-definable translations of Tx (relativized to all possible L2 definable is a finitely set of formulas. satisfiable domains) By com In other words, T2 satisfiable. then, it will be simultaneously pactness, a model without submodel for Tx. Q.E.D. possesses any definable PROPOSITION if each only Next, the model 'syntactic trick' mentioned in section 3.2 works as exem THE LOGICAL by the following plified special a 3x1/0 T2Htx(Tx)u> the first disjunct Call U(x) Notice = OF STUDY case. 465 SCIENCE Suppose that v (T2(TX)V> a 3x17z). a, and set dei(Ux(x) a a) v (U2(x) a la). for arbitrary Then, T2h3xU(x). = def a a) v (t2(P)(x) (ti(P)(x) T(P)(X) that P, Li-predicates set a la). that T2\-r(Tx)u. A simple argument about models of T2 establishes 'cross-structural' conditions Further refinements, using categorial as in II above, yield a duality result for interpretability 'tout court'. Instead of embarking upon this course, however, we prefer to review the whole topic from a different angle, namely that of combination of theories. One has theories Tx (in Lx) and T2 (in L2), got two first-order that both Lx n L2. Let us assume possibly sharing some vocabulary are intended to describe form the same kind of objects. The weakest seem to be: of combination would (1) the deductive union of Tx and T2 is consistent in Lx + L2. Theorem Joint Consistency tells us that this occurs just n \ when Tx Lx L2 and T2\ LXC\L2 contain no mutually contradictory that Tx U T2 need not be a conservative theorems. Notice extension of either Tx or T2: both theories may have learnt in the process. Some reflection upon Robinson's proof shows that, e.g., Tx U T2 will be a of T2 if and only if T? ?Lx n L2 is contained in conservative extension Robinson's T2\ LXD L2. from left to right is obvious. Conversely, for any (For, the direction if then T2U{"l<p} has got a model 3. By the L2-sentence <p, Ttf?, of 3 united with Tx will be then, the (Lx n L2)-theory assumption, - and hence it has a model 3'. satisfiable finitely Starting from 3 and one builds unions may be 3', chains, whose elementary alternating folded into an (Lx + L2)-structure Tx U T2 while together verifying and Keisler <p.Cf. Chang 1973.) falsifying Next, addition, (2) connections stronger such as: Tx U T2 is contained between in some Tx and T2 may definitional arise extension upon of T2. 466 JOHAN VAN BENTHEM ? of (Lx L2)-vocabulary, I.e., given certain L2-definitions Tx becomes from T2. This is the earlier definition derivable of 'interpretation' of possible to sub-domains, relativization leaving out the complication that is. In this case, automatically extension Tx U T2 is a conservative of T2 (cf. the proof of proposition 5 above), of though not necessarily that Tx. The latter need not even happen on the stronger connection occurs: 'definitional reducibility' Tx U T2 coincides (3) Contrary Extension theories with some definitional extension of T2. to our experience with the mis-match between Conservative and Ramsey the latter connection between Extension, admits of an elegant structural characterization: 7. Tx is definitionally PROPOSITION to T2 if and only if reducible each model for T2 admits of exactly one expansion to a model for Tx. to T2 - say, that Tx is definitionally reducible Proof: First, suppose 8 as above. Now, Tx U T2 is axiomatizable by T2 plus L2-definitions consider any model 3 for T2. By interpreting the (Lx L2)-vocabulary an + to 3+ which satisfies through 8, 3 is expanded (Lx L2)-structure for Tx: (T2 plus 8 and hence) Tx U T2. A fortiori, then, 2>+ is a model as for 'at most'. and the 'at least' has been taken care of. Next, If 3+, 3+f are any two expansions of 3 to models for Tx, then (Tx U T2, and the respective of hence) 8 will hold in both. Therefore, interpretations the - (Li L2)-vocabulary must coincide: i.e., 3+= 3*'. that the above structural condition holds. Its suppose Conversely, 'at most' side means that the implicit definability clause of Beth's Theorem is satisfied - and hence explicit L2-definitions 8 of the (Lx are U in derivable It follows that Tx T2. Tx U T2 may be L2)-vocabulary as of its the U ? union (re-)axiomatized L2-part (Tx T2) L2 with 8. For, both of these are derivable from Tx U T2. Moreover, vice clearly, if U U where versa, Tx L2) '?(Li)' T2f-<p(L,, L2), then T, T2\-?