Synthese (2008) 162:85–100 DOI 10.1007/s11229-007-9171-z
Structural realism and Davidson Jack Ritchie
Received: 11 April 2006 / Accepted: 17 April 2007 / Published online: 17 May 2007 © Springer Science+Business Media B.V. 2007
Abstract Structural realism is an attempt to balance the competing demands of the No Miracles Argument and the Pessimistic Meta-Induction. In this paper I trace the development of the structuralist idea through the work of one of its leading advocates, John Worrall. I suggest that properly thought through what the structuralist is offering or should be offering is not an account of how to divide up a theory into two parts—structure and ontology—but (perhaps surprisingly) a certain kind of theory of meaning—semantic holism. I explain how a version of structural realism can be developed using Davidson’s theory of meaning and some advantages this has over the Ramsey-sentence version of structuralism. Keywords
Realism · Semantic holism · Davidson · Worrall
Two arguments that pull in opposite directions set the stage for modern discussion of realism in the philosophy of science.1 On the one hand philosophers of science are drawn towards realism by the so called no miracles argument (NMA, hereafter). The best explanation of the predictive success of scientific theories is that they are true (or at least nearly true). On the other hand the history of science tells against such optimism. Past science is littered with predictively successful theories which turned out to be false; and not just a little bit false (or in other words nearly true) but apparently radically false. Gravitational forces, the electromagnetic ether, caloric were all part of the ontology of once successful theories—but by our current lights no such things exist. So we have no reason to infer (as advocates of the NMA do) that our current successful theories are true or nearly true. I shall follow the literature in calling this anti-realist argument the pessimistic meta-induction (PMI, hereafter). 1 I use the words realism and anti-realism in the standard way in the philosophy of science literature. Realists
believe scientific theories are literally true or nearly true, where truth is understood in a non-epistemic way and that we have good reason to believe in their truth. Anti-realists deny all or some part of this. So scepticism may be a form of anti-realism. Van Fraassen, the leading so called anti-realist, advocates precisely such a sceptical view. J. Ritchie (B) Department of Philosophy, Trinity College Dublin, College Green, Dublin 2, Ireland e-mail:
[email protected]
123
86
Synthese (2008) 162:85–100
An attractive response to this problem is structural realism. Structural realists claim that we can divide our theories into two parts: structure and ontology. The structure of the theory, they claim, is retained through scientific revolutions but often the ontology is discarded. NMA can then be enlisted to support the claim that the structure of the theory is approximately true and PMI can be used to dismiss the idea that science aims to get the ontology right; the best of both worlds. As good as this sounds, it has, unfortunately, proved rather hard to say what exactly the structure of a scientific theory is in a non-trivial way. In this paper I am going to trace the development of the structuralist idea through the work of John Worrall. I want to suggest that properly thought through what the structuralist is offering or should be offering is not an account of how to divide up a theory into two parts— structure and ontology—but (perhaps surprisingly) a certain kind of theory of meaning. The structure of structural realism is of a semantic kind. Structural realists should be semantic holists. I develop this idea through Donald Davidson’s work. I’ll end with a brief comparison of this reformulated structuralism with entity realism. The comparison will allow us to see what is distinctive about these highly general solutions to the PMI and better understand the role NMA-type considerations play in a realist philosophy of science.
1 Fresnel, Maxwell and the ether Worrall’s (1989) original argument for structural realism uses an historical example which nicely dramatises the powerful but opposing pulls of the NMA and PMI. At the beginning of the 19th Century, the Newtonian corpuscularian conception of light faced a number of serious problems; in particular it seemed unable to account for various diffraction phenomena. A French engineer, Augustin Fresnel, showed that many of these problems could (at least in principle) be solved if light is described as a transverse wave propagating through a luminiferous ether. Henri Poisson an advocate of the rival corpuscularian view, thought this led to a ridiculous result. If Fresnel’s theory were right there should be a bright white spot in the centre of the shadow cast by an opaque disc. The experiment was performed and the white spot observed; a surprising prediction gets a surprising confirmation. Surely such a result shows that light must be as Fresnel described it. Later physics does not support this view. Subsequent work by Maxwell and Einstein consigned the ether to history. If there’s no ether, there’s no vibrating in the ether and there is nothing in the world like Fresnel’s description of light. Surprising predictive success pulls us towards realism; knowledge of the history of science pushes us back to anti-realism. Structural realists claim we can have our cake and eat it here. If we look more closely at Fresnel’s theory we see that certain aspects of it are retained. Specifically, the mathematical equations with which Fresnel described the relative intensities of reflected and refracted light reappear in Maxwell’s theory unaltered. (See below.) Admittedly the meaning of the key terms is different. For Fresnel the equations described a mechanical oscillation in a jelly-like stuff; for Maxwell and his successors they describe a displacement current in an electric field. But this provides a philosopher with a neat way to parse the theory to respect both realist and anti-realist intuitions. What Fresnel got right was the structure of light as encoded in the mathematical equations; what he got wrong was its underlying nature. So we should be realist with respect to the structural claims of science and anti-realist about the claims that go beyond that structure.
