REVIEWS
Arne Naess, Communication and Argument. Elements of Applied Semantics, Universitetsforlaget, Oslo; Allen &Unwin, London, 1966, 136 pp., 25 N.Kr. Among philosophers professor Arne Naess is probably the leading representative of the view that if you want to know the facts about language and how it is used, look andsee; do not just sit back and excogitate what 'must' be the case. Thus (if we may be pardoned for overworking some phrases of Wittgenstein), do not say: 'The meaning of a complex expression must be a function of the meanings of its parts'; but look and see whether in fact it always is. D o not say: 'You cannot say such-and-such'; look and see whether we do not on occasion (successfully) say just that. D o not say: ' A n y fluent speaker of English must have internalized such-and-such rules'; decide what you want to mean by 'internalizing' rules, and then look and see whether the phenomena will bear the description you propose. Professor Naess's approach to the semantics of natural languages has been set forth at length in his treatise Interpretation and Preciseness. The present small book, which would be suitable as a text for a course in argumentation, concerns itself with the application o f his conceptual scheme to the practical task of improving the clarity and precision of cognitive discourse. Roughly the first third is devoted to defining and explicating the basic concepts to be employed, and in the remainder the author analyzes with their help various ways in which communication can break down (and be set right again). The three most fundamental notions - those of interpretation, reasonable interpretation, and precization - are defined in effect as follows. A declarative sentence U is an interpretation of a declarative sentence T i f and only if there is a person P and a situation S such that U can express the same meaning as T for P in S. Uis a reasonable interpretation of T (when T occurs in S) if and only if, when T occurs in S it usually means the same as U. Uis a precization of T if and only if all reasonable interpretations of U are reasonable interpretations of T, and there is at least one reasonable interpretation of T t h a t is not a reasonable interpretation of U. The presence of the word 'usually' in the second of these definitions makes it clear that the primitive notion is not ' U expresses the same meaning as T for person P in situation S', as the author sometimes appears to suggest, but the more complicated 'Uexpresses the same meaning as T for person P off occasion O of situation S'. (It is explained that 'situation' means "situation-type, i.e., a determinate pattern of events or state of affairs, and not necessarily restricted to one occasion".) All three definitions would be improved by the addition of
344
Synthese 17 (1967) 344-345. © D. Reidel Publishing Co., Dordrecht-Holland
REVIEWS enough quantifiers to make clear whether the reference is to all or some persons, situations, and occurrences of situations. Also, the combination of quantifiers and modality in the first definition is bothersome; on one understanding of this definition it would be plausible to say that every pair of declarative sentences T and U are interpretations of one another (since they couM mean the same for somebody on some occasion of some situation). Professor Naess is well aware that in many if not most circumstances of everyday life "strict accuracy is not called for". "When we are told", he says, "to make our expressions precise, this ... does not mean ... that we should precizate every expression up to the hilt, but only that our expressions should be sufficiently precise for the purposes at hand; which, of course, is quite another matter." He concedes that "the openness in the rules [for a natural language] allows words and sentences a perpetual variability in meaning, something which permits them to be vehicles for advances in understanding and sensibility", but he reminds us that this same flexibility allows language to be used as a cloak for ignorance and sloppy thinking. What is important, as professor Naess sees it, is that we be able to make our discourse more precise on those occasions when it appears likely that serious misunderstanding will otherwise exist. To this end the author includes his versions of a number of those admonitions with which logic books of the 'How to Think Straight' genre are filled: e.g., do not beg the question, do not evade the issue, do not define in vicious circles, do not explain obscurum per obscurius, be careful of catchphrases, metaphors, and slogans. In particular, to promote effective discussion he recommends six rules: keep to the point, avoid tendentious renderings of other people's views, avoid ambiguity that creates a real risk of misunderstanding, avoid ascribing to an opponent alleged consequences of views he holds, set forth the relevant facts in an objective way, endeavor to present the whole issue in a neutral context. N o doubt all of this is good advice. A t least, it is several cuts above the usual recommendations that we avoid such 'fallacies' as 'afftrming the consequent', 'using four terms', or 'arguing a dicto simpliciter ad dictum secundum quM'. But professor Naess has only himself to blame if those of us who have been influenced by his resolutely empirical approach to language begin to wonder whether, as a matter o f fact, any recognizable improvement in the quality of discussion and argumentation is ever effected by such advice. Maybe it is, and maybe it is not. The very task of formulating this hypothesis in such a way as to make it testable might lead to some desirable clarifications. But these doubts concern the entire genre. Of its kind, professor Naess's book is undoubtedly one of the best. It is short and sweet, and Alastair I-Iannay's translation reads very smoothly - so smoothly indeed that one would never guess that the book had not been written originally in English. The Oslo University Press, too, deserves congratulations for the attractive format and the beautifully clear type. University o f California, Berkeley
BENSON MArEs 345
REVIEWS Kazimierz Ajdukiewicz, Logika pragmatyczna [Pragmatical Logic], with a preface by Klemens Szaniawski, Pafistwowe Wydawnictwo Naukowe, Warsaw 1965, 409 pp.
