Cruse - Ramsey Sentences, Structural Realism And.pdf

Studies in History and Philosophy of Science Stud. Hist. Phil. Sci. 36 (2005) 557–576 www.elsevier.com/locate/shpsa

Ramsey sentences, structural realism and trivial realization Pierre Cruse Department of Philosophy, KingÕs College London, Strand WC2R 2LS, UK Received 21 December 2004; received in revised form 18 April 2005

Abstract Several recent authors identify structural realism about scientific theories with the claim that the content of a scientific theory is expressible using its Ramsey sentence. Many of these authors have also argued that so understood, the view collapses into empiricist anti-realism, since an argument originally proposed by Max Newman in a review of Bertrand RussellÕs The analysis of matter demonstrates that Ramsey sentences are trivially satisfied, and cannot make any significant claims about unobservables. In this paper I argue against both of these claims. Structural realism and Ramsey sentence realism are, in their most defensible versions, importantly different doctrines, and neither is committed to the premises required to demonstrate that they collapse into anti-realism.
2005 Elsevier Ltd. All rights reserved. Keywords: Ramsey sentence; Structural realism; Structuralism; Scientific realism.

1. Introduction A line of argument which has gained some currency in recent literature runs as follows. Structural realism about scientific theories involves the claim that while we do not have knowledge of the natures of unobservable entities, we can nevertheless E-mail address: [email protected] (P. Cruse). 0039-3681/$ - see front matter
2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.shpsa.2005.07.006

558

P. Cruse / Stud. Hist. Phil. Sci. 36 (2005) 557–576

know their structural features.1 Structural realists argue that their position can provide a synthesis of both scientific realism and anti-realist empiricism, and retains the advantages of each view. The question then arises of how we identify the ÔstructuralÕ features of the world of which we can allegedly have knowledge. An idea proposed originally by Grover Maxwell (1970a,b) is that we can elucidate the notion of structure using Ramsey sentences, formed by replacing the theoretical terms in the theory by existentially quantified second-order variables. According to Maxwell, Ramsey sentences express precisely the structural claims of scientific theories, so to be a structural realist is to believe that the cognitive (that is, potentially knowledge-expressing) content of a scientific theory is expressed in its Ramsey sentence. However, the resulting position is subject to a crippling objection deriving originally from an argument originally proposed by Max Newman in a review of Bertrand RussellÕs The analysis of matter (Newman, 1928). The conclusion of this argument is that Ramsey sentences are trivially satisfied, in the sense that if what they say about observables is true (and a cardinality requirement is satisfied) then they are guaranteed to be true. This appears to entail that structural realism collapses into the anti-realist position that only the observable is knowable, thwarting the realist ambitions of its proponents. Several recent writers on structural realism have told some more or less qualified version of this story, to the extent that it has almost become a consensus view on the issue (see for example Demopoulos & Friedman, 1985, Ladyman, 1998, Psillos, 1999, 2001, and most recently Ketland, 2004; to my knowledge only Zahar, 2001 has contested the efficacy of Newman-style reasoning in undermining structural realism, at least in its Ramsey sentence version). However, I believe that the story rests on two fundamental misconceptions. The first is that Newman-style reasoning necessarily demonstrates that Ramsey sentence-cum-structural realism collapses into some form of anti-realism. The second is that structural realism should be identified with the claim that the content of a theory is expressed in its Ramsey sentence in the first place. In this paper I will look in detail at these two issues. In the first part of the paper I will ask whether anything like NewmanÕs trivialization proof genuinely does refute either structural realism (henceforward SR), or the claim that the content of a theory is expressed in its Ramsey sentence (henceforward, RS). My answer will essentially be that it does not. More precisely, I will show that there are different motivations for holding SR and RS, which result in significantly different versions of each position. Some of these resulting views, I will concede, are refuted by the trivial realization argument. However, I will claim that the better motivated and more plausible versions of each position are not, since they are 1 I restrict my attention in this paper to Ôepistemic structural realismÕ which holds that our knowledge of the world is exhausted by knowledge of its structural features, while admitting that there are further ÔnatureÕ-constituting features of which we do not know. An ontic version of structural realism, on which the world is thought to be entirely constituted by its structural features has also been proposed (see e.g. Ladyman, 1998): I will have cause to mention this view below, but it will not primarily be under discussion in this paper.

P. Cruse / Stud. Hist. Phil. Sci. 36 (2005) 557–576

559

not committed to the premises required to get the argument off the ground. In the second part of the paper I will question whether Ramsey sentence realism and structural realism ought to be identified with one another. My claim here will be similar. Some versions of SR and RS, I will concede, are equivalent, but these are the less plausible versions of each doctrine. The more plausible and better motivated versions of each view are importantly different, and have distinct considerations in their favour.

2. Ramsey sentences and trivial realization I will begin simply by outlining the argument required to show that Ramsey sentences are trivially realized. We will then return to look in more detail at its philosophical significance. The relevant argument derives originally from a reply which mathematician Max Newman made to the position Bertrand Russell had developed in his book The analysis of matter (Russell, 1927). In this work Russell had argued for a form of structural realism, which asserted that Ôwhenever we infer from perceptions, it is only structure which we can validly infer; and structure is what can be expressed by mathematical logic, which includes mathematicsÕ (ibid., p. 254). NewmanÕs suggestion was that if you hold that scientific knowledge is exhausted by the claim that there exist relations which have a certain structure—in other words, by a Ramsey sentence claim—then this reduces scientific knowledge to knowledge of cardinality. This is because, in NewmanÕs words, . . . no important information about [an] aggregate A, except its cardinal number, is contained in the statement that there exists a system of relations, with A as field, whose structure is an assigned one. For given any aggregate A, a system of relations between its members can be found having any assigned structure compatible with the cardinal number of A. (Newman, 1928, p. 144) As far as I am aware, no-one has questioned the fact that NewmanÕs basic point— that relations with any given structure exist in any large enough domain—is correct. However, the exact moral we should draw from NewmanÕs reasoning has been disputed. William Demopoulos and Michael Friedman, James Ladyman and Stathis Psillos (Demopoulos & Friedman, 1985, Ladyman, 1998, Psillos, 1999) have all claimed that it demonstrates that RS reduces to the claim that we know nothing about the unobservable other than its cardinality. In the appendix to (Zahar, 2001), co-authored with John Worrall, Elie Zahar has contested this, on the grounds that NewmanÕs reasoning takes no account of the distinction between observational and theoretical terms on which Ramsey sentences rely. Most recently, Jeff Ketland has investigated the conclusion Newman style reasoning establishes in relation to Ramsey sentences, and shows that ZaharÕs point does not get the structural realist out of trouble (Ketland, 2004). As I believe KetlandÕs reasoning to be definitive here, I will begin by outlining the conclusion he draws, before commenting briefly on how it relates to the previous claims mentioned.