(8(Lx), refers to suitable L2-replacements the L2-formula through 8-, whence to (Tx U T2) \ L2. In conjunction with 8, then, the ?(8(Lx), L2) belongs latter theory reproduces the original formula <p(Lx,L2). to in order of Tx to T2, it now extension Thus, prove definitional to suffices show that (Tx U T2) ?L2 coincides with T2. Put differently, it remains to be shown that Tx U T2 is a conservative extension of T2. But this follows from the 'at least' side of the above structural condition. such that T2?<p, then T2 U {l<p} For, if <p is any L2-formula THE LOGICAL STUDY OF SCIENCE a model: can be expanded have which U Ti U T2 U {l<p} whence Tx T2V(p. Q.E.D. to will were These notions pleasing the theories - but, rather 467 some un-realistic. to be added will refer for model in actual For, to different (cf. Tarski kinds practice, usually of objects. Consider, e.g., Elementary 1959) Geometry with 'betweenness' and (quaternary) (ternary) 'equidis primitives = tance' for Tx, and the algebraic theory of the reals (IR (R, 0,1, +, )) for T2. As was noted already in section 3.2, a two-sorted combination the main is more appropriate sortal connections becoming possible issue. E.g., in this case, equidistance induces an equivalence relation a between of for notion of points allowing 'length' pairs bridge certain bridge principles for 'addition' satisfying (etc.). The relation between such more complex notions of reduction and the syntactic theorems will not be investigated usual representation here. Rijksuniversiteit Groningen NOTES * 1 I would like to thank David Pearce Rantala and Veikko comments. for their helpful to this rule is V. Rantala's and Non-Standard paper 'Correspondence exception As for 366-378. A Case in Acta Fennica 30 (1979), Study', Philosophica a variety of publications there is, of course, by Robinson 'ordinary' applications, An Models: himself and co-workers. 2 to justify their have used similar excuses defeatist (or elitist) Regrettably, logicians exclusive concentration upon pure mathematics. 3 in text book version of logic. E.g., What is usually is some static elementary criticized the otherwise work London, very (Batsford, of Reasoning interesting Psychology to show that and P.N. Johnson-Laird 1972), P.C. Wason example give the following two sentences 'if prices 'actual transcends formal The any modelling.' reasoning increase the firm goes bankrupt' and 'prices increase, only if the firm goes bankrupt' are stated to be 'logically the same 'i->B\ in equivalent', having But, logical form feel a difference, due to additional and so .... people practice, temporal or causal content, seem unaware The authors of the fact that observations like these are precisely at the basis of modern B' and 'I ? logical semantics. E.g., in tense past B\ respectively. And that without red carpets. expecting such philosophers to regard of science set theory, many: game theory, topology, 5 This is Popper's main insight applied logic, There logic the two logical seems as just forms would to be a growing one auxiliary system theory, to the philosophy be tendency discipline etc. of science 'I -? future itself. among among JOHAN VAN 468 BENTHEM 6 one sometimes In this connection, is in the air, fears that a New Aristoteleanism with understanding them. confusing labeling phenomena 7 It almost that her interest will also cover the most diverse goes without saying of reasoning: of information, extraction refu rationalization, purposes justification, or link with Hempelian In particular, the latter two form an obvious tation, explanation. Popperian 8 Notice of science. philosophy on this view, there is a continuous to that of scientific theories. language how, (natural) 9 Yet more have entities complex (Lakatos). programs' even But, been then, proposed, the level of spectrum from the logical like 'theory nets' (Sneed) or a convenient theories remains study of 'research starting point. 10 Speaking this view may have been that only doctrine historically, inspired by Kant's to (possible) which is related is free from contradiction, whereas knowledge experience runs the constant risk of antinomies. 