Pragmatical Logic was intended as an introduction to scientific methodology and addressed mainly to students of philosophy. Although the book includes a competent discussion of many 'technical' questions, still it is not limited to the problems which can be fully solved on the basis of formal logic or mathematics alone. Ajdukiewicz has emphasized that if we wish to understand what a scientist is doing when he is acting as a scientist and, in particular, if we wish to establish criteria for the evaluation of the procedures he applies, we must, beside observing his actual practice and the results he has obtained, take into account the goals which he tries to reach in his scientific work, even if he is himself unaware of them. This gives rise to some specific psychological and sociological (pragmatical) problems, which should also be taken into account in methodology. Pragmatical Logic was designed to satisfy this requirement. Let me say at the outset that, as no one would have doubted, professor Ajdukiewicz's book is rewarding reading for everyone who is interested in theory of science, not only for the beginner but also for the specialist. He offers clear and penetrating analyses of a wide range of problems, and he throws fresh light on many of them. The book is a long one and deals with many topics, so a useful summary is impossible. Instead I shall briefly list the topics covered, select a few of them for a brief exposition and some comments, and finish with certain general remarks. Before a presentation of the contents of Pragmatical Logic, let me point out that it is an unfinished book. Death interrupted the author's work before he could make his final changes and corrections, or write the last chapter. "It cannot be said for sure about any part of this book that it got its definitive form. This is clear for everyone who was familiar with the style of professor Ajdukiewicz's work. He was always seeking better solutions, and was never satisfied with the result obtained", we read in the preface. The book is divided into three parts. The first one, 'Words, Thoughts, and Objects', is devoted mainly to some preliminary questions. It deals with the following topics: expressions and their meaning, declarative sentences and their parts, the denotation and the content of names, ambiguity of expressions and defects in meanings, definitions, and interrogative sentences. There is no doubt that the chapter on definitions is the most interesting one. Once again the author advocates there the point of view that there exist two concepts of definition which have entirely different contents, while their denotations overlap (cf. e.g.K. Ajdukiewicz, 'Three Concepts of Definition', Logique et Analyse 1, 3/4, Bruxelles, 1958). They are the concepts of: (i) real definition, which is an unambiguous characterization of an object of any kind (e.g. an individual object, set, relation), (ii) nominal definition, which is a sentence allowing us to eliminate an expression from every context in which this expres-
346
Synthese 17 (1967) 346--349. © D. Reidel Publishing Co., Dordreeht-I-Iolland
REVIEWS
sion is included. The nominal definitions are still subdivided into synthetic (i.e. stipulative) and analytic ones. The nominal definitions of such expressions as do not denote any objects (e.g. definitions of connectives 'or', 'and', 'if .., then', and the others) are obviously not real definitions. But the question arises whether there can exist real definitions which are not nominal. This question is not discussed in the book. Let us note, however, that a negative answer would allow us to regard the real definitions as nominal ones of a particular kind, in spite of Ajdukiewicz's contention that there is no common concept of definition. Still, it seems that a real definition of an object is always a nominal one, namely the nominal definition of the denomination of this object. The second part of Pragmatical Logic, 'On Inference', begins with a short, six-page chapter devoted to the concept of logical consequence. In the following sections, Chapters II, III, IV, Ajdukiewicz conducts an inquiry into some problems of the theory of inference. After defining an inference (reasoning) as a mental operation which consists in accepting on the basis of premisses, accepted with a certain degree of certainty, a conclusion whose degree of certainty is higher than it had been previously, Ajdukiewicz distinguishes two kinds of inferences; subjectively reliable and subjectively unreliable ones. A reasoning of the person X is said to be subjectively reliable if and only if X accepts the conclusion of his reasoning (takes it to be true) with the same degree of certainty which he ascribes to the prernisses. If the conclusion is accepted with a degree of certainty lower than that with which the premisses were accepted, the argument is regarded as subjectively unreliable. The main concept of this part of the book is that of 'conclusiveness' of someone's inference. Ajdukiewicz defines it separately for subjectively reliable and subjectively unrdiable inference. A subjectively reliable reasoning is conclusive if it simply is a deductive inference (i.e. leads to the conclusion which is a logical consequence of the premisses). The definition of conclusiveness of a subjectively unreliable inference is more complicated. A subjectively unreliable inference of a person Xis conclusive if and only if X accepts the conclusion of his reasoning with a degree of certainty which does not exceed the logical probability of the conclusion with respect to the premisses. It is a pity that Ajdukiewicz's book does not include a more detailed examination of both the concept of logical probability and the concept of acceptance of a sentence with a certain degree of certainty. A sketchy information about the first and a few short remarks about the latter constitute a hardly sufficient basis for the full understanding of his thought. I am convinced that although the reader will be able to grasp the outline of the conception Ajdukiewicz presents, nevertheless there remains room for doubts and misunderstandings. Although it is not stated explicitlyin the book, the task of Ajdukiewicz's considerations of conclusiveness is to establish a criterion of rationality of someone's lending weaker or stronger credence to the conclusion of a reasoning (cf. in this matter an earlier work by Ajdukiewicz: 'The Problem of Rational Char347
REVIEWS
acter of Unreliable Modes of Reasoning', in Polish, Studia Filozoficzne, 1958). This remark leads us to the following doubts: (i) Since we cannot say that a person X behaves rationally when he accepts the conclusion of a highly convincing argument with a very low degree of certainty, say 0 (which obviously means that X rejects the conclusion), it seems that the definition of conclusiveness has to be changed as follows: Xinfers conclusively if and only if X accepts the conclusion of his argument with the degree of certainty which is equal in value to the logical probability of the conclusion with respect to the premisses. (ii) Rationality of our behavior is a matter of degree; we can behave more or less rationally. Accordingly, should not the concept of conclusiveness possess quantitative rather than qualitative character? In the final paragraphs of the last chapter, the author sums up his methodological investigations concerning inductive modes of reasoning. The third part, 'The Methodological Types of Sciences', consists of five chapters. The first, preliminary one, deals briefly with partition of the sciences into deductive and inductive ones (the latter and only those include statements obtained by inductive arguments). In the second chapter, after describing briefly intuitive, axiomatic, and formalized stages of evolution of deductive theories, Ajdukiewicz presents concisely some of the most important metamathematical notions, such as consistence and completeness of a theory, independence of axioms, and some others. The third chapter is divided into two sections: A and B. Section A is mainly devoted to the concept of observational statement and its status in the structure of science. Section B is a detailed and illuminating introduction into the theory of measurement. The fourth chapter offers a typology of statistical laws. The discussion of statistical laws, such as laws of distribution for qualitative variables, laws of distribution for quantitative variables, laws establishing correlations among variables, and some others, is especially instructive. The fifth chapter is an introduction to the theory of statistical inference. It deals with the following topics: some problems in the estimation of parameters, the correction of errors of observation, confidence limits and tests of certain statistical hypotheses. The book concludes with an appendix: consisting of a few pages of outline of the projected last chapter. It includes a short discussion of the concept of explanation and of the concept of proof. Pragmatical Logic could hardly be used as a textbook, mainly because it is not complete and self-explanatory enough. For instance, the work offers almost no information about the predicate calculus. We have already pointed out the lack of more extensive discussion of the concept of logical probability. Also there are some fundamental problems which have been scarcely mentioned in the book. Furthermore, Pragmatical Logic is short with bibliographical references. Nevertheless the work is undoubtedly valuable in many ways. The particular merit of the book is that while offering a wide exposition of statistical problems it reveals and emphasizes the natural and strong connections that