123
Synthese (2008) 162:85–100
87
Fresnel’s equations for the relative intensities of reflected and refracted light R/I = tan(i − r )/tan(i + r ) R /I = sin(i − r )/ sin(i + r ) X/I = (2 sin r. cos i)/ sin(i + r ) cos(i − r ) X /I = 2 sin r. cos i/ sin(i + r ) I 2 , R 2 and X 2 represent the intensities of the incident, reflected and refracted beams respectively for the component of light polarised in the plane of incidence. I 2 , R 2 , X 2 the same for the component of polarised light orthogonal to the plane of incidence. The angle of incidence of the beam is represented by i and the angle of refraction by r . Of course, the Fresnel–Maxwell example is atypical. It is not normally the case that successor theories retain exactly the same equations as the earlier theory. Nevertheless structural realists typically hope that by concentrating on the mathematical structure it is easier to see the continuities between earlier and later theories; and indeed, textbook derivations of, say, various equations of Newtonian Mechanics from Special Relativity by setting the speed of light to infinity suggest, perhaps, this hope is well founded. 2 What is structure? The idea clearly has some intuitive appeal but what does it amount to; what does it mean to believe only in the structure of the theory? Not much as it turns out. The clearest way to make sense of the idea of a mathematical structure is in set theoretic terms. A structure so defined just asserts that there exists some relation or relations which satisfy some domain of objects. Nothing is asserted of the relations beyond this. But this claim cannot be what a scientific realist wants as a description of what is held true in a scientific theory for it is too easy to satisfy such a structure. This is in fact a very old point. W.H. Newman made it against a similar proposal from Bertrand Russell.2 Newman puts it this way: No important information about the aggregate A, except its cardinal number, is contained in the statement that there exists a system of relations, with A as a 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 A. (Newman 1928, p. 140. Italics in the original) To know the structure is to know no more the cardinality of your domain. Scientific realists, one would hope, wish to be a little more ambitious in their claims.3 3 The Ramsey sentence approach Pure structure tells us nothing. So structuralism must be structure plus something else. The challenge for advocates of structural realism is to tell us what that something else is. One 2 Russell (1927) is taken by advocates of SR as an early version the structuralist project although its moti-
vations are quite different from those of Worrall’s paper. See Psillos (2001) for what he calls the upward and downward paths to SR. 3 In the case of an uncountably infinite domain, the Lowenheim–Skolem theorem shows us that if we know only the structure, then we don’t even know the cardinality. This result is the basis of Putnam’s (1981) modeltheoretic argument. See Ketland (2004) for a careful and rigorous presentation of the Newman objection.
123
88
Synthese (2008) 162:85–100
suggestion that has recently been defended by Worrall in work with Elie Zahar (2001) is to reformulate our scientific theories using the Ramsey sentence method. To do this we need to divide the predicates of our theory into two types: the observational and the theoretical. We can then eliminate any direct reference to theoretical terms by replacing them with existentially bound variables. So to borrow an example from van Fraassen (1997), consider the sentence: “water is composed of oxygen and hydrogen atoms”; and suppose further that we consider water to be an observational term and oxygen, hydrogen and atom to be theoretical terms. The Ramsey sentence equivalent of this mini-theory states: that there exists some property P, some property Q, some property R, such that P and Q are mutually exclusive and water is composed of things that have property P and R and things that have property Q and R.4 This remains structuralist about the theoretical part of our mini-theory since all we know about hydrogen, oxygen and atoms is that they stand in the relations described above to each other and the O-term water. However, what stops it being pure structuralism are those very O-terms. They ground the structure in the intended part of reality and, so Worrall and Zahar claim, this approach is not susceptible to Newman’s objection. There is some debate in the literature as to whether this account really does solve Newman’s problem (see Ketland 2004; Demopoulos and Friedman 1985).5 But even if it did, the price is hardly worth paying. To get this proposal off the ground we need a principled distinction between the observational and the theoretical terms and philosophy of science of the last hundred years or so has given us every reason to think there is no such principled distinction. Let me rehearse some of the arguments that have lead most philosophers to this conclusion. The distinction between observation and theoretical sentences is usually justified by the claim that observation sentences have a special epistemic status; they are incorrigible or the truth-value can be determined non-inferentially or they are free of any theoretical commitment. Those tempted by this view normally have one of two candidates for observation sentences in mind. The moderate view is to take observation sentences to refer to middle-size goods in the publicly observable world; sentences of the form “The needle is pointing to 5” are often offered up as examples. This is the view Worrall favours. However, such observation sentences seem to have none of the special epistemic features such an account requires. First there clearly is some kind of low-level knowledge required here to understand such a sentence. We need to know what needles are and that what they do is to point to numbers. And once we have admitted this much knowledge is necessary to formulate such observation sentences, then we start off down a slippery slope. Why not allow us to say that we equally well see that the voltmeter is registering 5 V since (let us suppose) we know it’s a voltmeter we are looking at and why not then allow us to say that we see that that voltage over the circuit is 5 V since we know that’s what voltmeters measure. Obviously as each further step requires more knowledge there is more chance that you might be wrong in what you say; it might be an ammeter instead of voltmeter that you see, perhaps the voltmeter is not working, etc. But the possibility of error exists at all stages and so cannot be used to arbitrate in any principled way between observational and theoretical sentences. Equally, it seems obvious 4 More formally and generally, given some theory formulated in a first-order language H(O1, . . . , On; T1. . .Tm) where O1, etc are the observation terms and T1, etc. are the theoretical terms, the Ramsey sentence – t1. . .C – tmH(O1. . .On; t1. . .tm). is: C 5 It is argued by these philosophers that the only extra thing you know is whatever the O-sentences say. Again knowing cardinality of your domain and certain truths about observables seems to be (to say the least) less than most realists hope for. For a response see Zahar (2001), Appendix II.