560

P. Cruse / Stud. Hist. Phil. Sci. 36 (2005) 557–576

Ketland considers an interpreted, second-order language, L2(ÆOiæ, ÆMiæ, ÆTiæ). This language contains three sorts of non-logical predicates: ÔÆOiæ = (O1,O2, . . .) (the ÔobservationalÕ relations), ÆMiæ = (M1,M2, . . .) (the ÔmixedÕ relations) and ÆTiæ = (T1,T2, . . .) (the ÔtheoreticalÕ relations)Õ. These predicates are individuated as follows. First we suppose that the universe of discourse is divided into two disjoint and exhaustive classes of objects—ÔobservableÕ and ÔtheoreticalÕ. An observational predicate is then a predicate whose field ranges only over the observable objects, a theoretical predicate one whose field ranges only over theoretical objects, and a mixed predicate is one that relates observable and unobservable objects. The language also contains first and second order variables. The first-order variables are divided into two sorts, ranging over observable and theoretical objects respectively. The second-order variables are divided into three sorts, ranging over observational (X1,X2, . . .), theoretical (Z1,Z2, . . .), and mixed (Y1,Y2, . . .), relations respectively. We can then write a scientific theory expressed in L2 as H(O, M, T). Finally, Ketland forms the Ramsey sentence RðHÞ of the theory by replacing the theoretical and mixed predicates in this theory with corresponding second-order variables, and then forming the existential closure of the resulting formula, giving, $Y$ZH(O, Y, Z). Ketland then considers a notion of empirical adequacy similar to that defined by van Fraassen (van Fraassen, 1980). According to van Fraassen, a theory is empirically adequate when it has Ôat least one model that all the phenomena fit insideÕ (ibid., p. 12). Ketland elucidates this by first defining the empirical reduct of a model as the result of subtracting from the model the theoretical domain, together with all its mixed and theoretical relations. A theory is then empirically adequate exactly when it has a full model that is empirically correct, where a model is empirically correct when its reduct to the empirical domain is isomorphic to the empirical reduct (formed similarly) of the world. On this basis Ketland shows that NewmanÕs basic point—that in any domain with a sufficiently high cardinality, there exist relations with a given structure—can be used to prove that: (1) RðHÞ is true if and only if H has a full model M which is T-cardinality correct and empirically correct, where a model is T-cardinality correct when its theoretical domain is of the same cardinality as the theoretical domain of the world. Essentially, (1) amounts to the claim that if the theory is empirically adequate, and the cardinality of the world is great enough, its Ramsey sentence is true (and vice versa). This appears to demonstrate that the claim that our scientific knowledge is expressed in a Ramsey sentence collapses into something like van Fraassen-style constructive empiricism. KetlandÕs result relates to its predecessors as follows. Demopoulos and Friedman (Demopoulos & Friedman, 1985) argue that if a Ramsey sentenceÕs observational consequences are all true, then if a cardinality requirement is satisfied, the Ramsey sentence is true (Ladyman, 1998, makes a similar claim). This is what Zahar denies (Zahar, 2001, pp. 240–241). However, Ketland shows that while a Ramsey sentence does not strictly follow from the claim that all its observational consequences are true, it does follow from the marginally stronger claim that it is empirical adequate, as defined above, so ZaharÕs denial does not specifically address this point (Psillos,

P. Cruse / Stud. Hist. Phil. Sci. 36 (2005) 557–576

561

1999, pp. 61–69 also claims only that the truth of a Ramsey sentence only follows from an empirical adequacy claim). I will therefore assume from now on that (1) is a definitive statement of what NewmanÕs basic reasoning can establish here. The issue I wish to look at is how this result relates to the philosophical motivations for holding structural realism and Ramsey sentence realism.

3. Ramsey sentences and the observational/theoretical distinction In order to do this, it will first be useful to make an observation about how KetlandÕs proof of (1) works. One salient feature is how it asks us to form the Ramsey sentence of a theory. The basic idea is that Ramsification involves replacing the mixed and theoretical relations in a theory with second order existentially quantified variables. However, the precise outcome of this procedure depends on how we taxonomize relations as observational, mixed or theoretical. As Ketland notes in a footnote, the division he describes does not correspond to any way in which we might intuitively class relations. For example, consider the relation denoted by the predicate Ôlarger thanÕ. On KetlandÕs taxonomy, there is no such single relation; there are three. First, there is the relation we might call observably larger than, which ranges entirely over observable objects. Second there is the relation we might call unobservably larger than, which ranges entirely over unobservable objects. Third, there is the relation we might call miscellaneously larger than, which applies to all and only pairs of objects such that the first is observable, the second unobservable, and the first larger than the second. On KetlandÕs terminology, only the first class of relations—those which range entirely over observable objects—count as observable, and thus, only relations in this class are left untouched in forming the Ramsey sentence of a theory. For reasons which will become clear later, I will call this the strong version of the observational-theoretical (O/T) distinction. Note that in order for (1) to have anti-realist implications, we cannot class a greater number of relations as observable than is allowed by the strong O/T distinction. The reason is that in (1) empirical adequacy is defined in terms of the fields of observable relations. A theory is empirically adequate, essentially, if it has a model whose restriction to the empirical domain is isomorphic to the restriction of the world to the empirical domain, where the empirical domain is the domain over which the observable relations range. Thus, if ÔobservableÕ relations range over unobservable objects as well as observable ones, it would follow that Ôempirical adequacyÕ, doesnÕt have its intended sense of Ôtruth about what is observableÕ: it would mean Ôtruth about what is observable and what is unobservableÕ. But this is just truth as a realist would conceive it. Thus, unless Ramsification (at least) removes all but the strongly observable relations—those relations that only range over observable objects—(1) has no anti-realist implications whatsoever. I believe there is a fairly obvious point behind this, which holds beyond the particular details of KetlandÕs argument. I donÕt think anyone would deny that if we have a stock of predicates available which apply determinately to unobservable

562

P. Cruse / Stud. Hist. Phil. Sci. 36 (2005) 557–576

entities, then it is possible to make nontrivial existential assertions about the unobservable. There is no more reason to deny this than to deny that we can make nontrivial existential claims in general. Now a Ramsey sentence is simply an existential assertion in which, for whatever reason, the predicates used in making it are classed as ÔobservationalÕ or Ônon-theoreticalÕ. So if our motivation for deeming such predicates observational or non-theoretical does not preclude them from applying determinately to unobservable entities, there is no reason to think that Ramsey sentences cannot make nontrivial existential assertions about the unobservable. Any proof that they do must tacitly be assuming that observational predicates do not apply determinately to unobservables; as indeed KetlandÕs proof does. To assess the significance of (1), then, we need to look in more detail at the theoretical motivation for forming Ramsey sentences according to the strong version of the O/T distinction, which restricts the range of observational predicates to observable objects. Only if there is any reason for thinking that someone who wishes to use Ramsey sentences to sustain some form of scientific realism is committed to this distinction will the proof hit its target.