348
REVIEWS
there are between the logical and the statistical approach to the questions of methodology. This is a worthwhile novelty in the scientific literature of the relevant kind. RYSZARDW6JCICKI
349
REVIEWS Jerrold L Katz and Paul M. Postal, An Integrated Theory of Linguistic Description, The M.I.T. Press, Cambridge, Mass., 1964. x + 178 pp., $6.00. Jerrold J. Katz, The Philosophy of Language, Harper&Row, New York, 1966. xii + 3 2 6 pp., 87.25. Very ambitious claims are currently being made for the relevance of linguistics to philosophy, and these two books together sketch what seems to be the most ambitious tine on this supposed relevance. The earlier book (hereinafter IT) is more squarely a work on linguistics, and the theory there argued for is treated by the later book (hereinafter PL) as a demonstrated conclusion. But IT has a short chapter devoted to the theory's implications for theories of linguistic universals, speech recognition and production, and language learning, and so shares to some degree PL's concern with philosophical issues other than the issue of how to give an account of the structure of a natural language. For the most part, 1T is an extended argument for two slightly-different forms for a generative grammar, here conceived as a device that generates all and only the sentences of some natural language as phonetic shapes, together with an account of each sentence's syntactic and semantic structure. Both forms have three main components: syntactic, semantic, and phonological. The input to the whole device is an initial symbol, 'Sentence'. Two sets of rewriting rules in the syntactic component are supposed respectively to produce, for each sentence in some language, a sequence (possibly unary) of 'underlying Phrasemarkers' and a 'final derived Phrase-marker'. The former are analogous to the atomic sentences of a formalized language, the latter to molecular sentences; but the latter are derived from the former not just by interposing connectives and preposing operators, but by transformation rules that permute and delete items in the underlying Phrase-markers. The main point argued for is that the output of the first set of rewriting rules (the 'base', in Chomsky's handy terminology) should be the input for the semantic component as well as for the second set of rewriting rules. 1 Earlier projects for a generative grammar had not contained this claim, and in fact supposed that the transformation rules would affect the semantic properties of the strings they operated on: particular cases where this supposition operated were those of the passive, negative, imperative, interrogative, and some nominalizing transformations. Katz and Postal present very careful and more or less convincing arguments in all these cases that enough information should be present in the input to the transformational rules that the same input coulji be used in an unaugmented form as the basis for a semantic account of the sentence. The two slightly different models mentioned earlier arise because of two alternative ways that the semantic component might be set up. Each morpheme found in an underlying Phrase-marker is to have a set of semantic markers and selection restrictions assigned to it by the lexicon, which is in this model a part of the semantic component. ~ Semantic markers (hereinafter 'sin') are intended to represent the bunch of concepts that are the meaning of the morpheme. 350
Synthese 17 (1966) 350--365. © D. Reidel Publishing Co., Dordreeht-Holland
REVIEWS
Selection restrictions represent the conceptual characteristics that other morphemes must have in order to combine with the morpheme underlain. F o r example, a selection restriction on 'think' might stipulate that only those nouns with a sm-representing capability of thought in their lexical entry can be used as subjects of 'think' in grammatical sentences; thus the grammar will generate 'People think' but not 'Stones think'. Thus far the two models are the same. The difference depends on whether the semantic account results from the use of one or two kinds of 'projection rules'. The syntactic account of a sentence before transformations may be represented by a tree diagram with 'Sentence' at the root and the items in the 'terminal string' of the underlying Phrase-marker at the tips of the branches. The projection rules' task is to state how the sm assigned to the tips are to behave when we consider the morpheme they underlie as part of a larger constituent, say a tensed verb, a tensed verb with its object and modifiers, or a whole sentence. The projection is conceived as combining the sm that terminate branches from a single point or node, then combining bunches of sm so combined at the next higher node, and so on to the highest node, the one labeled 'Sentence', at which point we will have a representation of the semantic structure of the whole sentence. A really astonishing lack in Katz and Postal's program is that of all the probably very large number of kinds of nodes one would get from an adequate natural-language syntax, they have only proposed projection rules for two kinds: the one subtending a modifier-head construction and the one where the reading for the informational nucleus of an interrogative sentence is combined with the reading for the interrogative morpheme. At any rate, the foregoing sketch fits the kind of projection rule they call 'PI'. The other kind of projection rule, P2, takes readings from clauses or phrases we might think of as derived by transformations from whole sentences and combines them with the readings from the sentence-frameworks in which they are embedded. If the syntax could be made to generate single 'Sentence'-rooted trees for any sentence, no matter how syntactically complex, P1 would be adequate without P2, and in fact this supposition is adopted by Katz in PL and by Chomsky in Aspects
of the Theory of Syntax. The total picture of a grammar is certainly an interesting framework for further research, particularly on the question (which I hope to make appear more mysterious later) what sorts of projection rules will be needed for nodetypes other than the two treated by Katz and Postal. A perhaps even more interesting consequence of the picture is that it claims that phrase-structure rules are adequate to generate structures for every sentence of a"natural language such that these structures will serve as the basis for a full semantic account of the sentences. The interest here comes from the fact that quantification-theory languages are phrase-structure languages too, that is, that they require no transformation rules in specifying their well-formed strings, so that perhaps the very powerful semantic apparatus of modern logic could be brought to bear on natural languages in a new way. But first more on Katz and Postal's new way. 351
REVIEWS
Some readers may find their argument unsatisfactory because it depends in many of its details on what sentences are judged to be grammatical or ungrammatical, while they propose no standard for such judgments. Less finicky readers may stick with them as long as their judgments coincide with the readers'. But even a studied tolerance might falter at Katz and Postal's opinion that exactly one sentence on the following list is grammatical no do not drive the car scarcely touch your food hope to be famous I request that you answer 'someone saw X' does he sometimes eat
(p. (p. (p. (p. (p.
