J Gen Philos Sci (2011) 42:17–31 DOI 10.1007/s10838-011-9153-1 ARTICLE
Semantic Challenges to Scientific Realism Holger Andreas
Published online: 13 May 2011 Ó Springer Science+Business Media B.V. 2011
Abstract This paper is concerned with connections between scientific and metaphysical realism. It is not difficult to show that scientific realism, as expounded by Psillos (1999) clearly qualifies as a kind of metaphysical realism in the sense of Putnam (1980). The statement of scientific realism therefore must not only deal with underdetermination and the dynamics of scientific theories but also answer the semantic challenges to metaphysical realism. As will be argued, the common core of these challenges is the proposition that a (metaphysical) realist semantics leads to semantic agnosticism in the sense that we are unable to grasp the proper meanings and referents of our linguistic expressions. Having established this, I will focus more specifically on the question of whether scientific realism—in its state-of-the-art account—has the resources to make reference to scientific concepts intelligible such that the semantic challenges can be answered. Keywords Causal-descriptive theory of reference Internal realism Metaphysical realism Scientific and structural realism Ramsey account of scientific theories Theoretical concepts
1 Introduction In investigation of the semantics of linguistic expressions, a distinction emerged between scientific realism and a more general conception of realism. The latter conception also includes, among other fields of discourse, statements about the observable world, observable in the sense that no explicit scientific knowledge is used to make an observation of spatiotemporal states of affairs. Scientific realism, by contrast, is concerned with the semantics of statements made in the context of an explicit scientific theory.
H. Andreas (&) Seminar fu¨r Philosophie, Logik und Wissenschaftstheorie, LMU Mu¨nchen, Geschwister-Scholl-Platz 1, 80539 Mu¨nchen, Germany e-mail:
[email protected]
123
18
H. Andreas
Prominent arguments that have been levelled against the broader conception of realism are Putnam’s famous model-theoretic argument as well as the acquisition and manifestation argument advanced by Dummett (1978, 1991) and systematised by Wright (1993). The target of these arguments is a certain kind of truth-conditional semantics, which is viewed as the essence of metaphysical realism by Dummett and Putnam. Taking up the strand of these arguments, the paper identifies the broader conception of realism with metaphysical realism. The thesis of underdetermination of scientific theories by observable evidence and the dynamics of scientific theories are taken as major thrusts to scientific realism, whereas the success of modern science seems to suggest a realist position about scientific statements. Focusing the debate about scientific realism on these themes is reasonable as they particularly pertain to an understanding of scientific language. It may nevertheless be important and necessary for such an understanding to investigate the relation between scientific realism and metaphysical realism as well. It will be shown that the semantic stance of scientific realism being explicated by Psillos (1999) is such that it is precisely the target of the major arguments against metaphysical realism. Therefore, Putnam’s model-theoretic argument directly applies. This may not come as a surprise, but Psillos (1999) and other major contributions to scientific realism do ignore this argument. Equally little attention received the acquisition and manifestation argument by Dummett (1978, 1991) in the search for an appropriate semantics of scientific language. As will be shown, a certain variant of the acquisition argument applies to scientific language to an even higher degree than Dummett’s original argument applies to everyday and mathematical language. Showing that the semantic challenges to metaphysical realism are an issue for scientific realism is not the sole purpose of the present paper, however. In addition to establishing this critical result, the paper is seriously concerned with the question of how reference to theoretical concepts can be made intelligible. This question will be approached from different and yet interrelated perspectives. First, from Psillos’s causal descriptive theory of reference, second from the problem of theoretical terms (Sneed 1979), which is presented in the form of a Dummettian acquisition argument, and third, from Putnam’s internal realism. A structuralist realist approach to science in the Ramsey-style will emerge as one promising answer to the question of how we can ever come to refer successfully to theoretical concepts. The constructive part of the paper can be read to advance an acquisition argument for structural realism.
2 The Semantic Stance of Scientific Realism Psillos (1999) has become the standard reference for scientific realism and the following characterisation is, of course, well-known (xix): (1) (2) (3)
The world has a definite and mind-independent natural kind structure. Scientific theories are truth-conditioned descriptions of their intended domain. Mature and well-confirmed scientific theories are, at least, approximately true.