4. Ramsey sentences and concept empiricism: the ÔupwardÕ path In a recent discussion of SR, Stathis Psillos has drawn a useful distinction between two arguments for SR, the ÔupwardÕ and ÔdownwardÕ paths (Psillos, 2001). The Ôupward pathÕ to SR refers to any argument that begins with empiricist assumptions about what can be directly referred to, and uses these to motivate some conception of the content of whole theories. ÔThe downward pathÕ denotes any argument that begins with a realist conception of the content of whole theories and then motivates the ÔsubtractionÕ of some of that content to get structural realism. Much the same distinction applies to arguments for RS. One can either start by holding that we can only directly refer to observational properties and use Ramsey sentences as a means to build ÔupwardsÕ to a conception of the content of whole theories, or one can start with a realist conception of the content of whole theories and use Ramsey sentences as a means to subtract ÔdownwardÕ to some conception of their genuinely cognitive or knowledge-expressing content. I will therefore adopt PsillosÕ terminology, and look at whether the upward and downward paths to RS and SR are likely to motivate the strong version of the O/T distinction. Let us begin by looking at the upward path to RS. There is, I think, a very intuitive case for RS along these lines. It goes something like this. There are properties we can observe, like redness and squareness, and properties we cannot, like strangeness and being a DNA molecule. We can imagine a story of how we might get to refer to properties like redness and squareness. It will presumably involve something like the fact that we can perceive when things are red and square, and can thus think ÔredÕ or ÔsquareÕ when something red or square is present to us. However whatever we say here wonÕt apply to strangeness or being a DNA molecule, since we can never perceive (directly, at any rate) that something has strangeness or is a DNA molecule, so we need a different explanation of how we get to talk about these things. RS says

P. Cruse / Stud. Hist. Phil. Sci. 36 (2005) 557–576

563

we get to talk about them by description, that is, by asserting the existence of something characterized using observational terms like ÔredÕ and ÔsquareÕ (plus mathematics and logic). Thus, we preserve the intuition that we do talk about unobservable objects—since unobservable objects have to exist, and have the properties we attribute to them for our Ramsey sentences to be true—while honouring the idea that our semantic capacities derive essentially from perceptual awareness. The alleged problem with this view, then, is that if Ramsey sentences are formed according to the strong distinction, then they amount only to cardinality claims, so we can only Ôtalk aboutÕ unobservable objects in a very attenuated sense. The issue we need to look at is therefore why anyone who embarked on the upward path would see themselves as committed to the strong O/T distinction. One well-known version of the upward path is clearly is committed to a version of the strong O/T distinction, or at least something isomorphic to it. This is the view held by Grover Maxwell (1966, 1970a,b), inspired by an earlier view of Bertrand Russell (1927).2 Maxwell and Russell based their view on an indirect theory of perception, and held that only ÔperceptsÕ or mental events and properties could be directly perceived, and thus referred to (Maxwell, 1966, p. 152, Russell, 1927, pp. 197–99). However, both also held that explanation of the structure of our perceptions motivated inference to theories about their unobservable causes. Maxwell made explicit the idea that these two doctrines could be reconciled using Ramsey sentences: although we can only directly refer to percepts, we can come to know that there exist properties in the unobservable world that are characterized in terms of their ÔstructuralÕ relations with those percepts, and this is sufficient to explain the structure our percepts themselves instantiate (Maxwell, 1970b). On the Russell–Maxwell view, the mental and physical domains are entirely disjoint, so no (non-mathematical) property which applies to a mental event applies to a physical event or vice versa. Similarly, no (non-mathematical) predicate which applies to a mental event will apply to a physical event. Thus, although the strong O/T distinction I defined above assumed that an observational term is one which ranges over (and only over) observable physical objects, the Russell–Maxwell view implies an isomorphic distinction, with mental events playing the role of observable objects, and physical events playing the role of unobservable objects. If we make the appropriate substitutions, it therefore follows from (1) above that the Russell– Maxwell view reduces our knowledge of the physical world to knowledge of its cardinality. Russell and MaxwellÕs view that only mental events can fall into the extension of directly referring (ÔobservationalÕ) vocabulary is certainly one influential empiricist claim. But this does not really do justice to the intuitive idea that the privileged vocabulary is observational in a general sense rather than mental, since intuitively at least some physical objects and their properties can be observable. To do this we would need something more like van FraassenÕs (1980, pp. 16–19) idea that an object or entity is observable if it is possible for a human to perceive it with their

2

See Demopoulos & Friedman (1985) for a detailed discussion of RussellÕs view.

564

P. Cruse / Stud. Hist. Phil. Sci. 36 (2005) 557–576

naked senses. This could be turned into a distinction between observational and theoretical terms by stipulating that an observational term is one that refers to an observable entity, and a theoretical term one that does not. However, van FraassenÕs distinction tells us only the conditions under which objects count as observable, and says nothing about properties and relations. If we assume that an observational predicate is one that denotes an observable property or relation (I will assume this from now on) we will get the strong O/T distinction only if observable properties and relations—and thus observational predicates—are those which range only over observable objects. Why draw the distinction this way? The sorts of terms we might intuitively class as ÔobservationalÕ—maybe ÔredÕ, ÔsquareÕ, and the like—fail to count as observational on the strong distinction, since we can meaningfully (and for a realist, truly) assert the existence of red blood cells, or microscopic square grids, for example.3 If we start with the intuition that terms like ÔredÕ and ÔsquareÕ are semantically unproblematic in a way that ÔelectronÕ and ÔstrangenessÕ are not, we will not get the strong O/T distinction. Someone who wanted to Ramsify on the basis that observational predicates were only meaningful insofar as they applied to observable objects would surely therefore need some additional motivation for doing so. I can only think of one source for this motivation. This would be a strong version of verificationism on which our capacity to speak about the world is entirely circumscribed by our ability to directly observe what is true of it. This would entail that we could only meaningfully call something ÔredÕ or ÔsquareÕ insofar as we could observe it is red or square, and thus would entail the strong O/T distinction. I donÕt think it is too controversial to say that verificationism of this strong form is very problematic, though I will not argue for this. But the more important point is that if you believe it, you have at your disposal a very strong argument for anti-realism about scientific theories, independent of trivialization worries related to Ramsey sentences. That is, if we can only refer to the world insofar as we can directly observe what is true of it then this is a strong reason to think that we cannot represent the unobservable at all. It would therefore be strange in this context to try to appeal to Ramsey sentences to bootstrap some form of scientific realism. In the context of verificationist pictures of scientific theories, Ramsey sentences have more often been thought of as a means to defend anti-realism, showing how the Ôdeductive systematizationÕ of theories can be preserved while removing their implications about unobservables.4 If this is your project (1) will if anything seem too realist, since it still entails that Ramsey sentences make cardinality claims about the unobservable. Aside from the Russell–Maxwell project, then, I can therefore see very little to recommend RS when based on the 3 This point was made against CarnapÕs conception of an observational term in Putnam (1962), pp. 217– 218. As Putnam also points out Carnap was never clear whether he conceived of the O/T distinction in its strong form. It is therefore not clear whether CarnapÕs later appeals to Ramsey sentences (Carnap, 1961, 1963, 1966) should be regarded as susceptible to (1). 4 See Scheffler (1961), Hempel (1958) for discussion, though neither endorses the proposal. Carnap (1966) argues that Ramsey sentences lead to an ÔinstrumentalistÕ view of theories.