123
Synthese (2008) 162:85–100
89
that a well-trained physicist can assert some supposedly theoretical sentences, “the voltage over the circuit is 5 V” non-inferentially—such sentences are incorrigible for her. So which sentences we take to be incorrigible or inference-free varies from individual to individual and context to context.6 The more extreme view is to push back even further into mind in an attempt to find a certain, incorrigible base. This is the view favoured by Zahar (2001) in his book on Poincare. It is difficult to see how such a view can avoid scepticism. Once we push back into the mind to find our observation sentences, then it is unclear how any theory can ever help us find our way back out into the world. But even setting sceptical worries aside, this view will fare no better. Sense-data or whatever the theorist posits as an indubitable base are not, whatever else they are, sentences or predicates and so not apt to be included in a Ramsification of the theory. Our O-sentences cannot be incorrigible sense-data but eminently corrigible sense-data reports. So here, as with Worrall’s suggestion, the same problems regarding fallibility will arise. The supposed special epistemic status of O-sentences turns out to be illusory.7 There are other obvious difficulties one could raise with these proposals as soon as one considers the matter in any depth. First, we should note that I doubt anyone other than a philosopher has ever actually explicitly entertained the propositions that are offered as Osentences. Surely few working scientists have ever had the thought the needle and the number nearly coincide, let alone that I am experiencing a mostly white colour patch with a long black streak and some other black squiggles or whatever the appropriate sense-data sentence is supposed to be; and that is (to say the least) rather surprising if such sentences are supposed to form the epistemic base of science. Moreover, the O-sentences which are supposed to be the consequence of theory would (being observer sensitive for the sense-data theorist) have to take into account a whole host of information that is not normally considered relevant in deriving a prediction from a theory. If you and I look at some meter reading at different angles, then in some sense we will see different things and this difference would have to be registered in our different O-sentences. But of course, it just never is part of any scientific prediction to make such bizarre calculations, nor do I think that it ever could be. Note that none of these points against the tenability of the observation/theory sentence distinction threatens the platitude that scientists make many observations and test their theories by making such observations. Peering through microscopes, looking at tracks in a cloud chamber and measuring neutrino interactions in vast containers of cleaning fluid might all be observations in appropriate circumstances. The arguments I have offered here and that we all learnt in Philosophy of Science 101 simply show us that there is no clear way to reconstruct the language of science by neatly dividing our terms into those that are observational and those that are theoretical8 ; and without such a principled division there is no such thing as the structure of a theory exemplified by the Ramsey sentence. 6 Moreover, it is not clear that Worrall’s suggestion picks out a well defined class of sentences. Clearly not all
observation in science is a matter of reporting meter readings. What are the equivalent O-sentences, to take one example, for imaging devices like microscopes, telescopes and at the fancier end of the spectrum cloud chambers? 7 In fact, a moment’s reflection would suggest our chances of being wrong here are much greater. How easy is it to convey the subtle distortions in shape and colour that objects have as they appear to us in any kind of report. One need only consider how poor many of us are in the art class room to understand the difficulties we have in recreating, let alone describing, the detail of what we actually see. 8 Even leading contemporary empiricists agree with this. Although van Fraassen does distinguish distinction between the observable and the unobservable in terms of what is visible to the unaided senses that is a distinction between observable and unobservable entities, not sentences. As he makes clear elsewhere (van Fraassen 2000) the language of physics remains as theory-infected as ever. There are no pure observational predicates.
123
90
Synthese (2008) 162:85–100
Before moving on, we should note that there is another ways to utilise the Ramsey sentence framework without invoking the old empiricist distinction between observation and theory sentences. Lewis (1979) offers the clearest account of this. Instead of drawing a distinction between observational and theoretical predicates, we draw the distinction in terms of those predicates that are antecedently understood and new, unknown theoretical terms that are introduced into our language. This sidesteps the concerns we have had with the idea of an observation sentence—but it can’t possibly provide the basis for an interesting account of structure. Such a distinction is obviously highly context sensitive and historically contingent: what a 21st Century scientist antecedently understands and takes for granted is quite different from what a 17th Century scientist would have understood. An attempt to find a set of antecedently understood terms free from such contingencies would push us back to a search for O-terms along the lines offered by Worrall and Zahar; and that as we have seen is hopeless. A second equally important reason for rejecting the Ramsey sentence method can be developed from a criticism of structural realism offered by Ladyman (1998). As Ladyman points out there is an apparent difficulty in regarding the Ramsey sentence approach as a solution to the problem of the PMI that structural realists hoped for. It is natural to understand the Ramsey sentence as explicating the meaning of theoretical notions,9 so terms like ‘is an electron’ or ‘light’ are implicitly defined by their place in the overall structure of the (Ramsified) theory. Theory change is (usually at least) driven by the empirical inadequacy of earlier theories. Such inadequacy can only manifest itself as entailing a false O-sentence. So a new theory must entail different O-sentences. That means the structural relations of the theoretical sentences and in turn terms must also change. So although Fresnel and Poisson might use the same term, ‘light’, in describing their theories, they in fact mean something different. Ladyman takes this to cause trouble for the Ramsey sentence view because changes in meaning imply changes in the purported reference of key theoretical terms. But if that is the case we have a quite radical form of discontinuity from one theory to the next. Rival theories are not even talking or purporting to talk about the same thing so it is difficult to see in what sense one could be an approximation of the other. There is something to this but I think it is open to the advocate of structural realism to claim that their account of theoretical approximation and approximate truth will not require constancy of reference.10 A Ramsey sentence may be approximately true despite the fact that there is no entity that corresponds exactly to the Ramsified description because the existential claims made using the Ramsey sentence are themselves approximately true. More concretely if we understand Fresnel’s theory in the Ramsey-sentence way, although there may be no thing which corresponds to his theoretical term ‘light’, there does exist something which nearly does. But even if something like that can be made to fly, the meaning variance Ladyman highlights leads to other problems. 9 This could be denied. One might claim that structuralists typically want to say that their kind of realism
involves asserting less than standard realism. So the Ramsey sentence must have less content than the unRamsified theory. Cruse (2005) suggests this is a way to distinguish his position, which he calls Ramsey sentence realism, from structural realism. Some passages of Zahar and Worral (2001) suggest this is their view (see Sect. [B] in particular); although they are far from clear about this. If the structuralist takes this line, he still owes an account of how theoretical terms get their meanings and how that is related to the Ramsey sentence. In other words, he replaces the problems I outline below with a mystery about the meaning of theoretical terms. It is clear, nonetheless, that some people who have called themselves structuralists have held the view that the Ramsey sentence also works as a theory of meaning, e.g. Maxwell (1970). 10 I have to thank John Worrall for making me see the force of this point. Cruse and Papineau (2002), while not identifying themselves as structural realists, make a proposal along these lines.