Of particular interest in our context is the commitment to truth-conditional semantics given by (2). Such a kind of semantics may come in two different forms, one going back to the Tractatus and the other being given by model-theoretic semantics of predicate logic. The Tractatus semantics rests on a substitutional reading of the quantifiers and does therefore depend on the assumption that every object in the domain is named by an
123
Semantic Challenges to Scientific Realism
19
expression in the object language. This requirement seems unacceptable for scientific realists because our scientific language may not contain a proper name or a definite description for every individual in the domain of scientific inquiry. Hence, the scientific realist must go for the referential reading of the quantifiers and the corresponding modeltheoretic semantics. What notion of truth comes with Tarskian model-theoretic semantics? Admittedly, model-theory by itself introduces no notation for asserting the truth of a sentence. Unlike Frege’s Begriffsschrift, model-theory does not contain an assertion sign. For this reason, model-theoretic semantics is best viewed as interpretational semantics in the sense that it tells us what it is for a sentence to be true only with respect to a particular interpretation of the non-logical symbols of the language and the specification of the universe in which the variables are interpreted. This view has been articulated, among others, by Etchemendy (1999). Likewise, Carnap pointed out that the model-theoretic apparatus of predicate logic is suitable for clarifying the notions of logical truth and logical consequence but makes no contribution to an understanding of the notion of factual truth (Carnap 1973, 98). How then is model-theoretic semantics connected to our representational understanding of truth? For this dimension of truth to be accounted for, the notion of an intended interpretation has been invoked. An intended interpretation of a formal system represents the meaning of the non-logical symbols. It can be made explicit by so-called rules of designation which assign either an intensional or an extensional interpretation to these symbols by means of expressions of a metalanguage, where every intensional interpretation uniquely determines an extensional one. The domain in which the variables are interpreted must be specified, moreover. Once the non-logical symbols of a language L are interpreted, we can say that a sentence / of L is true—in the representational sense of this notion—if and only if it is true in the intended interpretation of L. The conceptual apparatus of interpretational, model-theoretic semantics proved useful, to some extent, to deal with the question of how sentences of ordinary and scientific language get their truth-values assigned to. Let f1 be a function symbol mapping spatiotemporal macroscopic objects to real numbers. Assume f1 is interpreted by the following rule of designation: pf1 q designates the temperature function:
ð1Þ
Let us assume, furthermore, that the individual constants of the object language are interpreted by rules of the following kind: pai q designates ai
ð2Þ
where ai is an expression of the metalanguage being meaningful outright, that means, their expressions do not stand in the need of an interpretation. Moreover, for the individual variables of the object language an assignment of individuals may be assumed. By virtue of such an interpretation, which may be designated by I; atomic formulas such as f1(t1) = t2 have a well-determined truth-value. For we can say that f1(t1) = t2 is true if and only if the value of the temperature function f1 for the argument t1I is t2I ; where t1I and t2I stand for the individuals being assigned to t1 and t2 in the interpretation I respectively. The values of logically complex formulas are determined by the rules for logical connectives and quantifiers.1 1
Davidsonian truth-conditional semantics diverges slightly from (model-theoretic) interpretational semantics in that truth is taken as the basic undefined concept, with the intention to explicate meaning on the basis of the use of the language (Davidson 1984). Hence, in this semantics, truth is not understood as truth
123
20
H. Andreas
Do these considerations shed any light on the truth-value assignment to scientific statements? Obviously, scientific statements are viewed as being represented by sentences of a formalised object language. The object language has a well-determined intended interpretation, which is made explicit by rules of designation given by means of a metalanguage. Such an interpretation hinges on the expressions of the metalanguage being meaningful or having, at least, determinate referents. Meaning or sense is what determines reference. Given the expressions of the metalanguage are meaningful, what makes them to be so? One answer to this question is that it is our conceptual understanding of the metalanguage expressions that confers meaning upon them. Another is to invoke a more objectivist and Platonist conception of meaningfulness, which is independent of our understanding of linguistic expressions. Frege, famously, took this line of thought in his (1918/19). In the eyes of Frege, recourse to an objectivist account of meaning is necessary, since without we would have to think of logic as being concerned with laws of taking true as opposed to laws of truth. This view accords well with the contention of Dummett and Putnam, according to whom assertibility conditions are a more appropriate choice when it comes to account for our understanding of statements, provided no extravagant ontology of Platonic forms or medieval universals is assumed for that purpose. Dummett and Putnam merely disagree with Frege on whether logic should be concerned with laws of taking true or of making assertions as opposed to laws of truth. In sum, there are three options to account for the meaningfulness of the expressions of the metalanguage. First, it is our conceptual understanding of these expressions and only this understanding. Second, it is Platonic forms or universals in virtue of which such expressions are meaningful. Third, it is Platonic forms mediated through our conceptual understanding, where our understanding is directed at such forms. Of course, it remains to clarify what our conceptual understanding consists in if it is not seen to be related to some Platonist account of meaning. In contradistinction to meaning theories, causal theories of reference aim to circumvent the notion of meaning in the account of the assignment of extensions to linguistic expressions. 3 Putnam’s Model Theoretic Argument A collection of arguments in Putnam (1980) and subsequent writings has received the label model-theoretic argument against metaphysical realism. The common pattern of these arguments is as follows. If we adopt truth-conditional semantics for scientific and ordinary language, then the set of our affirmed sentences has numerous models, i.e., numerous interpretations under which these sentences come out as true. These interpretations can be divergent to the extent that the extension of ‘dog’ is the set of all cats. Such variations do not affect the assignment of truth-values. We are therefore not in a position to find out which interpretation of the language for which we adopted truth-conditional semantics
Footnote 1 continued under the intended interpretation. Putnam’s model-theoretic argument, therefore, will not apply directly. It is doubtful, however, whether Davidsonian truth-conditional semantics accords coherently with the realist distinction between truth makers of statements, i.e., truth-conditions that obtain, and statements themselves. Once such a distinction is made, truth does not seem to figure as a primitive notion any more. It is easy to see, moreover, that Dummett’s manifestation and acquisition challenge are an issue for Davidsonian truthconditional semantics.