77) 78) 78) 87) 88)
and that exactly one sentence on the following list is ungrammatical did who go home when did John see whom whom did John see when
(p. 101) (p. 106) (p. 106)
when it is decisions on such debatable cases that determine the content of some of the syntactical rules and so, ultimately, the form of the whole model. Still, the adduction and discussion of such bits of English and the attempt to fit them into a system have their value as further work toward a real demonstration of the orderliness of natural languages; the progress toward such a demonstration in I T seems to me considerable. P L is a different sort of book. Its first substantial chapter (Ch. 3) is a very much condensed and already controversial history of philosophy of language in the 20th century. Katz sees in that history two main movements, Logical Empiricism and Ordinary Language Philosophy, with Carnap as the representative figure of the former and Wittgenstein of the latter. The failure of either movement to give satisfactory solutions to philosophical problems is ascribed to inattention on one side to the facts of language use and on the other to the real underlying systematic nature of language. Katz maintains that his own approach, incorporating attention to both, will succeed where they failed. Katz maintains also that his own definitions of semantic concepts escape the charge of circularity that Quine made in 'Two Dogmas of Empiricism' against Carnap's definition. "The crucial point", says Katz, "is that their defining condition is stated solely in terms of formal features of semantically interpreted underlying phrase markers so that none of these concepts themselves need appear in the definition of any of the others" (pp. 172-173). Indeed, the twenty-five semantic concepts defined in P L 3 do not appear ineliminably i n each other's definientia, so the circle is not as small as the one Quine found in Carnap's work. But there seems to be a circle nevertheless. The terms designating "formal features of semantically interpreted underlying phrase markers" employed are these: 'reading', 'set of readings', 'assignment of a set of readings', 352
REVIEWS 'semantic marker', 'sentence-constituent', and 'semantically interpreted underlying phrase-marker' itself. The most important feature in this group is 'semantic marker', for the others are all definedin terms of it. The brief characterization of sm given above needs to be amplified. Since it is not entirely clear what "the philosopher's notion of a concept" is which groups of semantic markers are supposed to "reconstruct" (p. 157), perhaps the best way to get clear on sm is to inquire what they do. Two interesting answers are these: a dictionary entry made up of them "permits us to represent every piece of information about the meaning of a word required by the projection rules in order for them to operate properly" (p. 154); and the sm themselves "provide the theoretical constructs needed to reconstruct the interrelations holding between such conceptual elements in the structure of a sense" (pp. 155-156). As far as the first nine defined concepts are concerned, then, sm and selection restrictions have these joint functions: they prevent readings being assigned to constituents that a fluent speaker would judge semantically anomalous; they make it that semantically unambiguous and n-ways ambiguous constituents (again conformably to a fluent speaker's judgment) get exactly one and n readings respectively; and they make it that (fluent-speaker-judged) synonymous constituents get the same and semantically distinct constituents different readings. Similarly for the other sixteen concepts. Since sm and selection restrictions are theoretical entities, the theorist is constrained to choose them so that they fulfil these joint functions. Selection restrictions and sm that do not do the job are excluded from the theory. In fact, the only way Katz prescribes in which to choose sm is to choose just those that do the job. It is just here that the definitions begin to look circular. The six 'formal features' listed earlier are all defined in terms of sm, selection restrictions, and underlying phrase-markers. The last item is a contribution of the theory of the syntactic component; if we scrutinize the contribution of the semantic component, we find the following situation: sm and selection restrictions are those theoretical entities that result in ascriptions of the twenty-five properties that conform to ascriptions made by fluent speakers; and ascriptions of the twentyfive properties are made on the basis of the permissibility, number, and overlapping of assignments of semantic markers and selection restrictions. We will have established that there is a circle in definition here if it turns out that sm and selection restrictions not only have these tasks, but also have only these tasks; so we now inquire what else they are said to do. There appear to be two tasks: sm "enable us to construct empiricalgeneralizations about the meaning of linguistic constructions" (p. 157); and "Derived readings" (configurations of sm from different morphemes) "provide a characterization of the meaning of the sequence of words dominated by the node to which they are assigned" (p. 165). His example of the first is that the presence of the sm '(Male)' in the reading for each member of a list including 'bachelor', 'man', 'priest', 'the most unpleasant uncle I have', etc. and not in 'child', 353
REVIEWS
'mole', 'mother', 'the whole truth', etc. enables us to construct the empirical generalization that the members of the first list share the concept maleness, whereas those of the second list have not got it at all. The circularity of this task is made clear at the end of the passage, where he says that"the mode of expressing semantic generalizations is the assignment of readings containing the relevant semantic marker(s) to those linguistic constructions over which the generalizations hold and only those" (p. 158). There is a second circle in definition within the presumptive circle being explored, and one more far-reaching in its effects on Katz's position. The only way we have of getting a grip on a sm is to specify what we might call its 'extension': the class of morphemes in whose readings it will turn up. This is a natural result of a stipulation of Katz's that the labels, e.g. '(Male)', are not intended to represent English words, but only to be suggestive labels for the concepts underlying groups of English words. A sm might be characterized as that which the meanings of the members of a certain set of morphemes have in common insofar as they contribute to the meanings of expressions. This raises the question why we have a sm for the common meaning-contribution of some morpheme in the constitution of 'bachelor', 'man', 'priest', and 'the most unpleasant uncle I have', and not for the like contribution of 'apple', 'walks', 'until', and 'Charles V'. (We might suggestively label this '(Zero)'.) It seems to be Katz's position that we must depend on our intuitions to tell us when expressions have some elementary part of their meaning in common; that is, we must have some generalization about commonness of meaning in mind before we (as lexicographers) decide what sm to assign to a word. It seems strange, then, for Katz to say that sm enable us to construct empirical generalizations about commonness of meaning. The strongest statement that seems to be warranted is that the sm give us a new vocabulary in which to express such generalizations. The other thing that sm and selection restrictions are said to do - provide a characterization of the meaning of (e.g.) a sentence - is more interesting, since it is not very close to the twenty-five properties in its import. As an example, I will adapt one of Katz's sample 'sentoids': 'Criminals chase police'. To avoid some complications, though, I will use names (treated according to 1T, p. 90): 'John chases Jim'. Presumably the selection restrictions on 'chase' are at least something like this (I use 'X' and 'Y' as Katz does for variables ranging over subject and object respectively): ((X: (Physical Object) & (Animate) & (Capable of Intention)) & (Y: (Physical Object) &(Animate)) ). Among the readings for 'John' and 'Jim' should be something like '(Named "John"), (Human)', where '(Human)' entails the presence of the markers '(Physical Object)', '(Animate)', and '(Capable of Intention)' by redundancy rules of the kind introduced on p. 230 of PL. Together with the following 354
REVIEWS version of Katz's reading for 'chase', reconstructed from the remarks following the thoroughly misprinted version on p. 167, chase ~Verb, Verb Transitive, ... ; ((Activity) (Nature: (Physical)) of X), ((Movement) (Rate: (Fast)) (Character: (Following))), (Intention of X: (Trying to catch ((Y) ((Movement) (Rate: (Fast)))))); , and with a phrase marker something like this, Sentence Verb Phrase