123
Synthese (2008) 162:85–100
91
Consider someone who is converted from the old corpuscularian theory of light to Fresnel’s new theory. Such a person might say: “I used to think light wasn’t a wave but now I believe it is.” If we accept the Ramsey sentence account as giving the meaning of theoretical terms, then understanding such sentences becomes awkward. What does our imagined physicist mean when he uses the term ‘light’. If by ‘light’ he means to use the term as it appears in Fresnel’s theory, then the first half of his statement involves asserting he used to believe something contradictory since the meaning of the term ‘light’ involves, in part, the idea that it is a wave.11 But we don’t want to say that. We want to say only that he had a false belief or, more accurately, that he now believes his old belief was false. If, on the other hand, we understand this scientist to mean by ‘light’ the term as used in the old corpuscularian theory, then either he asserts something contradictory in the second part of his statement or the ‘it’ used in the second part of the statement does not refer to the same thing as ‘light’ in the first half of the statement. Either way we have some undesirable consequence. The individual is either asserting something contradictory or we are forced into some unnatural interpretation of his use of pronouns. What this shows us is that we require some degree of stability in the meaning of theoretical terms like ‘light’ not just to make sense of approximation or approximate truth (if we do in fact need it there) but also to keep track of some facts about ordinary language usage. We want to be able to say is that in cases like the one above a theoretical term still means the same thing even though some of the individuals surrounding beliefs have changed. The Ramsey sentence view does not allow us to make these simple claims. That, I suggest, gives us another reason to search for an alternative view.12 4 If not the Ramsey sentence, what then? Is there another way of saving structural realism from vacuity? The Ramsey sentence approach embodies two commitments. The first is epistemological. The Ramsey sentence method (at least as articulated by Zahar and Worrall) is a formal way of presenting a kind of empiricism. The arguments given above against the untenability of the observation/theory sentence distinction should lead us to reject this part of the approach. But there is also a semantic aspect to the Ramsey sentence method. As we have just seen, theoretical terms in the Ramsey sentence are defined by their role in the theory; that is by relation to other theoretical terms and ultimately their relation to O-terms. This is a version of semantic holism. That is to say, it is a kind of theory of meaning in which the idea of reference (at least for theoretical terms) plays no or only a secondary role. Language for meaning holists is related to the world in some other way than by first fixing the reference of individual terms and building up to sentences, theories, etc. from there. I suggest that a more promising version of SR should fix on this aspect of the Ramsey method. SR then is the claim, in effect, that the best way to be a scientific realist is to be a meaning holist. What is structural in such an account is the way words relate to the world.13 In the rest of the paper I explore this idea through the work of Donald Davidson. 11 I am for the sake the simplicity ignoring the fact that the term ‘wave’ is equally theoretical. 12 Papineau (1996) offers one way out of this problem: define an analytic core for some theoretical term. If
the same term in different theories satisfies those same core descriptions you are talking about the same thing. Zahar (2001) suggests that the problem arises from standard referential Tarskian semantics and we shall have to wait for the development of an alternative semantics in category theory. Below I show how we can develop something in the spirit of structural realism which does not rely on defending the analytic-synthetic distinction and employs standard Tarskian semantics. 13 Some people tell me this isn’t structural realism. Some of these people call themselves structural realists so I guess I should take their worries seriously. I’m not much concerned, to be honest, about labels but it should
123
92
Synthese (2008) 162:85–100
5 Radical interpretation and meaning holism Davidson, as is well known, broadly speaking follows Quine in his construction of a theory of meaning. The question which he begins with is: how can a competent interpreter come to understand the utterances of an alien language. The general strategy is as follows: The intrepid interpreter . . . seeks to assign a propositional content to the utterances of a speaker. In effect he assigns a sentence of his own to each of the sentences of a speaker. To the extent that he gets things right, the interpreter’s sentences provide the truth conditions of the speaker’s sentences. (p. 210, 2001) A radical interpreter tries to discern the attitudes the alien language user takes to various utterances made and then tries to correlate them to sentences in his own language, in attempt to generate sentences of the form p is T iff s, so called T-sentences, where p is the name of the sentence in the foreign language and s a sentence used by the radical interpreter. He is constrained in two further ways. Two key principles must be applicable if a speaker is to be interpretable: the Principle of Coherence and the Principle of Correspondence. The Principle of Coherence prompts the interpreter to discover a degree of logical consistency in the thoughts of the speaker; the Principle of Correspondence prompts the interpreter to take the speaker to be responding to the same features of the world that he (the interpreter) would be responding to in similar circumstance. Both principles can be (and have been) called principles of charity: one principle endows the speaker with a modicum of logic, the other endows him with a degree of what the speaker takes to be true belief about the world. (Davidson 2001, p. 211) These principles are not just evidence of the generosity of the interpreter. Unless we assume that people’s beliefs form a generally rational pattern, then we have no reason to think that they have beliefs at all; and since the way we identify people’s beliefs about their local environment is (normally) through the objects that cause those beliefs, there is no possibility of a systematic mismatch between what is believed by our interpretee and how things are.14 Being mostly right about the world about us and mostly consistent are preconditions of being a thinker at all. Davidson in later work has expressed the centrality of these principles to the possibility of language and thought using the idea of triangulation. Our acquisition of a language depends upon an interaction with another speaker and a shared environment. It is through Footnote 13 continued be clear calling such a position structural realism has a point. It is realist because (as we shall see below) it allows one to say that are current scientific theories are approximately true while avoiding the problems of the PMI. It is structural because the account of the relation of words to the world it gives is of a structural or holistic sort. Being a naturally conciliatory sort of person, I’d be quite happy if some people who thought of themselves as just straightforward unqualified realists felt they could embrace this position. Other people who have at least flirted with the label structuralism have also suggested that their position may be acceptable to those who call themselves unqualified realists. See Chakravartty (2004) for example. 14 Explained in this fashion the Principle of Correspondence may sound rather similar to the idea of the observation sentence. However as Davidson makes clear elsewhere it does not follow from the fact that in order to communicate we must share elements of a common environment that we must also have some basic non-theoretical vocabulary in common. See essay 10 of Davidson (2001) “I do not see how to draw the distinction between observation and theoretical sentences” (p. 145) and “Communication begins where causes converge: your utterance means what mine does if belief in its truth is systematically caused by the same events and objects.” (p. 151). See also his discussion of Schlick and Neurath in essay 11 of the same volume.