123
Semantic Challenges to Scientific Realism
21
hypothetically is the correct one. Hence, Putnam concludes, even complete formalisation of our beliefs does not determine a unique intended interpretation of the language we use. Among the sentences that we affirm (without significant disagreement) Putnam distinguishes between theoretical and operational ones. Since it is investigated whether and to what extent these sentences determine an intended interpretation, he is also speaking of theoretical and operational constraints (on the proper intended interpretation of the language). Theoretical constraints comprise axiomatic scientific theories, including mathematics and set theory. Operational constraints are given by a valuation that assigns the correct truth-value to any sentence Rða1 ; . . .; an Þ with R denoting an observable concept and a1 ; . . .; an being interpreted in the domain of observable objects. Observability may be understood in a very liberal sense here.2 Essential to both theoretical and operational constraints is that they are given in the form of a set of sentences and that these sentences are considered true. A concise formulation of the model-theoretic argument, thus, is that the union of the operational and theoretical constraints does not determine a unique intended interpretation of the language. Of course, if we let the term ‘dog’ refer to the set of cats, we realise that we are receiving an unintended interpretation. Awareness of this, however, comes only because we take the terms ‘cat’ and ‘dog’ as having sense and reference by virtue of our use of these terms. This strategy of ruling out unintended interpretations is not available to the scientific realist. For, if meaning is use, assertibility conditions in place of truth-conditions are to figure in the proper semantics of sentence meaning. This claim is adopted, perhaps a bit uncritically, from the work of Dummett. The argument for it goes as follows: (1) The use of linguistic expressions only displays what sentences are affirmed (theoretical and operational constraints) and what patterns of assertion and justification are used. (2) The only available empirical basis for properly learning the language is the use of linguistic expressions. Because of (1) and (2), procedures of justification (of which theoretical constraints may be constitutive) are what our conceptual understanding of linguistic expressions consists in. This, in a nutshell, is Dummett’s acquisition argument for assertibilist semantics, which will be dealt with in more detail in Sect. 6. Putnam is somewhat ambiguous about the relationship between constraints and the use of language. Occasionally he identifies the two, which does not seem to be correct (Putnam 1980, 466). As just indicated, the use of expressions may display something like patterns of justification, a feature that goes beyond theoretical and operational constraints unless we have a full-fledged axiomatic theory from which any pattern of justification can be derived. If the use of language has more aspects than theoretical and operational constraints, then one should complement Putnam’s thesis by the lemma that these aspects do not help either to determine a unique intended interpretation. For this lemma to be true, Dummett’s acquisition argument is decisive. The (1980) exposition of the argument starts with a discussion of Skolem’s paradox: By means of the downward Lo¨wenheim-Skolem theorem it can be shown that there is an interpretation of the axioms of set theory with only a countable domain and only countably many sets that satisfies a sentence saying, intuitively, that there are uncountable sets. In the course of the further exposition, Putnam speaks rather figuratively of Skolemisation, thereby meaning the procedure of varying a given interpretation of a formal language such that certain affirmed sentences remain true. On this reading, the model-theoretic argument does not depend on the validity of the Lo¨wenheim-Skolem theorem, which only holds for first order languages. Hale and Wright (1997), consequently, construe the argument as not 2
Operational constraints are also characterised as comprising all measurements of scientific magnitudes. They may even imply information about the outcome of counterfactual measurements.
123
22
H. Andreas
depending on whether we are dealing with first or higher order formalisations of science and everyday discourse. The model-theoretic argument can be used to show that truth-conditional semantics leads to semantic agnosticism in the sense that we are unable to identify the proper referents of the expressions we use in our language: (1) (2) (3)
(4)
(5) (6) (7)
Truth-conditional semantics is the proper semantics of our language. Knowledge of truth-conditions rests on knowledge of what the proper referents of our linguistic expressions are. Theoretical and operational constraints taken together with other aspects of the use of linguistic expressions do not determine what the proper referents of linguistic expressions are. If a creature lacks supernatural powers, her conceptual understanding of linguistic expressions rests exclusively on operational and theoretical constraints as well as further aspects of the use of these expressions. Human minds lack supernatural powers. Human minds are unable to find out what the proper referents of linguistic expressions are. Human minds do not possess knowledge of realist truth-conditions.
(3) is the kernel of the model-theoretic argument. (6) follows from (1), (2), (3), (4), and (5). (7), obviously, follows from (6) and (2). In a summary of the argument, Putnam himself states that it is the moderate realist position that seeks to preserve the classical notions of truth and reference without postulating nonnatural mental powers which is in deep trouble. The extreme Platonist position and a broadly verificationist account of meaning are ways out of the predicament (Putnam 1980, 464). Putnam’s critique of truth-conditional semantics has one important qualification. This kind of semantics is a sensible tool of explication on condition that we avail ourselves of linguistic expressions in the metalanguage whose meaning is taken to be settled along the lines of a prior verificationist or assertibilist understanding. Having made this qualification, Putnam (1980, 482) concludes: Models are not lost noumenal waifs looking for someone to name them; they are constructions within our theory itself, and they have names from birth. There is no consensus in the literature as to whether Putnam’s argument is successful. The most critical and controversial point is whether or not there are constraints other than theoretical and operational ones that determine the referents of our linguistic expressions. Causal contact with the appropriate referents is one candidate for such a constraint. Putnam’s response is that bringing to bear causal links between linguistic expressions and their presumed referents would add ‘just more theory’, theory which itself has far too many arbitrary, non-intended interpretations to deliver any determination of reference. Therefore, the situation is that whatever the realist offers as an additional constraint, Putnam will reply that this move amounts simply to just adding more theory. This may seem as not playing fair because, for there to be some sensible philosophical discourse at all, some fragment of our language must be exempted from the violent reinterpretations that Putnam demonstrates to be feasible. For this reason, Hale and Wright (1997) suggest that one should not rule out, prior to further proposals, the mere possibility of a successful strategy by the realist. The challenge of semantic agnosticism is in place, however, as long as the (metaphysical) realist does not find a way to demonstrate how additional constraints help to confer a proper understanding of linguistic expressions to human speakers.