P. Cruse / Stud. Hist. Phil. Sci. 36 (2005) 557–576

565

strong version of the O/T distinction, and even less to recommend trying to combine it with scientific realism. In fact I think there is a much better way of understanding the intuitive proposal outlined above that does not rely on the strong distinction. The intuitive proposal requires that the unproblematic class of observational predicates refer to, broadly speaking, perceptible, or observable properties such as redness or squareness. A natural understanding of this would be that these observable properties are unproblematic not because they are always observable, but simply because we can in at least some cases observe them, giving us sufficient epistemic grip to refer to them. This contrasts with properties like strangeness or being a DNA molecule which can in no case be observed or perceived. As before, we could then identify the observational predicates as those which denote properties whose instances can in at least some cases be observed. We can call this version of the O/T distinction the weak O/T distinction. Although I believe that this is a prima facie plausible way to draw the observational/theoretical distinction there are further issues one would have to address if one were to use it as a basis for RS. One issue would be whether terms which refer to observable properties or objects, but whose meaning seems essentially to involve theory—such as, maybe, ÔDVD playerÕ or ÔreptileÕ—are going to get classed as observational. This is clearly important, since the observational basis available for forming Ramsey sentences would differ significantly depending on the answer to this question. The answer will presumably depend on precisely which type of concept empiricism one starts off trying to defend. One relatively localized kind of empiricism would simply ask how it is possible for scientific theories to refer to entities or properties like strangeness, which are not instantiated in any perceived objects, and would demand an explanation referring to properties and objects which are present at the perceptible level. In this case one might be happy to class ÔDVD playerÕ as an observational term, since the property of being a DVD player is clearly instantiated in perceived objects. However, an empiricist who takes seriously the idea that all our conceptual capacities arise from experience might have the wider project of explaining how the meanings of our terms could be given with reference to properties which are not only instantiated in perceptible things, but directly and primitively represented by perceptual states. On such a view being a DVD player would presumably not count as an observable property, although properties like redness and squareness might.5 There is not room to investigate either form of concept empiricism any further here. The point I wish to make is just that both would allow observational terms

5 One way to identify properties which are directly and primitively represented in perception would be to appeal to modular theories of cognitive architecture of the sort developed by Jerry Fodor (see esp. Fodor, 1983, 1984). According to such theories the neural systems that generate perceptual representations function largely independently of systems that are responsible for belief formation, and will represent a subjectÕs environment in the same way regardless of what wider beliefs they hold. We could then say that a property F is (primitively) observable iff it is possible that there be some object o such that a human perceptual module can form a representation that o is F. I discuss this further, and argue that the idea might be used as a basis for a form of RS in Cruse (2005).

566

P. Cruse / Stud. Hist. Phil. Sci. 36 (2005) 557–576

to apply to unobservable entities, so neither would commit proponents to the strong O/T distinction.6

5. The downward path to Ramsification and structural realism The downward path, we will recall, begins by assuming that theories have the content a realist says they do. Next, some argument is put forward to demonstrate that not all of this content should be regarded as knowledge-expressing. Finally, Ramsification is used as a means to ÔsubtractÕ the surplus content so as to leave only the genuinely knowledge-expressing content of the theory. Does this line of thought lead to the conception of a Ramsey sentence for which (1) would be problematic? One example of this line of reasoning derives from the argument in favour of SR put forward by John Worrall (1989). Worrall sees his argument as a means to reconcile the Ôno-miraclesÕ argument and the Ôpessimistic inductionÕ in the scientific realism debate. The problem is that while the no miracles argument advocates that we infer to the approximate truth of our best theories as a means to explaining their empirical success, the existence of persistent and radical theory change in the history of science suggests the opposite conclusion. SR reconciles these two arguments by noting that through the radical changes in our scientific ontology, there has been significant continuity in the way in which the ÔstructureÕ of the unobservable has been described in successive theories, where structural properties are roughly what are described by the mathematical equations associated with a given theory. WorrallÕs example is the case of the aether—through differing ideas of the carrier of light waves (from the mechanical aether to the disembodied field) our knowledge of the basic mathematical structure of electromagnetic waves has been steadily cumulative. The key question the structural realist has to answer is what exactly constitutes ÔstructureÕ. The suggestion we are interested in here is that structural content can be characterized using Ramsey sentences. On this view Ramsifying a theory will remove that part of its content that tends to describe the ÔnaturesÕ of the entities it postulates, leaving only that part that describes their ÔstructureÕ. However, to make this precise, we need to say according to what version of the O/T distinction Ramsey sentences are going to be formed. We might note first the consequences of saying that Ramsey sentences expressing structural content should be formed according to the strong O/T distinction. The central point about Ramsifying according to the strong O/T distinction is that it allows no substantive (that is non-logico-mathematical) predicates in Ramsey

6

Though slightly peripheral to the concerns of this paper it is worth noting that one can argue for RS without immediately committing oneself to any version of the distinction between observational and theoretical terms at all, if one appeals David LewisÕs semantics for theoretical terms (see Lewis, 1970). On LewisÕs view the content of a theory is exhausted by its Ramsey sentence with uniqueness, where Ramsey sentences are formed on the basis of the distinction between old and new terms, which he argues cross-cuts the observational-theoretical distinction. LewisÕs semantics is used in the defence of scientific realism in Cruse & Papineau (2002) and Cruse (2004).

P. Cruse / Stud. Hist. Phil. Sci. 36 (2005) 557–576

567

sentences that apply to unobservable entities. Hence, if a structural realist held that the content of any theory was expressed by this sort of Ramsey sentence, they would be committed to the claim that the only respect in which we can knowledgeably describe unobservable entities like the aether is using mathematical predicates.7 I am not sure whether this is what structural realists such as Worrall actually intend to assert. One can certainly find passages where something like this claim is made. For example, Worrall says of the principle that mathematical equations of earlier theories are often preserved as limiting cases of later ones, that it applies Ôpurely at the mathematical level, and hence is quite compatible with the new theoryÕs basic theoretical assumptions (which interpret the terms in the equations) being entirely at odds with those of the oldÕ (Worrall, 1989, p. 160, authorÕs own emphasis). If this leads to the claim that all we know of unobservable reality is that it obeys certain equations (we know not how), I think it is genuinely rendered untenable by (1). This may be construed as an out-and-out refutation of structural realism. However, structural realists have also made a slightly different kind of claim. Consider for example WorrallÕs claim that from the point of view of MaxwellÕs theory, ÔFresnel was quite right not just about a whole range of optical phenomena, but right that these phenomena depend on something or other that undergoes periodic change at right angles to the lightÕ (ibid., p. 159). Note that Ôundergoing periodic change at right angles to the lightÕ is not just a matter of obeying mathematical equations. Something could conceivably obey FresnelÕs equations under some interpretation that does not equate the periodicity inherent in them with physical time or the mathematical angular co-ordinate with a physical angle, suggesting that Worrall he must be adding some partial physical interpretation to FresnelÕs equations. A similar suggestion also been made recently by Chakravartty (1998, 2004), who argues that structural realism should be understood as involving a realist attitude to detection properties as opposed to auxiliary properties. Detection properties Chakravartty defines as Ôcausal properties which we have managed to detect; they are causally linked to the behaviour of our detectorsÕ (2004, p. 162), while auxiliary properties are any other properties which theories attribute to entities.8 He suggests that detection properties can be identified by the fact that they are involved in a minimal interpretation of the mathematical equations of theories—that is, one which attributes enough content to understand how the equations can be used in making predictions, and no more. Thus, in the case of the aether, the detection properties are light intensities and directions of propagation, whereas the aetherÕs solidity or