123
Synthese (2008) 162:85–100
93
this three way interaction that the very notions of truth and objectivity emerge. Language and communication requires that we are able to share the same contents. What we initially share in the situation of radical interpretation is a causal interaction with some object in our environment. When our beliefs are systematically caused by those same objects (or events) then we share the same content. We learn the very idea of the separateness of the world and other people’s thoughts about it when we find ourselves triangulating but no longer agreeing with one another. That shows that you must represent the world differently from me; and possibly that the way I think of things is wrong. Here then we get the ideas of objectivity and truth. Such concepts require that we can form correct beliefs about others thoughts and the world. We find through the process of triangulating a difference between the way others understand the world and possibly, through further investigation, how the world really is. But to get this far we must have agreed upon a lot through our very process of triangulating. Error only makes sense against the background of largely true, shared belief which makes communication possible in the first place; and that communication must begin with elements of our shared environment with we which we both causally interact. In other words, a kind of externalism underwrites the principle of correspondence. The content of my belief is fixed by what is in common between the causes of my beliefs and the other speaker when we triangulate and initially begin communicating.15 With a sufficient number of these T-sentences, constrained by the Principles of Charity, the radical interpreter tries to construct a theory of truth along the lines of Tarski. When constructing the theory of truth, the radical interpreter aims to make the empirically discovered T-sentences come out as theorems. This, as Davidson often reminds us, inverts Tarski’s method. Tarski assumes a sameness of meaning relation between the language and metalanguage and defines truth via satisfaction. Davidson begins by assuming a univocal notion of truth and analyses sameness of meaning. Having postulated a theory of truth to make sense of his alien interlocutor, the radical translator will have a theory of how the parts (the words) of the alien language relate to form whole sentences and he can use this to generate sentences in the alien tongue he has never heard. He can then test the appropriateness of his interpretation out on the alien speakers and refine his theory as he goes along. What is significant about this approach for our purposes is that reference plays no essential role in linking language to the world. Reference, such as it is, falls out of the Tarski style formulation; it does nothing more than explicate how the parts of a sentence contribute to the truth-conditions of the sentence as a whole. This can be seen in two ways. First, there is no general characterisation of the notion of reference in Davidson’s account. Every time a new term or predicate is introduced a new recursive definition has to be added to the Tarski-style theory of truth. Second, Davidson’s account exhibits what he calls (again following Quine) the inscrutability of reference. Even given an ideal theory of truth for some alien language, that is a theory which satisfactorily interprets all a speaker’s utterances in all possible situations, there will be alterations we can make to the satisfaction (reference) relation which will not affect the truth-conditions of any sentence or the logical relations between sentences. “Since all the evidence for interpreting a language must come at the sentential level (for only sentences have a use in communication), the result is that there can be no evidence that one 15 Davidson (2001) explores these themes at length. See especially “Three Varieties of Knowledge”. Not everyone is convinced by these claims. See for example John McDowell’s (1996) classic criticisms in Mind and World. A full defence of them would be worthy of a separate article. For sceptics, we may chalk this up as another difficulty confronting Davidson’s programme which future workers on SR may hope to overcome or that may suggest some other holistic theory of meaning would better serve their purposes. See the closing parts of this article for an outline of other problems with Davidson’s theory of meaning. I want to thank a referee from Synthese for forcing me to elaborate on this point.
123
94
Synthese (2008) 162:85–100
of the satisfaction (or reference) relations is the right one.” (Davidson 2001 p. 78) Davidson has offered the following example. If we have a theory of truth which maps the word ‘Rome’ onto Rome, and the predicate ‘is a city in Italy’ onto cities in Italy, then our truth theory will then have as a consequence the sentence: ‘Rome is a city in Italy’ is true iff Rome is a city in Italy. But there will be other satisfaction relations which will do just as well. For example we might map ‘Rome’ in the alien language onto ‘an area one hundred miles south of Rome’ in our language and ‘is a city in Italy’ onto ‘an area one hundred miles south of a city in Italy’. So the biconditional we would get from our theory of truth would be: ‘Rome is a city in Italy’ is true iff an area one hundred miles south of Rome is one hundred miles south of a city in Italy.16 Compensating adjustments would have to be made in the rest of one’s theory of interpretation but, at least so claims Davidson, nothing fundamental would have changed as regards meaning. The meaning such as it would be what remains invariant in all these possible transformations or permutations of the satisfaction relation. Davidson employs a useful analogy to illustrate the indeterminacy which arises here. When measuring temperature we try to relate states of the world to the real numbers. There is obviously more than one way to do this. We may measure temperature in degrees Fahrenheit or Celsius, for example. Each of these represents different ways of keeping track of the same ‘facts’. Similarly interpreting the alien speaker in the ways illustrated above represents different ways of representing the same facts about a speaker’s attitude to states of the world. The only thing that changes is the language we understand the alien to be speaking. But this is no more significant than the change from Fahrenheit to Celsius.17 Reference only plays a role in allowing us to a construct a theory of truth which enables us to understand a potentially infinite number of sentences. But since many reference relations will do this job just as well there is no such thing as the reference of any term. What does the real work in Davidson’s theory of meaning is the identification of T-sentences constrained by the principles of coherence and correspondence. As Davidson (2001) himself puts it: “Correct interpretation keeps track of a complex pattern, and locates particular sentences and attitudes within it. But the counters we use to represent this pattern, namely our sentences, can represent it in more than one way.”(p. 79) We can, without doing violence to the meaning of this passage, replace the word pattern with structure and so see the sense of calling Davidson’s theory of meaning a kind of semantic structuralism. 6 The principle of charity How can this account of meaning help with scientific realism? In the general course of radical interpretation one must assume that most of what your interlocutor is saying is true. This is the only way one can get a purchase on what a speaker means given what you know about the environment. When we move from the mundane and everyday talk to the “exotic suburb”18 of our communicative practice that is science, there seems no need to continue to be charitable. When it comes to physics and cosmology, most people have a multitude 16 Such permissible transformations of the satisfaction relation involve no change in the truth conditions of the sentences being interpreted. That’s why by Davidsonian lights they involve no changes which affect meaning. 17 As Davidson points out there is an important disanalogy between these cases. When measuring temperature we can adopt as a convention whether to use to Fahrenheit or Celsius. In the course of radical interpretation we cannot similarly avoid indeterminacy by declaring: “I’m speaking English”, since this utterance like every other needs to be interpreted. 18 Davidson refers to science as such in Davidson (1986).