123
Semantic Challenges to Scientific Realism
23
Lewis (1984), in an attempt to counter the argument under consideration, took up the notion of an elite class of objects. It is only elite classes that make up the domain of interpretation of expressions designating properties, in his account. This move is intended to delimit or to exclude non-intended models of our theories. Which classes qualify as belonging to the class of elite classes is discovered by physics, where knowledge of physics is taken at face value. Non-physical properties are to be definable in terms of physical ones in order for them to qualify as elite. In this account, one must know a great deal about physics and about reductions to physical concepts if one wishes to acquire an understanding of one’s language. One may suspect, moreover, that the whole language of physics is exempted—by a stipulation—from the violent reinterpretations in the style of the model-theoretic argument. One point, however, should emerge as uncontroversial from our discussion: Scientific realism, as expounded by Psillos (1999), is a kind of metaphysical realism in the sense of Putnam (1980, 1981). This is because scientific realism—in the sense clarified by Psillos— rests on truth-conditional semantics and the availability of an ideal metalanguage forming the basis upon which sentences of scientific language are assigned to truth-values independently of our human strategies for justifying and proving assertions. The proponent of scientific realism therefore must find ways to counter the model-theoretic argument or, at least, concede that the statement of scientific realism depends on there being a successful strategy to counter this argument. This has scarcely been recognised in the literature.
4 A Causal-Descriptive Theory of Reference The model-theoretic argument, as presented here, aims to show that semantic agnosticism is inevitable under realist truth-conditional semantics. That means, we would be ignorant about the proper meanings and referents of our linguistic expressions. Let us have a closer look at the account of reference by Psillos (1999) in order to examine this proposition. This account has two elements: first, assumption of a natural kind structure of the world; second, adoption of a causal-descriptive theory of reference to natural kinds, which is loosely inspired by Lewis (1984). Causal and descriptivist elements are combined in a sophisticated way (Psillos 1999, 296): (1) (2)
A term t refers to an entity x if and only if x satisfies the core causal description associated with t. Two terms t0 and t denote the same entity if and only if (a) their putative referents play the same causal role with respect to a network of phenomena; and (b) the core causal description of t0 takes up the kind-constitutive properties of the core causal description associated with t.
In the exposition of this account it is clearly stated that the core-causal description associated with a theoretical term is to be taken from scientific theories in their state-of-the-art at the time. Two observations are in order here. First, no occult causal connection is postulated to determine reference. Therefore, Putnam’s objection that metaphysical realism assumes a mysterious connection between linguistic expressions and their referents does not apply. Since, however, it is scientific theories that provide information about the causal mechanism of the natural kind property, the ‘just more theory’ sub-argument by Putnam applies in a nonquestion begging way. Second, the account of reference relies on the kind-constitutive properties of the causal role of putative referents being correctly described by any theory that is successfully referring to natural kind properties. (If one uses an incorrect description of an
123
24
H. Andreas
entity in order to refer to that entity, one will not succeed, of course.) What could be meant by asserting or presupposing the correctness of such descriptions? Some historical reflections on the descriptivist theory of reference may help in answering this question. The descriptive theory of reference originates with Frege (1892) and Russell (1905). The differences between the two original accounts are marginal in comparison with the causal theories of reference advanced later by Kripke and Putnam. In particular, it seems fair to say of Frege’s examples of names having distinguished sense and reference that they implicitly rely on the sense being explicable by a definite description. Clearly, the evening star is the heavenly body that appears first in the evening. Likewise, we have a description of how to construct the intersection of the perpendicular bisectors of the sides in the equilateral triangle. There are different semantic readings a description can have in a descriptivist account of reference, be it partially or fully descriptivist. There are cases where the definite description has the status of a conceptual truth. That the evening star is the heavenly body that appears first in the evening does seem to rest on a convention entrenched in the use of the expression ‘evening star’. In other cases—most of Russell’s examples in his (1905) belong to this class— the use of a definite description does not imply the truth of this description; it is only the uniqueness condition that is implied. Claiming of an individual being definitely described by the property F that it has the property G is expressed in Russell (1905) as follows: 9xðFðxÞ ^ GðxÞ ^ 8yðFðxÞ ! x ¼ yÞÞ
ð3Þ
Finally, a definite description can be used as a factual statement in a descriptivist or causal-descriptivist account of reference. In sum, a description can have the following semantic readings. First, it can be read as a conceptual truth. Second, its use may imply that there is exactly one object satisfying it— this is Russell’s account, and third, a description may be interpreted as a factual statement. Which semantic reading of descriptions should we attribute to Psillos’s causal descriptivist account of reference? To answer this question, let us deal with a particular example. Arguably, the following statement was—at a certain stage of phenomenological thermodynamics—part of the causal description of the kind-constitutive properties associated with the scientific concept of temperature: If the temperature of an amount of mercury rises, ð4Þ then the volume of this amount of mercury will extend proportionally. Now, semantic realism is adopted by Psillos (1999, 11–15), which amounts to a truthconditional understanding of scientific statements in the correspondence sense of truth. According to this view, the sentence ‘If the temperature of an amount of mercury rises, then the volume of this amount of mercury will extend proportionally’ is true if and only if it holds that, if the temperature of an amount of mercury rises, then the volume of this amount of mercury will extend proportionally. Note that, for rather obvious and above explained reasons, the right-hand side of this biconditional must not be read along the lines of a verificationist or assertibilist account of meaning. (The left-hand side of the biconditional names nevertheless a sentence of scientific language as used by human scientists.) Nor does the right-hand side of the biconditional merely state conditions under which the proposition (4) is affirmed by speakers. Hence, the above description of causal properties of the temperature function is a factual statement whose truth-value depends on what the world is like quite independently of our theorising. From this it follows that the causaldescriptive account of reference hinges on the factual truth of the descriptions we are using in referring to scientific concepts and entities. If these descriptions were not true, we could not successfully refer to the scientific concept described.