Noun Phrase
I
Noun
Verb
Noun Phrase Noun
John
Present
chase
I
Jim
we are ready to get a reading for the maximal constituent, Sentence, by combining readings from the terminal nodes up, node by node. Neither of the projection rules cited earlier gives us help in this particular case, but from the informal material provided by Katz's discussion, we can carry out the process that seems to be intended. The end product is this (neglecting tense, which Katz does not treat explicitly): ((Activity) (Nature): (Physical)) of ((Named 'John') (I-Iuman))), ((Movement) (Rate: (Fast)) (Character: (Following))), (Intention of ((Named 'John') (Human)): (Trying to catch (((Named 'Jim') (Human)) ((Movement) (Rate) (Fast)))))). Perhaps with a clear formulation of the projection rule required in this case we could fred a way to compare this sentoid-reading with others and so, on the basis of their Boolean properties, to decide which of the twenty-five semantic concepts were predicable of it. But it hardly seems to "provide a characterization of the meaning" in any more interesting sense; considered as a string of English words in parentheses, it is more or less unintelligible, and since Katz assures us that it is not a string of English words, but a collocation of representatives of concepts, it is only more clearly unintelligible. And the meaning of a sentence is a fairly important concept to accommodate in a semantic theory. In case it be thought that this sentence reading, which is constructed from what Katz says and which does not actually appear in the book, is not a fair 355
REVIEWS
representation of the sort of string that will appear as a sentence reading, here is a specimen (the only one in the book) constructed by Katz himself: "(Some contextually definite), (Physical Object), (Nonliving), (With Blade), (With Handle), ( + Evalus: (ease of dividing substances softer than its blade))" (10. 300). This is a reading for 'The knife is good'. It seems slightly more intelligible than the reading for 'John chases Jim', but still, unaccompanied as it is by rules for interpretation, and couched as it is in symbols that are avowedly not English words, it is unintelligible enough. Again, however, it can b e used as a basis for predicating all of Katz's defined properties. I think it is fair to conclude that Katz's semantic theory is involved in as round, if not as small, a circle as the one Quine points out in Carnap's treatment of analyticity. The theory in fact extensively resembles Carnap's recommendation that we specify the predicates we are going to use and then construct (or rationally reconstruct) all sentences (or all we need) from there. An sm is the intersection set of the meanings (readings) assigned to its 'extension'. If the readings are assigned first, the linguistic intuitions and good will of the assigner could exclude sm like '(Zero)' above. If they are not, arbitrary sets could be sm 'extensions'. So, ifKatz's enterprise is to succeed, he must first decide what the vocabulary of sm is going to be and which morphemes are to have which sm in their readings. This is exactly tantamount to the Carnapian procedure, wherein we first settle on our primitive predicates and the meaning rules to translate them into ordinary language and then proceed to construct whatever sentences we need for our purpose. As a linguist, Katz would have to be guided in his decisions about the basic vocabulary of sm and selection restrictions by the judgments of native speakers on what sentences or smaller constituents are anomalous, ambiguous, synonymous, etc. Katz criticizes Carnap, perhaps rightly, for being too tolerant in this regard. But the formal similarity is still evident. Katz further claims as a merit of his definitions that they "avoid the charge of empirical vacuity that was correctly made against Carnap's constructions. ... Their empirical success depends on whether, in conjunction with the set of semantically interpreted underlying phrase markers for each natural language, they correctly predict the semantic properties and relations of the sentences in each natural language" (p. 173). But since Katz's semantic theory, except for Chomsky's syntactic contribution, is formally so similar to Carnap's, it is easy to see that Carnap's theory could be empirically tested in exactly the same way as Katz's. The Carnap language would have to be equipped with enough predicates to express every natural-language concept, and with the proper sort of meaning rules to connect the predicates with the natural-language words they are supposed to render. Then purportedly analytic (semantically anomalous, ambiguous, etc.) sentences constructed in the artificial language could be
356
REVIEWS
translated into the natural language in question and submitted to a native speaker for his judgment as to whether the sentence is indeed analytic (anomalous, ambiguous, etc.). On pp. 54--60 ofPL, Katz very cogently criticizes some empirical tests for semantic concepts proposed by Carnap in 'Meaning and Synonymy in Natural Languages' (Philosophical Studies 6 (1955) 33-47), but the tests proposed here are different in ways that ought to be acceptable to Katz, since they are analogues of his own tests. Thus neither advantage Katz claims for Iris definitions over Carnap's is a real advantage (and if specifying the full needed range of predicates is a will-o'-the-wisp for Carnap, so is specifying all the needed sm for Katz). When we compare the two kinds of theory on a further point, Katz's comes out at a definite disadvantage. Carnap tells us the meanings (though that is not his intention) of sentences in the artificial language by giving us natural-language sentences as renderings, so that, insofar as the rendering is adequate, and insofar as we understand the natural language, we understand the artificiallanguage sentences. Katz tells us the meanings of sentences in English by giving us unintelligible sets of nested expressions. Katz's main philosophical claim, excluding his claims about language structure, is contained in the last chapter of his book, 'Implications for Understanding Conceptual Knowledge'. The claim is that his theory provides the basis for 'fundamental progress' in solving four important philosophical problems, which he labels 'analyticity', 'semantic categories', 'innate ideas', and 'linguistic analysis'. By exhibiting this progress he hopes to lend weight to the theory itself. But in the three cases where semantic theory is employed to a large extent in the attempt, shortcomings of the theory render the solution unsatisfactory. The solution of the problem of analyticity depends on simple Boolean inclusion of predicate sm in the array of subject sm, and the predicate 'is analytic' is defined only for declarative copula sentences of a very simple sort and for conditional sentences composed of such copula sentences. Part of his defense of the definition is his demonstration that the negations of analytic, contradictory, and synthetic sentences are respectively contradictory, analytic, and synthetic, as one would expect. This demonstration, however, depends on a definition of negation (p. 201)4 that in turn rests on the notion of antonymy. The effect of negation on the sm underlying a predicate term is, roughly and with many qualifications, to replace them with all the other sm in their 'antonymy sets'. For instance, it seems to be a consequence of the definition that the configuration of sm underlying 'John is not a child' is the same as the configuration underlying 'John is a baby or a toddler or an adolescent or an adult' (see p. 196). Suppose now that '(Square)' appears in the reading of a predicate and that it is neither identical with nor antonymous to any sm in the reading of the subject (for these are some of the qualifications). Then clause (i) of the definition instructs us that when the containing sentence is negated, the effect is to re357