123
Synthese (2008) 162:85–100
95
of false beliefs. To turn this account into a version of scientific realism we need to add in some way to Davidson’s story. I suggest that answer is straightforward enough and already present in Worrall’s original paper. We extend the principle of charity when interpreting our old theories in light of our new theories just whenever and wherever that theory produces a surprising empirical success. Let me sketch how I see this as working by again looking at the Fresnel case. The analogues of the radical interpreter and alien speaker for us will be the advocate of the new theory and an advocate of the old. What we want to show is that it is possible to interpret Fresnel’s theory in broadly speaking Maxwell’s terms; and to show through such an interpretation that Fresnel’s theory is (approximately) true in important respects. Let us then consider the following three sentences which we may imagine Fresnel or an advocate of Fresnel’s theory may have uttered in explaining their theory. (A1) Light is a transverse wave (A2) Light is an oscillation in a luminiferous ether (A3) The luminiferous ether is a jelly-like substance. Our guide to interpretation should be the question: which parts of the theory are necessary in generating the surprising successful predictions of Fresnel’s theory. And of course we can see from our perspective that what was important was the idea that light was a transverse wave and not what Fresnel’s speculations were concerning the medium in which it vibrated. A reasonable interpretation of (Al) would be quite simply: (I1) “Light is a transverse wave” is true iff light is a transverse wave. Sentence (A2) is more difficult. One plausible translation would be: (I2) “Light is an oscillation in a luminiferous ether” is true iff light is an oscillation in something.19 Such an interpretation, obviously, will have the consequence that we understand ether as referring to just whatever light vibrates in. Thus we take (A2) to be true, although relatively uninformative. Then we should interpret (A3) as follows: (I3) “The luminiferous ether is a jelly like substance” is true iff the something light vibrates in is a jelly-like substance. Our interpretation obviously means that we take (A3) to be false. But that should not bother us since claims about what exactly Fresnel thought the ether was played no role in the surprising predictive success of the theory. We only need to apply the principle of charity where the success of the theory demands it. Note that our interpretation does not ascribe to Fresnel any knowledge that he didn’t have. We don’t need to pretend that he knew anything about the electromagnetic field. It just provides a sensible characterisation of the view that although Fresnel was right about important aspects of light (it was a transverse wave) he was wrong about others (it oscillated in an elastic medium). Given such an interpretation we can then understand why Fresnel’s equations are retained in Maxwell’s later theory.20 19 This is a bit like the basic structuralist idea in which we replace a theoretical term with an existentially quan-
tified variable. However, it remains a possibility on this account that as science develops and our knowledge increases such theoretical placeholders may be replaced by more specific theoretical terms. Indeed intuitively this has what has happened as our knowledge of light has increased. When we try to understand classical electromagnetism in terms of, say QED, we want to recover something of the concept of the classical field. 20 It will of course often be more difficult to correlate the two theories than this. In general the interpretation of the old theory in terms of the new will not make those sentences come out true but (in some appropriate sense) approximately true. I don’t discuss the details of this since this is an issue which all advocates of structural realism will have to address but is not explicitly discussed in the canonical presentations of the position.
123
96
Synthese (2008) 162:85–100
My Davidsonian account also saves the every day facts of usage which caused trouble for the Ramsey sentence approach. Using our theory of truth to understand Fresnel, we map the term ‘light’ in his language onto ‘light’ in our language. So we can say, as we want to say, that Fresnel’s is talking about the same thing as we are when we use the term.21
7 Whig history versus clairvoyant philosophy I expect many philosophers of science will be unimpressed by this account. “Look,” they’ll say, “on your account you may be able to tell a nice story about what is and is not retained from one scientific theory to the next but that is because it is entirely post hoc. It is an account of the old scientific theory told from the perspective of the new theory. It is Whig history of science.” To which I can only reply: yes, it is—but so what. Most structural realists presumably think post hoc rationalisation is a bad thing because they believe that the job of the realist is to provide some concrete account of what part of our current scientific theories it is that one should hold true; and, crucially, one should be able to say what this is in advance of the theory being replaced by some better theory. Structural realism as originally conceived by Worrall looks like such an answer—keep the equations. The obvious but important point to make against anyone who is tempted by such thoughts is that interpreting past theories in terms of our current theories is (since we don’t know what the final true theory is) all we can ever do. If realists (or anti-realists) think that realism is only tenable if one can describe the bit of one’s theory which is now true and will never be revised, then they are engaged in a hopeless quest for clairvoyance. A quest which not even the original (and recall empty) version of structural realism satisfies since there are changes in the mathematical equations in all but the most special cases. The charge of Whig history is only serious if one is involved in imputing beliefs to past scientists that it is impossible or unreasonable to believe they held. But our story doesn’t do this. We have just given a very minimal interpretation of the term ‘ether’ as it features in Fresnel’s theory; one which is in fact made entirely reasonable in the context of the development
21 There may be cases where the best interpretation of the scientists in question takes them to fail to disambiguate key notions that we take to be different. Perhaps we would need to do that with the term ‘mass’ for example when interpreting pre-Einstein physicists. In that case, in our best interpretation we may have past scientists making inferences which from our perspective are fallacious since they equivocate on the meaning of mass. Similarly their may be occasions where we wish to say that although two individuals are using the same term, they mean different things by it. To take one extreme example, we would wish to say something like this with regard to the early history of the term ‘meson’ and its multiple meanings. (See Hacking’s 1983 discussion of this, p. 88ff.) Neither case creates a problem for this sort of analysis. We can readily explain the cause of the apparently fallacious inferences in the former case; and in the latter case we don’t in fact want to say that such scientists hold contradictory views. In all our accounts of the beliefs of past scientists we need to balance our interpretation of key theoretical terms by maximising the truth of success-relevant claims (as judged from our present theory) against overall coherence of the theory offered. We can allow occasional incoherence (through failure to disambiguate different notions of mass, for example) if there is a reasonable explanation for it. In this regard, this Davidsonian account is similar to Braddon-Mitchell’s (2005) (or more accurately Braddon-Mitchell is elaborating an account which has a lot in common with Davidson’s work) in that we allow context of use to play a role in our determinations. However, what we are determining is not reference but the truth conditions of various sentences. It is also compatible with this suggestion that there may be no best unique interpretation. This does not, however, threaten our claim that the old theory is approximately true.