123
Semantic Challenges to Scientific Realism
25
What reason do we have to suppose that descriptions used according to Psillos’s causaldescriptive theory of reference are true? The correct answer seems to be: none. For in order to assess the truth of a description—of causal or non-causal properties of a scientific concept—there would have to be a means of referring to this concept without using the stock of descriptions whose truth is to be evaluated. In other words, assessment of truth of a description requires direct acquaintance with the entity being described. But, by assumptions made in the work of Russell and in Psillos (1999), there is no direct acquaintance with theoretical scientific concepts. We have no knowledge of theoretical concepts by acquaintance. Hence, we can say that the truth of definite descriptions we are using to refer to scientific concepts does—in the realist picture—in principle escape our means of recognition. From this we can conclude that the challenge of semantic agnosticism stays in place because the truth of descriptions is constitutive of our means of referring to scientific concepts in Psillos’s account.
5 Another Semantic Reading of Descriptions Even though semantic realism commits Psillos to taking descriptions of causal properties as factual statements, it is highly illuminating to see what the consequences would be of siding more closely with the style of the original descriptivist account of reference by Russell (1905). If the scientific realist were to take this alternative, proposition (4) would have to be converted to an existential statement: There is a natural kind function X1 such that if X1 of an amount of mercury rises, then the volume of this amount of mercury will extend proportionally.
ð5Þ
Note that any proposition /(f1) concerning the temperature function would have to be translated, then, into a statement of the following form (see Andreas 2010): There is a natural kind function X1 such that if X1 of an amount of mercury rises, then the volume of this amount of mercury will extend proportionally and wðf1 =X1 Þ and /ðf1 =X1 Þ:
ð6Þ
w(f1) is the conjunction of sentences that further characterise the natural kind property temperature if the description of this property by (4) should not be exhaustive. w(f1/X1) and /(f1/X1) are the sentences that one obtains when all occurrences of f1 are replaced with X1. If it is the case that the sentences describing the causal properties of the temperature function have occurrences of theoretical property terms, then these terms need also to be replaced by higher order variables. The expression ‘volume’ in (4) clearly qualifies as such a term since measurement of volume rests on the introduction of a spatiotemporal metrics. Moreover, mercury should qualify as a clear-cut example of a natural kind. Taking these additional qualifications into account leads—in our semiformal exposition—to the following formulation of the assertion of /. There are natural kind functions X1 and X2 and a natural kind property Y1 such that if X1 of an amount of Y1 rises, thenX1 of this amount of Y1 will rise proportionally and wðf1 =X1 ; f2 =X2 ; P1 =Y1 Þ and /ðf1 =X1 Þ:
ð7Þ
f2 designates the volume function with spatiotemporal objects as arguments and P1 the property of being mercury. It is assumed that /(f1) has neither occurrences of f2 nor of P1.