7

Or more precisely, mathematical predicates plus strongly observational predicates. However since strongly observational predicates by definition apply to no unobservable entities, I take it this amounts to the same thing. 8 French and Ladyman (see esp. their 2003, pp. 45–46) also suggest that structural realists should not take the structures which they believe to be purely mathematical, but as physical (though not in any way ontologically dependent on objects); indeed they speak of a blurring between mathematical and physical structures.

568

P. Cruse / Stud. Hist. Phil. Sci. 36 (2005) 557–576

elasticity, for example, are auxiliary (see Chakravartty, 1998, pp. 395–397; 2004, p. 163). This is clearly an idea similar to the interpretation of Worrall I suggested above.9 A second point in favour of a broader conception of structure would be if SR had ambitions to apply outside the realm of mathematical physics.10 For example, it does not look implausible to apply some sort of ÔstructuralistÕ understanding to DarwinÕs theory of natural selection (right about the ÔstructureÕ of the process of natural selection, wrong about the ÔpangeneticÕ nature of the mechanism) or nineteenth-century atomic theories of matter (right about the atomic structure of matter and chemical composition, wrong about the intrinsic nature of the atom). It seems unlikely that what is correct about the theories in question can be expressed in purely mathematical terms. I donÕt intend to assess the merits of this proposal here, though for reasons already expressed I do consider it much more promising than Ôpure mathematicalÕ structural realism. Instead I will comment on how it relates to Ramsey sentences. First of all, I think it is reasonable to suppose that this modified structuralist approach might at least in some cases need to be implemented using Ramsey sentences. The approach recommends that we restrict our realism to belief in descriptions of detection properties such as light intensities and directions of propagation rather than auxiliary properties like the solidity and elasticity of the aether. However, it would be odd to conclude from this that light actually has no further properties than these; that these intensities and directions can somehow subsist on their own. A more reasonable claim is surely that while we are ignorant of what further properties light might have, we can know that light has some further properties which stand in certain relations to the intensities and directions we know about, even if we are ignorant of what these properties are.11 If so, our knowledge could in this case be represented in Ramsey sentence form. I will return below to the relationship between Ramsey sentences and this modified form of structural realism. For now I will simply note that if we do interpret it in 9 Psillos (1999), pp. 157–161, uses a similar argument against structural realism, arguing that it shows that more than ÔstructureÕ as the structural realist conceives it is preserved in the transition from FresnelÕs theory to Maxwellian electromagnetism. However, I am suggesting that the structural realist should subsume the retained properties Psillos regards as Ônon-structuralÕ under the wider notion of structure I am suggesting. French & Ladyman (2003), p. 35, also suggest that PsillosÕs insight might support a form of structural realism, though they are referring in this case to the ontic form of SR that they hold. 10 That structuralism (in this case ontic structuralism) should apply outside physics is also suggested by French & Ladyman (2003), p. 32, who claim that the view can also be applied (inter alia) to qualitatively expressed biological theories. 11 Chakravartty (1998), p. 397, seems to make a similar suggestion: ÔNote that the distinction between detection and auxiliary properties . . . does not necessitate that light itself be rectilinear, transversely oscillatory, etc. Rather, light has the property that (or properties such that), when subjected to specific forms of detection, certain characteristics are causally manifested and detectedÕ. This seems to be in the spirit of the idea that our knowledge in the aether case might be expressible in Ramsey sentence form. It is also worth noting that ontic structural realists would regard the structural relations described by FresnelÕs equations as ontologically subsistent, though this may not be best conceptualized as the idea that properties of intensity and direction are subsistent.

P. Cruse / Stud. Hist. Phil. Sci. 36 (2005) 557–576

569

terms of Ramsey sentences as I have suggested, those Ramsey sentences will attribute at least some determinate physical properties—such as oscillation at right angles to the direction of light propagation—to unobservable entities such as light waves. Clearly, then, no commitment is generated to the strong O/T distinction, on which unramsified terms can only range over observable objects. Thus, there is no reason to think that this type of structural realism would suffer from the trivialization problems engendered by (1).

6. The relationship between structural realism and Ramsey sentences I now hope to have demonstrated that there are forms of both RS and SR that involve no commitment to the strong version of the O/T distinction, and therefore remain untouched by trivialization problems. What I will now look at is the relationship between RS and SR: should they be identified with one another, or are they distinct? 6.1. Is (defensible) SR a version of RS? We saw above that on one interpretation of structural realism, structural properties are not just mathematical properties, but include a broader range of physical properties. I argued that if this is Ramsey sentence realism, it is Ramsey sentence realism of a sort that would not fall foul of trivialization problems. However, I think that the position we arrive at here is different from the kind of Ramsey sentence realism we would get from any version of the Ôupward pathÕ, and may not best be conceptualized as a form of Ramsey sentence realism at all. The reason is this. Ramsification essentially shows us how much theoretical content it is possible to preserve while doing without a certain class of term. Now in the case of the upward path to RS—that is, the route beginning from concept empiricism—we wanted to retain terms which referred to observable objects or properties, broadly construed. However, in the Ôdownward pathÕ, we wanted to retain those terms which are (inter alia) involved in a minimal physical interpretation of the equations of theories, and likely to be preserved in theory-change—I will continue to use ChakravarttyÕs term Ôdetection propertiesÕ for this class. The reason I think RS and SR are non-equivalent is that the class of observational terms, however delineated, will not be coextensive with the class of terms which refer to detection properties. The point can be illustrated using the case of the aether. We have been supposing that the structural realist will wish to say that the detection properties postulated by aether theories were the transverse oscillations it associated with light, and that the auxiliary properties were the solidity and elasticity it attributed to the aether. For this to be expressible as a Ramsey sentence claim, Ramsifying aether theories would have to siphon off the terms that refer to the auxiliary properties (elasticity, solidity, composition out of atoms etc.) leaving us with an assertion of the existence of something with the detection properties (oscillatory motions, maybe conserved energies, and so on).