123
Synthese (2008) 162:85–100
97
of Fresnel’s work since he never had a developed theory of what the ether might be beyond a suggestion that since light was a transverse wave the ether must be an elastic solid.22 We do well to recall that the theories we have now are only as good as the tests we have subjected them to; new tests may force us to produce new theories. What these theories will be we cannot tell. That is not to say that scientists do not often have a good sense of the bits of their current theories that are most under strain. Most particle physicists, for example, agree that the need to postulate so many unexplained constants in the Standard Model indicates that there is something wrong or inadequate about their current theories; and thus much energy is expended in trying to reformulate their theories to avoid these problems. But the history of science also shows that changes in theory sometimes involve revisions in parts of science that many scientists have taken to be the most secure. The scientific revolutions that took place at the beginning of the twentieth century led to revisions in the fundamental categories of space, time and causation. What we should hope for, or at least the realist must hope for, is that however things turn out there will be a reasonable story to tell through which we can interpret some, perhaps most of the sentences in the old theory as true (or approximately true) in the new one.23 This means that realism is in part an attitude. But that should not embarrass anyone. We can say in some detail how things are if we are realists. We just need to admit that as, to use Peirce’s phrase, “contrite fallibilists” we might be wrong in some of the detail of these claims. Whatever the drawbacks of Whig history are thought to be, they should not force us to attempt clairvoyant philosophy.
8 Compare and contrast—Ramsey, Davidson and Hacking I want to end by rehearsing the benefits of this approach as compared to the Ramsey sentence method and by highlighting some interesting points of comparison with entity realism. The Ramsey sentence method has two objectionable consequences that we have noted. First it requires an untenable distinction between observation and theory sentences. Second if two theories disagree in observational consequences, since the theoretical terms get their meaning by relation to each other and the O-sentences, then none of the terms in the two theories can mean the same thing. This arguably makes it difficult to see how one theory can be an approximation of another and it certainly, as we have seen, does violence to our intuitions about what’s going on when an individual changes their mind. The Davidsonian version of structural realism avoids these unwelcome consequences. No distinction between observation and theory sentences is required to set up a Tarski-style theory of meaning and we can, in constructing our theory of meaning, avoid the absurd consequences of the unconstrained holism of the Ramsey sentence. Although meaning is fixed by the truth-conditions of sentences and the pattern of inferences between those sentences, we can using Davidson’s theory make sense of the very reasonable thought that Fresnel was talking about light and its properties but got some of those properties wrong. What allows us to do this is the principle of charity. In fact all we have added to Davidson’s account is a claim about when we should apply the principle in the history of science; we should take past theories to be (mostly) 22 As Kitcher (1993), pp. 146–149 reports the history of ether theories is of one of continual failure. Each new suggestion as to what in more detail the ether might be was eventually shown to be untenable for various reasons. 23 My account does not guarantee this can be done. Perhaps in other cases where there have been scientific revolutions there is no reasonable story to tell which can explain the success of the past theory in terms of its relevant similarity to the new theory. It is on such ground that the realism- anti-realism controversy should be fought.
123
98
Synthese (2008) 162:85–100
true in so far as they have had striking empirical successes and seek to retain the elements essential to that success as judged from our new theory. Entity realism also has at its heart a theory of meaning; Putnam’s (1975) causal theory of reference. When applied to the issue of scientific realism Hacking (1983) adds to Putnam’s account a story about when to apply the principle of charity or the benefit of the dubbed as Putnam calls it. When we can manipulate the entities in question we should be confident that no matter what theoretical changes our account of entities undergoes, we are still talking about the same thing. The table below provides a useful means of comparing entity and structural realism. Entity realism vs. Structural realism Entity realism
Structural realism
Theory of meaning
Causal theory of reference
Principle of Charity
Benefit of the dubbed
When can the Principle of Charity be applied?
When the entities can be manipulated
Holistic theory of meaning (Davidson, conceptual role semantics: Peacocke, Brandom.) Principles of Coherence and Correspondence (for Davidson) When there is a surprising predictive success
PMI presents a problem to realist philosophy of science because it highlights discontinuity in the development of the sciences, particularly mathematical physics. Realists want to be able to say that what looks like discontinuity is in fact continuity. To do that they need to be able to say that scientists from the past are (to put in the most neutral way) saying similar things to present scientists. In other words they need a theory of sameness or sufficient closeness of meaning or reference. Structural and entity realism as I reconstruct them offer different ways to do this; and this I believe provides the appropriate contrast between the two views that many structural realists intuitively feel there should be. But notice too the fundamental similarities. Both theories involve an open-ended appeal to a principle of charity; a principle which licences us to say that alien utterances are broadly about the same thing or mostly true. To make these theories applicable to the scientific realism debate then, we need a story about when it is appropriate to extend these principles to reference to unobservable entities (in the case of entity realism) and to high-level scientific theories (in the case of my Davidsonian version of structural realism). Hacking provides one plausible story for entity realism; I have sketched one for structural realism.24
9 Conclusion I have tried to make sense of one way of being a structuralist—by elaborating a scientific realist position making use of an holistic theory of meaning. We might call this position semantic structuralism. It has, as I have stressed, many advantages over the Ramsey sentence approach. 24 My table implies that their can be disagreement among realists along two dimensions. First in deciding upon the correct kind of theory of meaning; then in formulating the grounds for application of the principle of charity. Another possible strategy, not covered by my table, is to claim that there is more than the right theory of meaning for understanding a particular scientist’s utterances varies from context to context. This is appears to be Kitcher’s (1993) view.