123
26
H. Andreas
One may finally take the uniqueness condition of Russell’s theory of description into account in the style of (3), which is not difficult to accomplish. In an attempt to show how theoretical terms can be defined, Lewis (1970) introduced such a uniqueness condition using what he called the Carnap-Ramsey sentence. This approach may only then succeed if the domain of interpretation of the higher order variables is restricted to extensions of natural kind properties and functions. Otherwise, theoretical and observational constraints would certainly not uniquely describe the extension of a theoretical concept. This is because our measurements are insufficient to determine the extension of a scientific quantity completely (cf. Schurz 2005). If the assumption of a natural kind structure is not made and a corresponding restriction on the domain of interpretation of the higher-order variables is not introduced, the Ramsey account must be read as using indefinite descriptions of theoretical concepts. This reading has been argued for by Bohnert (1968). Ramsey (1950) himself pointed out that truth-value gaps may well occur at the level of theoretical atomic sentences. From this it follows that, in his view, the Ramsey sentence uses only indefinite descriptions of theoretical concepts. Dropping the assumption of a natural kind structure and a corresponding restriction on the domain of interpretation of the higher-order variables would require the following modification of (7): There are functions X1 and X2 and a property Y1 such that if X1 of an amount of Y1 rises, then X1 of this amount of Y1 will rise proportionally and wðf1 =X1 ; f2 =X2 ; P1 =Y1 Þ and /ðf1 =X1 Þ:
ð8Þ
To sum up, the adoption of a strict Russellian style of using descriptions in a causaldescriptive account of reference leads to the Ramsey view of scientific theories. This result does not depend on whether or not the assumption of a natural kind structure is made. On this assumption, which Psillos adopts, the causal-descriptive account may yield definite descriptions of theoretical concepts, whereas one has to be content with indefinite descriptions if this assumption is dropped. In a certain sense, the causal-descriptive account of reference by Psillos is wholly descriptivist with the qualification that causal properties of natural kinds are what matters in describing natural kinds. As is well known, the Ramsey view has been proposed as a formal clarification of structural realism. The corresponding expositions commonly do not make much use of natural kinds. One may therefore hold that the Ramsey approach to scientific theories with assumption of a natural kind structure is an intermediate position between Ramsey-style structural realism and scientific realism. This view squares well with Psillos (1999, 68n). For a recent exposition and defence of structural realism in the Ramsey-style see, e.g., Cruse (2005). One central argument against structural realism in the Ramsey-style is the charge of trivial realisation (Ketland 2004). 6 The Acquisition and the Manifestation Argument Having dealt at some length with Putnam’s model-theoretic argument, let us move on to Dummett’s acquisition and manifestation argument. The target of the latter arguments is again truth-conditional semantics, as was pointed out above. The commitment to this kind of semantics by the scientific realist has far-reaching consequences. The considerations of this and the next section merely confirm results obtained in the preceding ones. The acquisition argument may best be encapsulated in the claim that truth-conditional semantics cannot be extended to a general theory of meaning which accounts for how we
123
Semantic Challenges to Scientific Realism
27
come to properly understand the meaning, i.e., the Fregean sense, of our linguistic expressions (Dummett 1991, 340). Language learning, so (Dummett 1978, 1991) argues, is guided by assertibility conditions and not by truth-conditions. To this, the realist can still reply that assertibility conditions converge with truth-conditions, or that our methods of justifying statements rest on truth-conditions of these statements. Then, the debate continues, there is, for the antirealist, divergence between truth-conditions and assertibility conditions for non-decidable statements, i.e., statements with potentially evidence-transcendent truth-conditions. For this reason, the antirealist maintains an antirealist position only with respect to a disputed class of statements, such as mathematical statements about infinite domains, statements concerning spatiotemporally very remote events or statements about theoretical entities. The manifestation argument does not start with a particular kind of semantics but with an observation concerning the linguistic behaviour of speakers. Suppose Wittgenstein is correct in holding that meaning is use. Then the question arises whether the understanding displayed in the use of language can be made to harmonise with truthconditional semantics (Wright 1993, 16–23). Use of language and particularly our ways of affirming and negating sentences are, according to the antirealist, guided by assertibility and not by truth-conditions, and the two kinds of explanations of sentence meaning diverge in the realm of statements with potentially evidence-transcendent truthconditions. Hence, truth-conditional semantics is not acceptable as a global theory of meaning. Both the acquisition and the manifestation argument are justifiably called challenges to the semantics of realism as they are no disproof of truth-conditional semantics. The debate on this is ongoing, and several strategies have been developed by either side to gain some victory over the other. A promising defence strategy by the realist is to insist on the compositionality of sentence meaning and then to show that we can grasp evidencetranscendent truth-conditions through semantic knowledge of the linguistic components of the corresponding sentences and an understanding of the principles of compositionality. This consideration will further concern us below.3 The labels realism and antirealism have become entrenched in the philosophy of science. It should be noted, however, that the semantics of scientific language has, aside from the language of mathematics, received surprisingly little attention in assessment of the manifestation and the acquisition argument being at the core of the antirealist position. It does therefore seem worth taking a look at the semantics of theoretical concepts, an all too prominent subject matter in philosophy of science, from the perspective of these arguments. The primary focus will be on the acquisition argument.
7 The Problem of Theoretical Terms In this section, the problem of theoretical terms—in the sense of Sneed (1979, 31–40)—will be presented as a Dummettian acquisition argument. Sneed observed that for almost any scientific theory containing some mathematical apparatus the following holds true:
3
Dummett himself also advocates the principle of compositionality but under an intuitionistic reading of the logical constants.
123
28
(1)
(2)
H. Andreas
There is at least one predicate or function symbol t in the language of the scientific theory T such that the extension of t can only be determined with the help of some axiom of T. The truth-values of those axioms which can be used for the determination of t are determinable if and only if the extension of t is known.