570

P. Cruse / Stud. Hist. Phil. Sci. 36 (2005) 557–576

My basic point here is that it is not clear that the terms which describe the auxiliary properties postulated by aether theories will always describe auxiliary properties in other theories. Take the case of solidity. What did aether theorists mean by saying the aether was a solid medium? Granted, if they meant to attribute to it some ineffable nature in virtue of which it has elastic properties, the structural realist would do well to rule this out as a possibly knowledge-expressing claim. But so, presumably, would any realist, so this wonÕt isolate what is distinctive about the structuralist view. A better interpretation is surely that the ÔsolidityÕ of the aether amounted to the fact that it was thought to be subject to deformation under certain forces, and that when deformed its parts would experience further restoration forces that tended to return them to their original relative positions. These features are, I assume, still auxiliary properties, but will the structural realist want to say that the terms involved describing them are universally ÔbadÕ? I suggest not. The reason is that the same terms seem likely to resurface in describing parts of theories in other parts of science that the structural realist would want to believe in. For example, it is often noted that the continuity between classical mechanics and quantum theory is exhibited in the fact that certain relations in Newtonian mechanics are preserved as Ôlimiting casesÕ of quantum mechanical relationships. One example is EhrenfestÕs theorem in quantum mechanics, which relates the expectation values of force, mass and acceleration in the way that NewtonÕs second law relates the quantities themselves. In this case if the structural realist wants to say that more than just the maths that was preserved it would look very implausible to deny that some physical notion of force has been retained through the theorychange; if so, the term ÔforceÕ would be necessary to provide a minimal physical interpretation to EhrenfestÕs theorem, so in this context would refer to a detection property. Similarly, it seems we cannot say what was right about different versions of the atomic theory of chemistry without stating that atoms are parts of physical objects. Together, though, as we saw, the terms ÔforceÕ and ÔpartÕ can be used to describe some of the auxiliary properties of aether theories. So whether a term describes detection properties or auxiliary properties will depend on the particular theory in which it occurs. One could even argue that the terms that are needed to state the ÔstructuralÕ content of aether theories can also be used to state part of its own non-structural, ÔauxiliaryÕ content. For example, the claim that light involves a periodic oscillation at right angles to the direction of its propagation is part of aether theoriesÕ structural content. But the claim that there exists a frame of reference (i.e. fixed in the aether) movement relative to which is detectable through the differential behaviour of light, is part of the non-structural—or at least part of the non-retained—content of aether theories. The terminology required to state these claims, however, is essentially the same—ÔlightÕ plus some kinematic vocabulary. If this is right then there is some doubt as to whether the claim that the structuralist wishes to make about aether theories can be made with reference to classes of good and bad terms at all. Contrast this with the Ôupward pathÕ to RS. Here the terms we can use in forming Ramsey sentences are ÔobservationalÕ terms which refer to observable properties.

P. Cruse / Stud. Hist. Phil. Sci. 36 (2005) 557–576

571

Now which properties are actually observable—for a concept empiricist at least— will be fixed by general facts about human perceptual capacities, so a termÕs status as observational will not depend on which particular theory it occurs in.12 This is certainly true on the weak O/T distinction suggested above, on which a term is observational if any of the things it applies to are observable. On this conception, it would be incoherent to say that ÔforceÕ say, was observational in one theory but not in another. This results from a more fundamental disanalogy between the upward path to RS and the downward path to SR. On the upward path, the basic motivation for picking out a privileged set of properties to occur in Ramsey sentences is semantic, in that the observational properties are those to which we can assume unproblematic reference. On the other hand, the downward path is motivated by epistemology—detection properties are those which our theories about the unobservable give us warrant to believe in. Now the fact that certain terms are semantically able to ground theoretical content does not in itself provide warrant for any particular theory we form using them: it implies theories containing them could express knowledge, but not that they actually do. On the other hand, there is no obvious reason why the properties which structural realists claim we have warrant to believe should be semantically privileged for an empiricist: why, for example, should an unobservable property like strangeness not turn out to be required to minimally interpret the equations in parts of particle physics? As a result, I claim that while a structural realist might in some cases wish to make use of Ramsey sentences to express theoretical content, there is no reason to think that this will work in all cases, or even if it did, that the resulting position would coincide with the result of the Ôupward pathÕ to RS. 6.2. Is (defensible) RS a version of SR? We have seen that RS would be a viable position as long as it was assumed that Ramsey sentences were formed according to the weak, rather than the strong, O/T distinction. But should someone who appeals to Ramsey sentences in this way conceive of their position as a form of SR? To address this issue we first of all need to say something about what distinguishes structural realism from ÔfullÕ or ÔstandardÕ realism. We have noted that a standard idea is that these positions differ because the latter holds that we can have knowledge both of the structure and the nature of unobservable reality, whereas the former holds we can only have knowledge of its structure. This itself is a controversial idea, but we can begin by asking whether RS commits us to anything like it.

12 Of course, there are many philosophers—such as Hanson (1958), Kuhn (1962) and Churchland (1979)—who have thought that what we can observe does depend on what theories we hold at the time. However, this idea is normally thought to be antithetical to concept empiricism, since it denies the possibility of an Ôobservational basisÕ from which theoretical content can be constructed.

572

P. Cruse / Stud. Hist. Phil. Sci. 36 (2005) 557–576

One philosopher who thought that that it did was Grover Maxwell. He said: Structural characteristics may be taken to be just those that are not intrinsic and can be described by means of logical terms and observation terms. Intrinsic properties are those that are, or could be, direct referents of predicates . . . although the Ramsey sentence gives us knowledge only of the structural properties of the theoretical and leaves us ignorant of its intrinsic (first order) properties, it does refer by means of description to unobservable intrinsic properties. That is, the ignorance concerning intrinsic nature is ignorance concerning what the intrinsic properties are . . . (Maxwell, 1970, p. 188). MaxwellÕs view is based on the Russellian distinction between definite descriptions and logically proper names. His suggestion is that in a ÔfullÕ theory both observational and theoretical predicates could name or directly refer to theoretical properties and entities, whereas a Ramsey sentence names only observable properties, and denotes unobservable properties and entities only by description. Maxwell seems to have held that this distinction entailed the claim that Ramsey sentences restrict our knowledge to structure alone. However, it is difficult at first glance to see how the distinction between descriptions and names can bear this epistemological weight. Suppose I know that there exist entities with all the properties modern physics attributes to the electron, and you know in addition only that electrons have those properties. It is not at all clear why in this situation your knowledge is substantively greater than mine. Intuitively you would only know more than me if you knew more about electrons than I did, but in this case, your knowledge by description would also differ from mine, so the difference would not be due to the mere difference between naming and description.13 One might respond here by arguing that while the name/description distinction is not epistemologically significant per se, it is nevertheless associated with further consequences that are. For example, Maxwell held, following Russell, that being in a position to refer to something is associated with our being perceptually acquainted with it, so the fact that we cannot name unobservable properties implies that we are not thus acquainted with them. But while this mean that restricting belief to the Ramsey sentences of theories entails some sort of epistemological limitation, this doesnÕt seem relevant to the scientific realism debate as it is normally conducted, since I take it that not even Ôfull realistsÕ would think we can be acquainted with unobservable properties in this way.14 I therefore believe that there is a strong case for saying that realism about Ramsey sentences is not, epistemologically speaking, a 13