'pure reason' 11 one argue to the opposite Or should effect: non-constructive for logic is harmless one has once non -constructive to be more domains: careful constructively given domains are considered (the intuitionist position)? As so often, metaphors either point way. 12 There If T\ is axiomatized be a problem here. of universal might by means to use universal statements and one is only allowed in proofs, could not it statements, be that certain universal theorems become their predicate-logical because unprovable: intermediate the steps of higher requires quantifier complexity? Fortunately, proof answer as may be seen, for instance, is negative: semantic tableaus. by using 13 case is, e.g., Goldbach's to the effect An interesting that each border-line Conjecture even natural number do not know any than two is the sum of two primes. We greater n equals such that each even functions number only primes explicit /, g yielding there are (trivially) recursive functions But, /(n) + g(n). /, g such that, if n can be as a sum written make of two Goldbach's Conjecture recursive Skolem primitive 14 A relevant such Or a sum. Does should one this require In the real numbers, it is is the following. (A.S. Troelstra) + x = 0 has solutions that the equation for each value of provable y3-3y x. But, there is no continuous function such that /: IR-?IR example constructively the parameter Vx(/x)3-3(/x) (A 'cusp catastrophe' 15 here By the way, can at all, then f(n) + g(n) will be primes an acceptable 'concrete' statement? functions? + x=0. occurs.) is an instructive to the claim that philosophers counter-example a knowledge of just logical results. For, as it happens, Craig's method a 'tricky' set of axioms for the observational without any produces sub-theory, use. Thus, a it is good to know that a 'reduction' in principle, exists afthough get by with of proof practical lot of work remains 16 This contribution to be done to produce and interesting convincing examples. in the philosophy exhausts of by no means Suppes' important work To mention his papers on measurement science. just one other example, theory have a terms. clear interest for the semantics of 'observational' 17 is negative, Cf. Przet?cki 1969. The answer maybe surprisingly. 18 Notice that, in this case, T\-<p if and only if T0h<p, for each L0-sentence <p (T0 being THE Hence, complete). be and (3) would LOGICAL STUDY also the example a syntactic to at least, it would be a decidable matter, This eliminable. is Ramsey conjecture 1978. in van Benthem however, 20 As for C2, it would have to be decomposed U y) Vxy m(x = 469 SCIENCE of (1) that the conjunction the conjecture the structural condition that MOD(To) refutes counterpart \ L0. equals MOD(T) 19 was it that, hoped Originally, T, if its theoretical vocabulary OF given was a theory refuted, into a law 4- m(y) m(x) (with 'U! denoting physical join) plus the identity constraint 21 from a quite of this type of requirement For an example Kreowski and P. Padawitz: H.-J. 'Stepwise Specification Cl. different and area, cf. H. Ehrig, of Implementation Lecture Notes 5 (1978), in Proceedings in Data ICALP Springer Types', are algebraic the morphisms Science 62, 205-226. Here homomorphisms. Computer 22 a common and Di will have in both D\ and Di. Now, Let p occur D\ empirical at the theoretical D containing this relation must be preserved p. Since superstructure Abstract level, D2. 23 p will receive would One also of semantics the same theoretical function like to see the Kripke structures of modern 'empirical structures' theories. scientific (The values in the expansions tense of logic recent of D, D\ applied work and in the by R. are of this kind.) discussions parallel W?jcicki 24 There are the status of natural laws: linguistic concerning or structural Cf. Suppe 1976. relations'? 'sequencing expressions 25 on Sneed's is intelligible set-theoretic constraints the 'subset condition' (and E.g., a universal even formulation. has ('had'?) correct) explicit only if the constraint occur in at least one like 'any two particles 'Universal-existential' constraints together situation' empirical Another finitely case many undefinability the ultimate for us the condition. satisfy with in equilibrium in Sneed 1971 of a balance is the example on will take the predicate-logical it. Some structuralists objects to be of this simple of finiteness situation) type of empirical (and hence verdict is force do not in point one do such an analysis would But, what case. E.g., a 'finite' in this particular on the a recursion would involve example formalisation. upon first-order to think about the cash value