In short, the problem of theoretical terms consists in the mutual dependency between the extension of a non-logical symbol t belonging to the language of a theory T and the truthvalues of one or several axioms of T. On the one hand, we are to know the extension of t in order to find out whether the axioms of T are true. On the other hand, it is simply impossible to determine the extension of t without assuming the truth of some sentence being an axiom of T in advance. This, in essence, is what Sneed expounded as the problem of theoretical terms.4 Again, truth-conditional semantics may be cited as the culprit being responsible for the problem. Suppose this semantics gives a correct account of meaning for scientific language. Note, then, furthermore that our methods of determining the extension of scientific terms simply rest upon using some axiom of T. Take, for example, the above mentioned temperature function. There are several methods of measuring this function available. We have liquid, gas, and resistance thermometers, to name only the most entrenched ways of measuring temperature. In using one of these thermometers, one has—implicitly—to assume that a certain law-like proposition holds for the underlying process of measurement. In the case of a liquid thermometer, it is the constant ratio of the spatial extension of a liquid to the rise of temperature. Gas thermometers are based on the ideal gas law, whereas resistance thermometers rely on a law-like proposition about the electrical resistance and the temperature of the conducting material. Similar considerations apply to other physical quantities, such as force, mass, air pressure, the intensity of the electromagnetic field and so on. The challenge for the realist now is to show that our procedures of determining physical quantities is truth-conditionally well-founded, that means, she has to show that the sentences that we use to determine physical quantities are true according to the standards of truth-conditional semantics. This is not possible, however, since a truth-conditional evaluation of these sentences requires knowledge of the extension of theoretical concepts in advance, i.e., prior to the very act of measurement. Note, furthermore, that the ideal gas law or the law-like correlation between temperature and volume for liquids cannot be interpreted as defining temperature in the strict sense of a definition. These laws have—in the context of further sentences—empirical consequences and therefore do not qualify as conservative extensions of the language. By contrast, a definition must be conservative in the sense that it does not extend the valid consequences among sentences that have only occurrences of undefined symbols. The lesson that can be drawn from Sneed’s analysis may be seen to be the following. Reference to theoretical concepts is always based on a prior assignment of truth-values to sentences with occurrences of the corresponding concept terms. For this reason, the novice will be introduced to scientific concepts through learning axioms about them. These 4
It would have been more precise to say that the value of t cannot even partially be determined without accepting some application of T to an empirical system as being successful without any justification, where such an application implies the statement that some axiom of T holds true for this system. But to fully appreciate the content of this formulation, it would have been necessary to give a more fine-grained exposition of the manner in which Sneed employs techniques of the semantic approach to scientific theories. The above formulation presents the core of the problem of theoretical terms.
123
Semantic Challenges to Scientific Realism
29
axioms form the conceptual foundation of our measuring methods. The resulting order of understanding contrasts with truth-conditional semantics, in which we would have to know first what temperature is and could only then understand what it means that, if the temperature of an amount of mercury rises, then the volume of this amount of mercury extends proportionally. If the truth-conditional order were to be adopted for the purpose of learning scientific language, we would never come to know what temperature is. In particular, we would have no reason to suppose that it is temperature that we are measuring with our thermometers. Thus, we formulated an acquisition argument against truth-conditonal semantics, one that is particularly concerned with scientific language. As a solution to the problem of theoretical terms, Sneed (1979) proposed to adopt the Ramsey view of scientific theories, that is, to replace the theory with its Ramsey sentence. Why does this amount to a solution to the problem under consideration? Now, by the Ramsey sentence it is only asserted that there is an extensional interpretation of the theoretical concept symbols under which the axioms of T come out as true but not that these axioms are true in the sense of a truth-conditional evaluation. No assumption of a natural kind structure of the world is made and no corresponding restriction on the range of the higher-order variables is introduced by Sneed. So, there is another argument suggesting a Ramsey view of scientific theories and a corresponding deviation from the semantics of metaphysical realism, viz., the observation that the problem of theoretical terms arises in this semantics but does not in the Ramsey view. It was claimed in the introduction that the acquisition and the manifestation argument apply to scientific language to an even higher degree than to other fields of discourse. What is the basis of this claim? As indicated above, compositionality may be the key to understanding statements with evidence-transcendent truth-conditions: First, we master a fragment of our language, where truth- and assertibility conditions converge. Then, we enlarge our semantic capacities through capturing principles of compositionality such that knowledge of evidence-transcendent truth-conditions comes within reach of our understanding (Wright 1993, 16). This strategy does not seem to work for scientific languages, aside from the language of (non-applied) mathematics. For as careful investigations of our measuring procedures show, the semantics of scientific language may not be compositional. At least, our patterns of grasping the meaning of scientific concepts do not conform to the principles of compositionality because we start learning general sentences about scientific concepts and then move on to determine their extension (cf. Schurz 2005). This point matters since it is fair to require any theory of meaning to be in harmony with a proper theory of understanding.
8 Internal Realism Putnam himself recommended an assertibility account of meaning as a way out of the semantic agnosticism that proved to be an implication of metaphysical realism. In advocating such a view, he walked in the footsteps of Dummett but suggested we should not be as revisionary as Dummett would have liked. Intuitionistic logic is not adopted in place of classical logic by Putnam. Assertibility conditions are rather a weakened form of justification conditions, one that does not require us to be in a position to verify conclusively or to refute any meaningful statement. Admittedly, Putnam did not get too far in making precise how the weakening of justification conditions ought to be understood and what it involves.