Psillos makes a similar point in his (2001), pp. 19–20. An exception here might be Ian Hacking (see for example Hacking, 1982, 1985), though it isnÕt clear. Hacking certainly believes that experimenters, through manipulating entities like electrons, gain knowledge of electrons which is distinct from the knowledge implied by theoretical models of the electron, so in this sense experimental knowledge would differ from that implied by the Ramsey sentences of current theories. However, he also speaks of experimentersÕ knowledge of electrons in terms of descriptions of the causal properties of electrons (see for example Hacking, 1982, p. 168), so it is not clear that such knowledge would not merely be expressed by different Ramsey sentences, rather than not by Ramsey sentences at all. 14

P. Cruse / Stud. Hist. Phil. Sci. 36 (2005) 557–576

573

significantly weaker position than realism about theories, since saying otherwise relies on a strong claim about the epistemological significance of naming or reference, which I donÕt think even full realists would wish to maintain.15 If the foregoing is right, there is no reason to think that realism about Ramsey sentences is significantly different from realism about theories. But one might respond that full realism requires more than just belief in scientific theories; it requires that we also provide a realist metaphysical account of the entities that satisfy our theories. This claim has recently been made by Steven French and James Ladyman as part of their defence of their ÔonticÕ form of structural realism (see Ladyman, 1998; French & Ladyman, 2003). Ladyman (1998) argues that the ontology of many physical theories—for example the issue whether or not quantum particles are individuals— is underdetermined by empirical evidence. As a result, he claims that a realist who believes the theory of quantum mechanics (QM), say, but remains agnostic about the status of its ontology is ÔersatzÕ, in that they fail, essentially, to specify what they are being realist about. Standard or ÔtraditionalÕ realism avoids this by adding an ontology described in terms of categories such as individuality or identity to bare belief in the theory. But because the choice of this interpretation is underdetermined by the empirical evidence for QM, French and Ladyman recommend instead an ontology of primitively subsistent structures rather than individuals, and regard the underdetermined ÔtraditionalÕ ontologies as representations of these structures. I will not comment here on the merits of the ontic structuralist view. The question at issue is whether this argument shows that RS is only ÔersatzÕ realism, on the grounds that it imposes an agnosticism about the metaphysical status of certain theoretical entities. One reason to doubt this is that it is not clear that agnosticism about the ontological implications of theories is incompatible with genuine scientific realism. A challenge to this claim, raised by Chakravartty (2004), is that the same argument does not seem to apply in other contexts—the fact that physics underdetermines whether bare particular theory or bundle theory is true doesnÕt turn you into anything less than a realist about ordinary material objects, for example—so it is not clear why it should apply specifically to QM.16 But whether or not this argument is right, I doubt whether RS implies agnosticism about ontological questions in the first place. As I pointed out above, RS is essentially a view about semantics—it says that the content of any theory we believe is given by its Ramsey sentence. It does not tell us anything about the conditions under which we are warranted in believing theories. Thus, provided it doesnÕt imply that metaphysical interpretations of QM 15 A further view might be that direct reference is epistemologically significant, since the ability to name something covaries with the existence of certain types of causal connection between subject and referent (see for example Kripke, 1980; Putnam, 1973). However, I have argued in detail elsewhere that causal connections do not have this kind of epistemological significance (Cruse, 2004). Briefly, my argument is that if we hold the descriptive knowledge possessed by a scientist fixed—as expressed in the Ramsey sentence of their theory—then the presence or absence of causal connections with unobservable entities makes no difference to our ability to understand their ability to predict and explain empirical results, and therefore is irrelevant to our grounds for attributing theoretical knowledge to them. 16 See French & Ladyman (2003), pp. 50–51 n. 14, for a reply to (a predecessor of) this argument, and Chakravartty (2004), pp. 159–160, for his response.

574

P. Cruse / Stud. Hist. Phil. Sci. 36 (2005) 557–576

and other theories are meaningless, it could not also imply that such interpretations are not warranted. And as far as I can see it does not. The most plausible version of RS is based on the idea that the terms used in forming a meaningful Ramsey sentence have to be (weakly) observational. But the terms involved in stating at least the Ôtraditional realistÕ interpretation of QM—ÔindividualÕ, ÔidentityÕ, and so on—do seem to be observational in this sense; indeed, as Ladyman points out (Ladyman, 1998, p. 423), the whole point of such an interpretation is to try to understand QM using categories applicable in the world of experience. Thus, I believe that RS is perfectly compatible with at least the traditional realist interpretation of QM. One might nevertheless question whether RS is consistent with ontic structural realism, which Ladyman claims does not assume that Ôthe structure of the mind-independent world be imaginable in terms of the categories of the world of experienceÕ (ibid., p. 423). If it is not, then its compatibility with a traditional ontology would be little consolation if the case for ontic SR were compelling. However, it strikes me as equally plausible to turn this objection around. One could reasonably require that if ontic SR is to be comprehensible, the ontology it recommends be in some way explicable using terms applying to the world of experience, since this is just the requirement that it be meaningful at all. If it is not, then it would be ersatz in the sense that its proponents would have failed to specify adequately what it is actually claiming, in which case arguments for or against it would not even arise. I have tried to argue, then, that RS does not support any kind of structure/nature distinction, so there is no justification for equating it with SR on these grounds. However, one might question whether this really means that RS should be distinguished from SR. Could it not instead be taken to show that RS is structural realism, and the supposed added content of Ôfull realismÕ is illusory? I donÕt think so, since I think the foregoing shows that even if we repudiate the structure/nature distinction, there are still ways of distinguishing full realism from structural realism. One is that full realism, unlike structural realism, allows Ôtraditional realistÕ metaphysical interpretations of theories. Another is that full realism may say we have warrant for parts of theories which a structural realist calls ÔauxiliaryÕ, and therefore doesnÕt believe in—for example, a realist might claim that nineteenth-century scientists did at the time have warrant for believing in the aether. RS, in both these senses, is at least consistent with the fully realist view, since Ramsey sentences can postulate both traditional realist metaphysics, and, in some cases—such as the aether—auxiliary properties. Thus I think that insofar as there is any content to the distinction, RS is better aligned with full realism than SR.