of the above formalisation simple first-order in a natural number of objects, corresponding, on the balance. weights of fashion, to progressive addition of that the incorporation into in Pearce of linguistic 1980B it is argued aspects Finally, it 'still more and comprehensive'. would make the Sneedian frame-work expressive 26 one to classical Cf. Vaught It need not even commit semantics. logic on the Tarski true in all countable structures with the class of formulas 1960, where classically to be non-recursively is shown axiomatizable. decidable predicates 27 the preferred role of classical strength logic may be due largely to its excessive E.g., a way from premises to conclusions that ignore the possibility blasting making people for the same theory might have been organized allowing logical means, using weaker subtler metic, as a distinctions which enables (mathematical) to be made. one One to derive theorem. A traditional the Law recent example of Excluded exciting example Arith Heyting statements for atomic is intuitionist Middle is the proof in M. Dunn: JOHAN VAN 470 Mathematics' 'Quantum Arithmetic axiomatized laws as mathematical 28 like to see at least One would theme. 29 Although of (Department with quantum theorems. one BENTHEM Philosophy, logic produces text book ... would you board a rocket in the theory of differential inconsistency 30 of Chang The organisational principle to Mars, Bloomington, all additional of model theory immediately 1980) that Peano classical logical organized after around the publication this of an equations? and Keisler 1973, viz. that of various methods seems less suitable from the present construction, (cf. note 28). perspective a feeling awareness for the ontological of such Still, principles' gives 'proliferation universe. of the structural abundance 31 is necessary: without The latter addition be given. E.g., it, counter-examples may = take for T2 the first-order theory of (N, O, S) and for T, :T2 plus {c# s'0, i 0, 1,2,...}. is not as trivial as it seems at first sight.) calculation (The required 32 with To get better acquainted these various notice that Z is isomorphically notions, in Q, as well as being to one of its quotients all 'pins' embeddable (contract isomorphic of model (n, n + 1] for reducibility of Q- and converses fail. As for n). Both integer to be introduced Z is not isomorphic below, the converse the model-theoretic to any definable notion of substructure as well. fails as translations ti, ..., Tk of the ,Uk as well U\,... a 3xUi) into such that L, L2 non-logical T2(-(ti(Ti)Ui v ... v (Tk(Tx)Uk a 3xUk). 34 One might dualities embed try to restore symmetry by formulating concerning rather than interpretability. 'for each Li-sentence if <p,Tih<p if and only dability E.g., on MOD(Ti), to: [some symmetric is equivalent condition t, U]\ MOD(T2), T2Kt(<p))1/, 35 once remarked to his political As the famous Dutch Calvinist prime minister Colijn 'In negation lies our strength'. opponents: 36 A prime example is the Leibnizian view of Space and Time; cf. Suppes 'relational' 1973. exist there 33I.e., L2-formulas constants in REFERENCES Arbib, M. Ashby, Balzer, and Padulo, L.: 1974, System Theory, to Cybernetics, R.: 1976, Introduction W. and Sneed, J. D.: 1911 IS, Generalized Studia Barwise, Benthem, Benthem, siteit, Benthem, W. Saunders, Philadelphia. London. Methuen, Net Structures of Empirical Theories', and 37, 167-177. 36, 195-211, L?gica Amsterdam. J. (ed.): 1977, Handbook North-Holland, of Mathematical Logic, J. F. A. K. van: 1978, 'Ramsey Eliminability', Studia L?gica 37, 321-336. J. F. A. K. van: 1980, The Logic of Time', Filosofisch Instituut, Rijksuniver Groningen. J. F. A. K. Instituut, Benthem van: Rijksuniversiteit, L. S. Jutting, 1980a, Modeltheorie Groningen. van: 1979, voor Wetenschapsfilosofen', Filosofisch in the Landau's 'Checking 'Grundlagen' Mathematical Centre Tracts 83, Amsterdam. System', an Up-to-date E. W.: 'Towards of the Natural Sciences', Beth, 1949, Philosophy Methodos 1, 178-185. AUTOMATH THE Beth, E. W. 1966, Mathematical J.: Piaget, OF SCIENCE 471 and Epistemology D. Psychology, Dordrecht. Reidel, Chang, and STUDY LOGICAL C. C. and Keisler, H. R. L.: and Vaught, Craig, W. Journal of Symbolic Logic N.: 1957, Syntactic 1973, Model Theory, Amsterdam. North-Holland, Predicates', using Additional Den Haag. Structures, Mouton, Truth and Logic', 1975, 'Vagueness, 30, 265-300. Synthese - The J. (ed): 1978, K. Ajdukiewicz World-Perspective Scientific Dordrecht. Reidel, Chomsky, Fine, K.: Giedymin, Essays', Giles, R.: 'Formal 1979, L., Henkin, J.: 1958,'Finite Axiomatizability 23, 289-308. P. Suppes, and Physics', Languages A. and Tarski, (eds.): and Other in Hooker 1959, (ed.) 1979, pp. 19-87. 