REVIEWS place '(Square)' in the reading by a Boolean union of all the sm for every notsquare shape. It is easy to see that this set is infinite in size. Similar examples could be constructed using any notion selected from a continuum, e.g. '(Seven)', '(Red)', '(Middle C)', '(Now)', etc. There is, of course, nothing to keep the theorist from denying that these will be sin. But Katz's own treatment of the semantic properties of 'good' requires that the symbol (q-) be assigned a degree to account for the behavior o f ' g o o d ' in 'x is better than y' or 'x is as good as y' (pp. 308-309); such a treatment would likely require a continuum of degrees. Moreover, color words would not seem to lend themselves to a plausible treatment as underlain by complex meanings, and clearly the colors (and so at least possible color-words) form a continuum. Difficulties such as these seem to diminish the impressiveness of Katz's proof that his treatments of analyticity, contradiction, and negation make it that the negation of an analytic sentence is contradictory and vice versa, and that the negation of a synthetic sentence is synthetic. The solution of the problem of semantic categories is vitiated by a different difficulty: the method of determining which sm underlie which words in a language. Katz criticizes Aristotle for relying on "intuitive judgments about what are and what are not the most general answers to questions like 'What is Socrates?' or 'What is greenness?' ", and for presupposing "our intuitive understanding of the notion of maximal generality ... without providing any analysis of it" (p. 225). Katz's approach is to introduce 'redundancy rules' that record such facts as that whenever the sm '(Human)' appears in a reading, so does '(Physical Object)', but not vice versa. This provides a formal representation of the supposition that one can truly predicate 'is a physical object' of everything of which one can truly predicate 'is human'. Besides recording such facts, the rules allow us to rewrite dictionary entries so that we can omit '(Physical Object)' where '(Human)' occurs, knowing that we can resupply the omitted sm when necessary by reference to the redundancy rules; thus we get a smaller dictionary. Katz represents the form of such rules in this way: [(IVI.1) v (MO v . . . v (M=)]-->(M~), which might best be read 'If a reading contains sm (M1) or (M2) or ... or (Ms), then it also contains (Mk)', though that is not quite Katz's version. He then identifies the semantic categories of a language as the set of sm that appear on the right of some arrow, but on the left of no arrow, in the redundancy rules of the language. Here, he says, is a rigorous formulation of how to find semantic categories, with an explication of the notion of maximal generality. But consider the fact about sm pointed out earlier, viz. that the only way to get a grip on them is to consider their 'extensions'. A consequence of this is that the only way to get a grip on what will be to the left or right of Katz's redundancy-rule arrows is to consider whether the 'extension' of one sm is contained in the 'extension' of another. Take the case, then, where (Mk) in the above 358
REVIEWS
schema is the total reading for some single word or expression (as, presumably, '(Physical Object)' is for 'physical object') and (M1), (Ms) ... (Mn) are the total readings for single words or expressions (e.g. '(Human)', '(Plant)', '(Artifact)', for 'human', etc.). Then the sort of question we must consider to establish the redundancy rules is 'Is every human, plant, or artifact a physical object, whereas not every physical object is a human, nor every physical object a plant, nor every physical object an artifact?' Surely questions of this kind were well within the logical powers of Aristotle. It is hard to see that Katz has made a real advance in this regard, except that his promise of a set of redundancy rules (and so an enumeration of categories) is shored up by a promise of an exhaustive dictionary of the language in the sm notation, so that the whole enterprise may momentarily seem more impressive than Aristotle's. But it hardly relies less on intuition, since at its base are intuitive judgments as to which sets of morphemes do share a semantic marker. Katz points out that even with this much accomplished, we would have ascertained the semantic categories only for a single language, whereas former theories of the categories seem to have envisioned something that transcended a single language. He has a project for the latter task, but it seems that it will not work in the way he supposes. "There will be a set of redundancy rules that is the intersection of the sets of redundancy rules found in the dictionaries of linguistic descriptions of each particular language" (p. 235). Then the semantic categories of language generally will be those markers that occur to the right of some arrow and to the left of no arrow in this intersection set. Katz does not consider how one could determine the intersection set, but it is clearly a difficult problem. Suppose the English redundancy rules contain (1)
(Red) v (Blue) v ... v (Color) v ... v (Soft) v (Hard) v ... -->(Quality) and the French redundancy rules contain
(2)
(Rouge) v (Bleu) v . . . v (Couleur) v ... v (Mou) v (Dur) v ... ~ (Qualit6)
and that French and English are the only languages there are. We are not clearly justified in counting these as one rule, since the 'suggestive labels' for the sm are different. But on what grounds can we make them the same? Suppose that both rules appeared as (1) or that they both appeared as (3)
(P1) v (P~) v . . . v (P~) v . . . v (P~) v (P~+l) v . . . ~ ( P n ) .
We would then clearly be justified in including such a rule in the set of universal redundancy rules. But how could we have discovered such a redundancy rule for French in case (1) or for either language in case (3)? The course that immediately suggests itself is translation. But even if we allow 359
REVIEWS
that English 'blue' translates into French as 'bleu', we cannot both translate and take seriously Katz's assurance that what looks like 'blue' in '(Blue)' is not an English word. We have no rules or traditions to help us translate sm (if they are even the sort of thing that can be translated). The only other apparent course is the one we have already used to get a grip on sm: since weknow what an sm is when we know its 'extension', we can conclude that two sm with the same 'extensions' are identical. F o r this course to do us any good, we might do two things: (a) allow words to be translated, so that the sm that has as its 'extension' 'blue', 'bluish-green', 'the blue of the sky', etc. may be supposed to be the same as the sm that has as its 'extension' 'bleu', 'vert quasi bleu', 'le bleu du ciel', etc. ; or (b) amalgamate the languages so that (P~) has as its 'extension' 'blue', 'bleu', 'bluish-green', 'vert quasi bleu', etc., relying on the intuitions of a bilingual. But there are difficulties with either course. Suppose we chose course (a) and attempted to determine the intersection set of redundancy rules for English and some language that did not contain a word for the color puce; it is plausible that there are many such languages. Assuming again that simple color words will not lend themselves to an analysis as underlain by complexes of sm, we are faced with the English rule (4)
(Red) v (Blue) v (Puce) v (Green) v ...--> (Quality)
and the Other-Tongue rule (5)
(Red) v (Blue) v (Green) v ... ~(Quality);
and we must conclude that (Quality) is not a semantic category for the two languages, and so not a universal category. This specific result is not by itself very important, but it suggests a general way of generating counterexamples to the supposition that two languages share a redundancy rule: find a vocabulary item in the 'extension' of a sm on the left of the arrow in one language's rule for which there is no translate in the other language; then the rules fail to coincide at the point where this sm occurs. It may be objected that some sort of translation, however lengthy, must be constructible in any language for expressions in any other language; suppose, for example, that we run across a Kwakiutl color-word 'grue'. N o matter whether any current English colorword is an acceptable translation for it, we can render it as 'the color a Kwakiutl speaker designated by the word "grue" '. The maneuver does not work, though, for the English expression would probably be underlain by a fairly complex group of sm, whereas 'grue', on our earlier assumption about color-words, would not. A n d even if he sm underlying 'grue' were complex, it is reasonable to doubt that they would include the sort of components (particularly in regard to their syntactic structure) that underlie the relative clause about a Kwakiutl speaker. Color words may be an especially easy case to exhibit as counterexamples, but the technique seems to admit of general application. 360
REVIEWS