7. Conclusion In this paper I hope to have disentangled some of the relationships between structural realism and the view that the content of a theory is expressed in its Ramsey sentence, and shown the extent to which proponents of these theories should be worried by trivialization arguments. In particular I hope to have demonstrated that:

P. Cruse / Stud. Hist. Phil. Sci. 36 (2005) 557–576

575

(i) RS is only a version of anti-realism if Ramsey sentences are assumed to be formed according to the strong version of the O/T distinction, on which only terms that range entirely over observable objects count as observational. SR also only collapses into anti-realism if it is committed to this form of RS. (ii) Some versions of RS and SR are committed to this form of the O/T distinction, for example Russell–Maxwell structural realism and structural realism based on the claim that our knowledge of unobservable entities is purely mathematical. Verificationist versions of RS are committed to the distinction, but would not regard it as an objection that this commits them to anti-realism. (iii) There are also versions of RS and SR that are not committed to the strong O/T distinction. RS is not committed to the distinction if it is motivated by the upward path, but based on the more intuitive weak version of the O/T distinction. SR is not committed to the strong O/T distinction if it entails we know some physical properties (such as ChakravarttyÕs detection properties) of the unobservable, rather than purely mathematical properties. (iv) SR is in this less extreme form not the same as RS, when the latter is based on the Ôupward pathÕ. This is because RS identifies the un-Ramsified terms semantically, as referring to observable properties, whereas SR identifies them epistemically, in terms of their likelihood of surviving theory-change. These categories need not coincide. (v) RS, in its defensible form, is not a version of SR, since structural realism on any interpretation involves a concession to anti-realism, attributing less content to theories than a full scientific realist would, whereas RS does not. For this reason, the claim that RS is not distinguished from full realism is not an objection. I would claim, then, that two positions survive the trivialization arguments discussed at the outset—modified SR, and RS based on the weak O/T distinction. Both views have considerations in their favour, and indeed they are not inconsistent; though nor, as I have tried to show, are they identical. The question whether either is actually true will have to be addressed elsewhere. Acknowledgements Many thanks to Jeff Ketland for very useful discussion on an earlier version of this paper. Thanks also to an anonymous referee for this journal for helpful feedback which helped me improve the paper.

References Carnap, R. (1961). On the use of HilbertÕs e-operator in scientific theories. In Y. Bar-Hillel, E. I. J. Poznanski, M. O. Rabin, & A. Robinson (Eds.), Essays in the foundations of mathematics (pp. 156–164). Jerusalem: Magnes Press. Carnap, R. (1963). Replies and systematic expositions. In P. Schilpp (Ed.), The philosophy of Rudolf Carnap (pp. 859–1013). La Salle: Open Court.

576

P. Cruse / Stud. Hist. Phil. Sci. 36 (2005) 557–576

Carnap, R. (1966). Philosophical foundations of physics. New York: Basic Books. Chakravartty, A. (1998). Semirealism. Studies in History and Philosophy of Science, 29A, 391–408. Chakravartty, A. (2004). Structuralism as a form of scientific realism. International Studies in the Philosophy of Science, 18, 151–171. Churchland, P. M. (1979). Scientific realism and the plasticity of mind. Cambridge: Cambridge University Press. Cruse, P. (2004). Scientific realism, Ramsey sentences, and the reference of theoretical terms. International Studies in the Philosophy of Science, 18, 133–149. Cruse, P. (2005). Empiricism and RamseyÕs account of theories. In H. Lillehammer, & D. H. Mellon (Eds.), RamseyÕs legacy (pp. 105–122). Oxford: Clarendon Press. Cruse, P., & Papineau, D. (2002). Scientific realism without reference. In M. Marsonet (Ed.), The problem of realism (pp. 174–189). London & Aldershot: Ashgate. Demopoulos, W., & Friedman, M. (1985). Critical notice: Bertrand RussellÕs The analysis of matter: Its historical context and contemporary interest. Philosophy of Science, 52, 621–639. Fodor, J. (1983). The modularity of mind. Cambridge, MA: MIT Press. Fodor, J. (1984). Observation reconsidered. Philosophy of Science, 51, 23–43. French, S., & Ladyman, J. (2003). Remodelling structural realism: Quantum physics and the metaphysics of structure. Synthe`se, 136, 31–56. Hacking, I. (1982). Experimentation and scientific realism. Philosophical Topics, 13, 154–172. Hacking, I. (1985). Representing and intervening: Introductory topics in the philosophy of natural science. Cambridge: Cambridge University Press. Hanson, N. R. (1958). Patterns of discovery. Cambridge: Cambridge University Press. Hempel, C. G. (1958). The theoreticianÕs dilemma: A study of the logic of theory-construction. In H. Feigl, M. Scriven, & G. Maxwell (Eds.), Minnesota studies in the philosophy of science, Vol. 2. Concepts, theories and the mind–body problem (pp. 37–98). Minneapolis: University of Minnesota Press. Ketland, J. (2004). Empirical adequacy and ramsification. British Journal for the Philosophy of Science, 55, 287–300. Kripke, S. (1980). Naming and necessity. Oxford: Blackwell. Kuhn, T. S. (1962). The structure of scientific revolutions. Chicago & London: University of Chicago Press. Ladyman, J. (1998). What is structural realism? Studies in History and Philosophy of Science, 29A, 409–424. Lewis, D. K. (1970). How to define theoretical terms. Journal of philosophy, 67, 427–446. Maxwell, G. (1966). Scientific methodology and the causal theory of perception. In I. Lakatos, & A. Musgrave (Eds.), Problems in the philosophy of science (pp. 148–160). Amsterdam: North-Holland Publishing Company. Maxwell, G. (1970a). Theories, perception and structural realism. In R. Colodny (Ed.), The nature and function of scientific theories (pp. 3–34). Pittsburgh: University of Pittsburgh Press. Maxwell, G. (1970b). Structural realism and the meaning of theoretical terms. In S. Winokur & M. Radner (Eds.), Minnesota studies in the philosophy of science, Vol. 4. Analysis of theories and methods of physics and psychology (pp. 181–192). Minneapolis: University of Minnesota Press. Newman, M. H. A. (1928). Mr. RussellÕs Ôcausal theory of perceptionÕ. Mind, 37, 137–148. Psillos, S. (1999). Scientific realism: How science tracks truth. London: Routledge. Psillos, S. (2001). Is structural realism possible? Philosophy of Science, Supplemental Vol. 68, S13–S24. Putnam, H. (1962). What theories are not. In E. Nagel, P. Suppes, & A. Tarksi (Eds.), Logic, methodology and philosophy of science (pp. 215–227). Stanford: Stanford University Press. Putnam, H. (1973). Explanation and reference. In D. Pearce, & S. Maynard (Eds.), Conceptual change (pp. 199–221). Dordrecht: Reidel. Russell, B. (1927). The analysis of matter. London: Routledge & Kegan Paul. Scheffler, I. (1961). The anatomy of inquiry. New York: Alfred A. Knopf. van Fraassen, B. C. (1980). The scientific image. Oxford: Clarendon Press. Worrall, J. (1989). Structural realism: The best of both worlds? Dialectica, 43, 99–124. Zahar, E. (2001). Poincare´Õs philosophy: From conventionalism to phenomenology. Chicago & La Salle, IL: Open Court.