123
Synthese (2008) 162:85–100
99
This Davidsonian account frees structural realism from an unattractive version of empiricism and offers, I believe, a more promising answer to the PMI. But there are of course problems too. Davidson’s specific approach to meaning requires a language which is amenable to a Tarski-style theory of truth. As is well known, this presents certain problems for Davidson in developing his theory as a general account of all the idioms of a natural language. For example, there is as yet no satisfactory account of how to deal with modality within Davidson’s framework. Moreover, some philosophers have suggested that science itself cannot be adequately represented in a first-order language.25 These may or may not prove to be insurmountable problems for a Davidsonian theory of meaning.26 But even if my particular suggestion fails, my analysis of what is distinctive about structural realism opens up many possible avenues of research. Davidson’s theory is just one possible form of holism on offer. Brandom’s (2005) development of Wilfred Sellars’ inferential role semantics and Peacocke’s (1996) conceptual role semantics to name only two of the most recent and famous offer alternative ways to develop semantic holism. I suggest that by engaging with these views more robust and nuanced formulations of structural realism will emerge. Structural realists have unnecessarily burdened themselves with a philosophy of language which has its origins in the empiricisms of the early 20th century. They have more allies amongst contemporary philosophers than they perhaps realise. Acknowledgements I’d like to thank the helpful comments of one reviewer from Synthese which improved my discussion of Davidson. I’d also like to thank John Worrall for encouragement and many useful comments on an earlier draft of this paper. He does not agree with me but the paper has been vastly improved by his close critical attention. Part of this research was undertaken while I was an IRCHSS postdoctoral fellow. I would like to thank them for their support.
References Braddon-Mitchell, D. (2005). The subsumption of reference. British Journal for the Philosophy of Science, 56, 157–178. Brandom, R. (1994). Making it explicit. Cambridge: Harvard University Press. Chakravartty, A. (2004). Structural realism as a form of scientific realism. International Studies in the Philosophy of Science, 18, 151–171. Cruse, P. (2005). Ramsey sentences, structural realism and trivial realization. Studies in the History and Philosophy of Science, Part A, 35, 557–576. Cruse, P., & Papineau, D. (2002). Scientific realism without reference. In M. Marsonet (Ed.), The problem of realism. London, Aldershot: Ashgate. Davidson, D. (1986). Reply to Quine. In E. Lepore, & B. Mclaughlin (Eds.), Actions and events: Perspectives on the philosophy of Donald Davidson. Oxford: Blackwell. Davidson, D. (2001). Subjective, intersubjective and objective. Oxford: Clarendon Press. Demopoulos, W., & Friedman, M. (1985). Critical notice: Bertrand Russell’s the analysis of matter: Its historical context and contemporary interest. Philosophy of Science, 52, 621–639. Hacking, I. (1983). Representing and intervening. Cambridge: Cambridge University Press. Ketland, J. (2004). Empirical adequacy and Ramsification. British Journal for the Philosophy of Science, 55, 287–300. Kitcher, P. (1993). The advancement of science. Oxford: Oxford University Press. Ladyman, J. (1998). What is structural realism? Studies in history and philosophy of science, 29, 409–424. 25 This is suggested in Ladyman (1998). 26 One possible response is of course to take such objections as a challenge. One might claim that just as it is,
I think, often rightly complained that Davidson’s account pay too little heed to empirical results in psychology or cognitive science which might or probably should force him to change the way he understands how it is we come to formulate a theory of meaning, the same I think can be said of contemporary physics. To take Davidson’s project further one would need to be sensitive not only to how contemporary science describes our minds and the minds of others but how it describes the non-human world around us.
123
100
Synthese (2008) 162:85–100
Lewis, D. (1979) “How to define theoretical terms”. Journal of Philosophy, 67, 427–446. Maxwell, G. (1970). Structural realism and the meaning of theoretical terms. In S. Winokur, & M. Radner (Eds.), Analyses of theories and methods of physics and psychology: Minnesota studies in the philosophy of science (vol. IV). Minneapolis: University of Minnesota Press. McDowell, J. (1996). Mind and world (2nd ed.). Harvard University Press. Newman, M. H. A. (1928). Mr. Russell’s ‘causal theory of perception’. Mind, 37, 137–148. Papineau, D. (1996). Theory-dependent terms. Philosophy of Science, 63, 1–20. Peacocke, C. (1996). A study of concepts. Cambridge: MIT Press. Psillos, S. (2001). Is structural realism possible? Philosophy of Science, 68, S13–S24. Putnam, H. (1975). The meaning of ‘meaning’. In Mind, language and reality: Philosophical papers (vol. 2, pp. 215–271). Cambridge: Cambridge University Press. Putnam, H. (1981). Reason, truth and history. Cambridge: Cambridge University Press. Russell, B. (1927). The analysis of matter. London: Kegan Paul, Trench, Trubner. van Fraassen, B. (1997). Structure and perspective: philosophical perplexity and paradox. In M. L. Dalla Chiara et al. (Eds.), Logic and scientific methods (pp. 511–530). Dordrecht: Kluwer. van Fraassen, B. (2000). Literate experience: The [de-, re-] interpretation of nature. Versus, 85/86/87, 331–358. Worrall, J. (1989). Structural realism: The best of both worlds? Dialectica, 43, 99–124. Zahar, E. G. (2001). Poincare’s philosophy: From conventionalism to phenomenology. Chicago, IL: Open Court. Zahar, E. G., & Worrall, J. (2001). Appendix IV: Ramseyfication and structural realism. In Poincare’s philosophy: From conventionalism to phenomenology. Chicago, IL: Open Court.
123