An example drawn from Katz's own work is 'anaesthetic'. It seems reasonable to assume that there are languages without non-discursive translational equivalents of this word, and yet Katz assures us that it is underlain by a simple sm, viz. "(Evah (Effective in producing a temporary loss of feeling in a part of a body))" (p. 284). Since the noun 'anaesthetic' and the adjective 'anaesthetic' presumably belong respectively to the categories (Stuff) and (Quality), it follows that neither (Stuff) nor (Quality) will be a universal semantic category. Katz claims (p. 294, fn. 1) that hypnosis is an anaesthetic, also, so that (Process) cannot be a universal semantic category. Of course, it may be doubted that these are semantic categories, but the occurrence of 'anaesthetic' in English and not in some other language hardly seems the right sort of grounds for the doubt. Now suppose instead that we take course (b). If we amalgamate any two languages, the objection against course (a) for the universe consisting of those two languages would clearly be met, since the objection rests on a distinction between the 'extensions' of sm in different languages (or lists of the vocabulary of different languages). But the difficulties arise again with a third and a fourth language, and so on; when they arise, our only move is to accept the newcomer as a new part of a larger amalgam, and we shall end up depending for our rules on the intuitions of a panglot. Besides the practical situation here being far from what Katz seems to envisage, the theoretical situations has one very unattractive feature: our amalgamation will have effectively given us one world-language, with a very redundant vocabulary, and the notion of a theory of language as a reconstruction of the competence of a native speaker is no longer viable, since no one is a native speaker of all languages. So Katz's solutions of the problems of analyticity and semantic categories are unsatisfactory and can hardly be claimed as merits of his semantic theory. There is already a sizeable and growing literature on the third problem Katz treats, viz. innate ideas. There is not mlxch point in adding remarks to it that must necessarily be brief and so probably inadequate. The fourth problem Katz works on is that of 'linguistic analysis': "how to formulate the meaning of a philosophically significant word so as to clearly exhibit every facet of its meaning that might be involved in those epistemological questions in whose statements the word enters essentially. In other words, how should we go about analyzing the meanings of words, and with what kind of representation should we express the results of such analysis" (p. 283). If the last part of the b o o k did succeed in the two tasks, and succeeded in terms of Katz's theory, one could maintain that he had marshalled impressive support for the theory. However, there is no discussion of the first task, and for the second, he refers us back to the discussion of Ch. 4, where the apparatus of sm and selection restrictions is explained. The evidence of the text is that the first task is done thus: I intuitively decide that a certain group of words is underlain by such-and-such sm and selection restrictions by asking myself questions about or depending upon the ordinary361
REVIEWS
language equivalents of the twenty-five defined properties, e.g. 'Are all bachelors really unmarried?' or 'Is chasing an activity?' I then 'test' my conclusions about what sm and selection restrictions are present by seeing whether they combine in such a fashion as to render ambiguous sentences ambiguous, analytic sentences analytic, etc. A n d if I want a real command of what sm there are, I must decide of every word of morpheme in the language (and so, presumably, decide by recursive rules of every phrase, clause, and sentence in the language) what sm underlie it. Katz goes on in the chapter to provide a sort of case study of the analysis of a philosophically significant word ('good' as an adjective or adverb). In one respect it differs in procedure from the sketch just given: there is no attempt to get a complete command of what sm underlie 'good'; in fact, Katz explicitly excludes consideration of the occurrence of 'good' in 'He is a good man'. This may be the reason that Katz suggests late in the b o o k that perhaps the philosophical implications of the meanings of 'good' and 'bad' are "nonexistent or trivial as far as the substantive issues of moral philosophy are concerned" (p. 311). But though the case study may appear to diverge from the first part of my sketch, it does not. Katz describes his technique on pp. 288 and 289. The two points at which his description diverges from my sketch are a mention of projecting the meanings of infinitely many sentences and a mention of showing the paraphrase relations between sentences. As was pointed out above, this is the only section of the book in which Katz gives a full reading for a sentence, viz. "(Some contextually definite), (Physical Object), (Nonliving), (With Blade), (With Handle), ( + Evalus: (ease of dividing substances softer than its blade))." We are given a basis from which a large number of such readings can be constructed, although we should have ourselves to guess at the non-evaluative sm underlying the other words for objects we might wish to call 'good'. But since recursive devices for constructing readings are only hinted at (e.g. on pp. 290--291), we have hardly been given the basis for an infinite number of such readings. Most important, no step is taken in this section to make it any clearer how a reading such as the one provided gives us the meaning of the sentence, unless we happen to guess right. The mentions of paraphrase relations are not very complete or helpful either. One would suppose that on Katz's theory of the meaning of 'good', 'This knife is good' and 'This knife divides substances softer than its blade with ease' would be mutual paraphrases. But it seems extremely implausible that the whole verb phrase in the second sentence would be underlain by a single sin, as would be required were the theory to mark this paraphrase relation. A n d if the relation were so marked, there would not be enough structure in the second sentence's semantic analysis to m a r k its semantic relations with 'It divides the room', 'The sofa is softer than the chair', 'This knife's blade is missing', etc. Perhaps Katz 362
REVIEWS
could get around this difficulty by giving more semantic structure to the phrase embedded in the '(Evalu~: (0))' marker, but such a course would require a more complicated formal apparatus than anything in the theory so far. Some moves are made in this direction on pp. 304-307, where Katz begins an explicit discussion of paraphrase relations between sentences. The two cases he discusses are (1) that of the pairs of sentences 'The knife is good'/'The knife cuts well' and 'The knife is bad'/'The knife cuts badly', and (2) that of "paraphrase relations between copula-sentences and sentences with transitive main verbs" (p. 306). In case (1), one of the sm underlying 'cuts' is '(Evalu~: (Ease of dividing substances softer than blade of X)', where 'X' marks the slot for the reading of the subject of 'cuts'. This innovation, together with an apparatus to prefix a plus or minus to 'Eval', will "yield derived readings that are the same as those for the corresponding copula-sentence paraphrases" (p. 301). But if we follow Katz's directions as given in the passage, part of the set of sm underlying 'The knife cuts well' will be "(-k Evalu~: (Ease of dividing substances softer than blade of ((Some contextually definite), (Physical Object), (Nonliving), (With Blade), (With Handle), (Evalus: ease of dividing substances softer than its blade))))", which is unlike what he cites as underlying 'The knife is good' (see above). It might be argued that we need only alter the reading Katz gives in one respect to bring it into line with this reading: change 'its' in the evaluation sm to 'of X' where 'X' is a slot to be filled by the reading of what is indicated by 'it', viz. 'knife'. The unhappy result can be only partially represented thus: "(Some contextually definite), (Physical Object), (Nonliving), (With Blade), (With Handle), (+Evalu~: (ease of dividing substances softer than blade of (Physical Object), (Nonliving), (With Blade), (With Handle), (Evalus: (ease of dividing substances softer than blade of (Physical Object), (Nonliving),
...)))) ...",
since from this point on, the reading for 'knife' must be infinitely many times embedded within itself. We must conclude that Katz's treatment of case (1) is not successful. Insofar as we credit Katz's claim that the treatment of case (2) is "essentially the same", we can expect it to have the same infirmity. The difference is that instead of filling one slot with a reading for the verb's subject, we fill two slots, one for the subject and one for the object. He does not treat any examples of this process, but it is reasonable to suppose that it (like case (1)) would either fail on one side to show the proper paraphrase relations because of the presence of 'its' in the sm, or fail on the other because on an infinite enlargement of the subject reading. The object reading would probably not introduce any new 363
REVIEWS