'The Axiomatic Method: to Geometry and Physics, Amsterdam. North-Holland, Reference D.: der Geometrie, 1899, Grundlagen Teubner, Hilbert, Leipzig. U.: Its J. and Remes, Hintikka, 1974, The Method of Analysis, D. Reidel, Dordrecht. and Geometrical Origins, Significance with Special C. A. Hooker, Dordrecht. 1979, Physical (ed.): Theory as L?gico-Operational Philosophical D. Reidel, Structure, in Quantum S. and Specker, E.: of Hidden Variables 1967, 'The Problem Journal and Mechanics 17, 59-67. Mechanics', of Mathematics I. M.: and Refutations, Lakatos, 1976, Proofs Press, Cambridge. Cambridge University on Interpretability', P.: of Philosophy, Lindstr?m, 1979, 'Some Results Department of G?teborg. University Kochen, P. Lorenzen, and Lorenz, Darmstadt. gesellschaft, 1978, Dialogische Wissenschaftliche Logik, Buch Yale University 1974, Formal Press, New Haven. Philosophy, Brace & World, New York. 1961, The Structure Harcourt, of Science, D.: of the Structuralist of Reduction' 1980, Properties Concept 'Logical R.: Montague, E.: Nagel, Pearce, Erkenntnis to as 1980A). (referred (forthcoming) 'A Mathematical Characterization D.: Pearce, Definitional Pearce, 380. K.: of Interpretation, and Reduction to as 1980B). to appear in Studia L?gica (referred Equivalence', between 4, 363 1981, 'Some Relations Systems', Epistemolog?a Empirical 1980, D.: Perelman, C: J.: Piaget, London. Przet?cki, London. 1976, Logique 1973, M.: Juridique, Jurisprudence of Genetic Epistemology, Principles 1969, The Logic of Empirical G?n?rale Theories, Paris. Dalloz, Routledge Routledge and Kegan Paul, and Kegan Paul, a Model-theoretic versus to the Logical M.: 1974, 'A Set-theoretic Approach Przet?cki, Structure of Physical Studia L?gica Theories', 33, 91-105. in the K. and W?jcicki, R. (eds.): M., Szaniowski, 1976, Formal Methods Przet?cki, D. Reidel, Dordrecht. Sciences, Methodology of Empirical W. V. O.: 1951, 'Two Dogmas in Quine of Empiricism', 1963, pp 20-46. a W. V. O.: From Point and View, 1963, Row, New York. Quine, Logical Harper of F. P.: in Philosophy, Mathematics and 1978, Foundations: Ramsey, Essays Logic, and Kegan Economics, Paul, London. Routledge Quine, V.: Rantala, 213-222. 1977, 'Definability in Empirical Sciences', Acta Philosophica Fennica 29, JOHAN VAN 472 V.: Rantala, 1977, Aspects BENTHEM North-Holland, of Definability, Amsterdam (referred to as 1977a). Basis of the Logical the Structuralist Philosophy 15, 33-53. H.: and Time, Dover, New York. 1956, The Philosophy Reichenbach, of Space D. Reidel, E. (ed.): 1979, Game-theoretical Dordrecht. Saarinen, Semantics, K.: in Classical Modal Studier Filosofiska 1971, An Essay Logic, Segerberg, V.: Rantala, 1980, Erkenntnis Science', of 'On 13, Uppsala. in Barwise 'The Incompleteness Theorems', 1977, pp. 821-865. and Row, New York. 1977, The Tao Is Silent, Harper D. Reidel, Structure Dor J. D.: 1971, The Logical Sneed, Physics, of Mathematical drecht (2nd ed. rev. 1979). Berlin. View of Theories, W.: 1979, The Structuralist Springer, Stegm?ller, C: Smorynski, 1977, R. M.: Smullyan, F.: 'Theoretical 1976, Suppe, 1976, pp 247-267. P.: 1957, Introduction Suppes, Laws', in Przet?cki, Szaniowski, and W?jcicki (eds.) to Logic, Van Nostrand, New York. P.: 1960, 'A Comparison of the Meaning and Uses in Mathematics of Models Suppes, and the Empirical Sciences', 12, 287-301. Synthese P. (ed.): 1973, Space, Time and Geometry, D. Reidel, Dordrecht. Suppes, and Tarski in Henkin, Tarski, A.: 1959, 'Wliat is Elementary (eds.) Geometry?', Suppes 1959, pp. 16-29. Toulmin, Vaught, Logic Wessel, 1958, The Uses of Argument, Press, Cambridge Cambridge. University True in All Constructive Journal Models', 1960, 'Sentences of Symbolic 25, 39-53. Berlin H. (ed.): 1977, Logik und empirische Akademie Verlag, Wissenschaften, S. E.: R. L.: (DDR). H.: Weyl, York. Zeeman, Physics 1963, E. C: Philosophy 1964, 5, 490-493. of Mathematics 'Causality Implies and Natural the Lorentz Group', Science, Journal Atheneum, New of Mathematical