123
30
H. Andreas
Internal Realism became later on in Putnam’s writings the label for the view that he suggested as an alternative to metaphysical realism. Two points are central to this view: First, reference is always relative to a conceptual scheme; second, the classical notion of truth is replaced with the notion of justification under ideal epistemic conditions (Putnam 1981). What are the ideal epistemic conditions under which it would be feasible to justify the proposition that, if the temperature of an amount of mercury rises, then the volume of this amount increases proportionally? I do not think that internal realism is much better off here than metaphysical realism. Replacing the classical notion of truth with the notion of justification under ideal epistemic conditions suggests in the case of proposition (4) that there is a means of referring to the concept of temperature without taking some statement of thermodynamics, broadly conceived, as a conceptual truth. Even the semantics of intuitionistic logic, which Dummett recommends as an alternative to truth-conditions, does not seem to have the resources for making reference to theoretical concepts intelligible. In this semantics it holds that the assertion of a general sentence is semantically based on the justification of all the particular instances of this sentence. To give such a justification in the case of proposition (4), we would have to refer to the concept of temperature without taking some statement of thermodynamics as conceptual truth, which seems not feasible according to the central arguments presented here. Even though the problem of referring to theoretical concepts can be given the form of a Dummettian acquisition argument, the semantics of intuitionistic logic is no solution to it. The view to which I would like to point to tentatively has it that reference to scientific concepts always rests on taking some axiom of a scientific theory as a conceptual truth. Such truths form the conceptual schemes that enable us to refer to theoretical concepts, where propositions expressing them may be admitted to have some empirical content in the context of the whole theory. Hence, these propositions are constitutive of reference to theoretical concepts and holistically defeasible at the same time. As I understand Poincare´’s philosophy of science, his notion of a convention has precisely such a twofold function.
9 Conclusion In the eyes of Dummett and Putnam, there is an unbridgeable gap between meaning and understanding in realist semantics. Their arguments are thus construed as showing that realist semantics leads to semantic agnosticism, i.e., ignorance about the proper meanings and referents of our linguistic expressions. Psillos’s causal descriptive theory of reference proved not to have the resources to counter these arguments, provided descriptions are taken as factual statements. There is another semantic reading of descriptions, however, one that is more faithful to Russell’s original descriptivist theory of reference. There, descriptions are not factual statements but used to pick out an entity in a domain. Pursuing this semantic reading of descriptions allowed us to make a transition from the critical part to a more constructive part of the paper. In the latter part, it could be shown that a strict Russellian reading of descriptions leads to the Ramsey account of scientific theories. This account is by far more promising when it comes to explaining how human scientists can ever grasp the meaning of scientific concepts, for it does not presuppose that theoretical concept terms have reference independently of the theoretical context in which such terms are
123
Semantic Challenges to Scientific Realism
31
introduced.5 On the assumption that the Ramsey account explicates (epistemic) structural realism, our analysis of the semantics of descriptions in a causal-descriptive theory of reference presents a novel case for (epistemic) structural realism. Descriptions may finally be taken as conceptual truths in a weak sense, not implying a priori knowledge of metaphysically necessary truths. Hence, there is a third semantic reading of descriptions. The corresponding semantic convention view of scientific theories could only be adumbrated here. Its potentialities in face of acquisition considerations remain to be explored. Acknowledgments I am grateful to an unknown reviewer for very helpful comments on a first draft of this paper.
References Andreas, H. (2010). A modal view of the semantics of theoretical sentences. Synthese, 174(3), 367–383. Bohnert, H. G. (1968). In defense of Ramsey’s elimination method. The Journal of Philosophy, 65, 275–281. Carnap, R. (1973). Introduction to symbolic logic and its applications. New York: Dover Publications. Cruse, P. (2005). Scientific realism, Ramsey sentences and the reference of theoretical terms. Studies in History and Philosophy of Science, 36(3), 557–576. Davidson, D. (1984). Enquiries into truth and interpretation. Oxford: Oxford University Press. Dummett, M. (1978). Truth and other enigmas. London: Duckworth. Dummett, M. (1991). The logical basis of metaphysics. Cambridge, Mass: Harvard University Press. Etchemendy, J. (1999). The concept of logical consequence. Stanford: CSLI Publications. ¨ ber Sinn und Bedeutung. Zeitschrift fu¨r Philosophie und philosophische Kritik, 100. Frege, G. (1892). U Frege, G. (1918/19). Der Gedanke. Beitra¨ge zur Philosophie des deutschen Idealismus, 2, 58–77. Hale, B., & Wright. C. (1997). Putnam’s model-theoretic argument against metaphysical realism. In: B. Hale & C. Wright (Eds.), A companion to the philosophy of language. Oxford: Blackwell. Ketland, J. (2004). Empirical adequacy and ramsification. British Journal for Philosophy of Science, 55, 287–300. Lewis, D. (1970). How to define theoretical terms. Journal of Philosophy, 67, 427–446. Lewis, D. (1984). Putnam’s paradox. Australasian Journal of Philosophy, 62(3), 221–236. Psillos, S. (1999). Scientific realism. London: Routledge. Putnam, H. (1980). Models and reality. Journal of Symbolic Logic, 45(3), 464–482. Putnam, H. (1981). Reason, truth and history. Cambridge: Cambridge University Press. Ramsey, F. P. (1950). Theories. In: R. B. Braithwaite (Ed.), The foundations of mathematics and other logical essays (pp. 212–236). New York: Humanities Press. Russell, B. (1905). On denoting. Mind, 14, 479–493. Schurz, G. (2005). Semantic holism and (non-)compositionality of scientific theories. In: M. Werning, E. Machery, & G. Schurz (Eds.), The compositionality of meaning and content (Vol. I, pp. 271–284). Frankfurt: Ontos-Verlag. Sneed, J. (1979). The logical structure of mathematical physics, 2nd ed. Dordrecht: D. Reidel Publishing Company. Wright, C. (1993). Realism, meaning and truth, 2nd ed. Oxford: Blackwell.
5
As for the semantics of non-theoretical concepts and sentences, it is advisable to adopt a verificationist account of meaning in order to block the model-theoretic argument by Putnam.
123