complications, since the object is not being evaluated and the difficulties are introduced by the evaluation sm. The elaboration o f details and supposed illuminating consequences of PL's semantic theory is in a pretty unsatisfactory state. It is difficult to see h o w to support the claim that it will stay in that state, but it is just as difficult to see h o w its one great infirmity - the lack of a way to show intelligibly what a sentence means - can be cured with the existing apparatus. As an alternative, one might take D o n a l d Davidson's position that a truth-definition is an adequate theory of meaning, since it would give the meaning o f each sentence on the basis o f its structure (see ' T r u t h and Meaning', Synthese 17 (1967) 304-323), and add to it a Katz-type 'semantic theory' as a m o d e l to show in a clear way what some very simple sentences have in c o m m o n that renders t h e m susceptible to c o m m o n semantic judgments. This alternative makes more tenuous the links between semantics and theories of speech production and recognition, but that may be a step in the right direction. This alternative would not, however, tend to diminish the great interest and importance of the w o r k on language done in I T and popularized in P L )
University of California, Berkeley
BRUCE VERMAZEN
REFERENCES 1 The output of the second set is then the input for the phonological component, which produces a phonetic representation of each sentence. The two books' neglect of phonology will be fully mirrored in this review. 2 Noam Chomsky, in Aspects of the Theory of Syntax, calls a subcomponent of the syntactic component the 'lexicon', but he envisages it as assigning only some of a morpheme's semantic markers and selection restrictions to it - enough to keep the output of the grammar from containing such 'ungrammatical' strings as 'Some idea is green'. The rest of the assigning is done by the semantic component, however, so Chomsky's model still has a Katz-Postal 'lexicon' (unnamed) in addition to a Chomsky 'lexicon'. 8 The concepts are 'semantically anomalous', 'semantically ambiguous', 'semantically ambiguous n-ways', 'synonymous on a reading', 'fully synonymous', and 'semantically distinct' for any constituent, plus the first and third of these and 'semantically unambiguous' for sentences; then 'analytic on a reading', 'fully analytic', 'contradictory on a reading', 'fully contradictory', 'synthetic on a reading', and 'fully synthetic' for what he calls declarative copula sentences, 'entails on a reading', 'fully entails', 'analytic on a reading' for conditional sentences, the first, third, and fifth of this second group for sentences with a relative clause modifying the subject, 'indeterminable on a reading' and 'fully indeterminable' presumably for declarative copula sentences, and 'metalinguistically true' and 'metalinguistically false' for sentences with singular terms describing sentences as their subjects and with one of the first twenty-three defined properties as their predicates. 4 There seems to be a mistake in clause (ii) of the definition. The expression 'k + 2" seems not to make sense, whereas 'iq-k + 1' would. This is also a good place to point out that the definition of 'antonymous n-tuple' on p. 197 (section (iii)) seems to have 364
REVIEWS nothing to do with the succeeding definitions, though it is presented as one of their bases. What wouM be useful is a definition of 'antonymous n-tuple' that applied to n-tuples of sm. 5 In preparing this review, I have benefited from conversations with Diana Axelsen and Donald Davidson of Stanford University.
365
REVIBWS Rudolf Carnap, Philosophical Foundations of Physics. An Introduction to the Philosophy of Science (ed. by Martin Gardner), Basic Books, New York, 1966, x + 300 pp., $6.50. Carnap's book fills a great gap in the literature of the philosophy of science. Hitherto, there has been nothing which was really satisfactory as a text for a beginning course. Most philosophy of science books are either too specialized or too advanced for that purpose, and one or two are too elementary. Carnap's work is just right. It is just right, that is, provided the instructor already knows a great deal about the philosophy of science. Otherwise, one of the main virtues of Carnap's book in its role as a text will turn into a vice. Carnap never pursues any subject far enough into details to make it wholly clear to a novice what is going on. The job of pursuit and clarification must be carried on by the instructor. That is what makes Carnap's book such a good one to teach from. The instructor is not made useless by the text; and, at the same time, the book furnishes an excellent foundation for the discussion of several philosophical problems. The issues which Carnap deals with reflect his own interests in the philosophy of science. There are two main themes, namely the application of formal logic to science and the relation o f geometry to relativity theory. The logical discussions include treatments of probability statements, causal modalities, and the analytic-synthetic distinction. The discussions of geometry and relativity theory include analyses of measurement, non-Euclidean geometries, and the geometry of empirical space-time. In addition to his two main themes, Carnap takes rapid glances at the notion of explanation and at determinism in relation to quantum theory. The chapters on logical topics need the most simple amplification by an instructor. There are at least three points where that need is apparent: (1) A beginning student would likely come away from Carnap's examination of probability with the idea that there is one, rather unsatisfactory, kind of probability theory usable by gamblers; another, more satisfactory, probability theory usable by insurance companies; and a third theory which Carnap likes best but it is not clear why. (2) In his treatment of causal statements, Carnap translates them into universal material implications. He has a good discussion of whether material implication might omit some element of necessity present in causal statements. But he does not consider objections to such translation arising from the truth table definition of material implication. (3) Carnap says he has a hunch that conditionals which assert scientific laws and conditionals which do not can be distinguished by the inspection of some semantic property of statements (not of propositions). And he throws in some symbolism involving a language-metalanguage distinction. The reliance upon hunches, the references to semantic properties, the difference between statements and propositions, and the distinction of language from metalanguage come without advance notice and can only bewilder a beginner. The symbolism does no more 366
Synthese 17 (1967) 366-367. © D. Reidel Publishing Co., Dordrecht-Holland
REVIEWS
than show that what can be said in words can also be expressed by arbitrary symbols. While the logical discussions must be amplified by an instructor, the treatment of geometry and space has to be straightened out in a couple of places: (1) In one chapter, Carnal) warns strongly against using the word 'curvature' in anything but a technical sense. One must not attribute to it the usual connotations of ordinary language. Yet, in another chapter, Carnap talks of geodesics as being the 'shortest' distance between two points; and he says, without any warnings about technical terms, that no gravitational force need be postulated since planets simply travel in the shortest paths they can take. (2) Carnap is admirably insistent that space curvature is a physical concept, not a purely mathematical one, and that, in relativity theory, the physical concept of curvature replaces the physical concept of gravitation of Newtonian theory. But, further on, Carnap also claims that relativity theory is simpler than Newtonian theory; for relativity has eliminated the need for talking about something with universal physical effects, namely gravity. He fails to realize that, in accord with his own point, relativity theory introduces a substitute something with universal physical effects. A few miscellaneous comments upon some other points in the book go as follows. There is one place where Carnap's past of logical empiricism raises a potential difficulty that is minor in relation to the book but major in relation to philosophy. He holds that the correspondence rules for theoretical terms cannot be specified once and for all. By contrast, his whole attempt to define analyticity rests upon the assumption that statements about material objects can be translated once and for all into statements about conjunctions of sensible qualities. There is no reason to suppose that the latter translation is any more possible than the specification of a complete set of correspondence rules. Carnap's discussions of measurement, geometry, and the creation of physical concepts are excellent. His brief discussion of free will, white philosophically much too simple, is enough to show that quantum theory is irrelevant; and that is all that need be done in a book on the philosophy of science. The last chapter, on quantum theory itself, is clear but hardly more than an afterthought.
University of Cincinnati
ROLLIN W. WORKMAN.
367