Review of Suppe: The Semantic Conception of Theories and Scientific Realism (1989)

Let’s continue our investigation on semantic conceptions of scientific theories with a review of Suppe’s well named “The Semantic Conception of Scientific Theories and Scientific Realism” from 1989.

Summary of the Book

The book starts with a helpful prologue that gives some historical context: how the semantic conception came about, roughly from philosophers with a scientific training interested in elucidating the structure of scientific theories, but unsatisfied with positivist reconstructions.

The first chapter deals precisely with what, according to Suppe, is problematic with the old “syntactic view”. I found the succint presentation of it helpful. For the record (and even more succintely), this is the idea that, assuming a language for science:

• One can distinguish observational and theoretical terms
• The observation language is a language without quantification using only observational terms (merely describing individual facts for observation reports)
• Only this observation language is given a direct semantic interpretation (in terms of correspondance with observable objects and their properties)
• Theoretical terms are only partially interpreted by means of:
  • Theoretical postulates expressed using only theoretical terms (a structural, purely internal interpreation, so to speak)
  • Correspondence rules, in finite number, linking theoretical and observational terms, which allow for (i) transcribing observation reports into theoretical descriptions and (ii) translating theoretical predictions into observational predictions.

A scientific theory is precisely composed of these theoretical postulates and correspondence rules. The implication of partial interpretation is that given any set of observation reports (any model of our observations), there are always many possible models satisfying the theory. Observations never settle a full theoretical description of what is going on. Nevertheless, it is sufficiently narrow to have empirical content.

Suppe then rehearses standard objections against this view, but interestingly, he argues that they are not fatal. This is the problem that most terms of natural languages are mixed: temperatures, distances, etc. are observable in a range but not outside a range. The division between observable or not is not one of predicates, but one of instances of their applications. But, Suppe argue, we could maintain the distinction by bifurcating our vocabulary, calling something “Observable-red” when it is visibly red, and “Nonobservable-red” when not: this is artificial, for sure, but feasible if really it helps shed light on epistemological aspects at play in science. The main problem, according to Suppe, is precisely this: it does not, actually, shed any light on scientific practice, so the problem is not so much that the division is impossible than that it is irrelevant for the philosophy of science.

Suppe then proposes his own view, first summarised in this chapter and then developed in subsequent ones. Just as with the structuralist programme reviewed in the previous post of this blog, a theory is considered extra linguistic: not a set of statements, but the collection of structures that this set describes. More precisely, we can distinguish between:

• a set of models, which are abstract structures defined on state-spaces (e.g. sequences of states), where these state-spaces are constructed from a set of properties deemed relevant from the point of view of the theory (position and momentum for classical mechanics)
• a set of physical systems, which are purely extensional conceivable states of affairs (an object following a trajectory)
  • they are theory-induced if they correspond to a model of the theory
  • they are causally possible if they correspond to how an actual phenomenon P would behave if “the idealized conditions (if any) imposed by the theory were met and the phenomenon P were influenced only by the selected parameters.”
• a set of actual phenomena that have the properties deemed relevant and define the intended domain of application of the theory

The theoretical statements can be understood to describe (be true of) any of these three kinds of entities. The first really constitutes the theory and, I guess, serves the purpose of analysing its content. One can define various kinds of laws on the basis of these structures: laws of coexistence, of succession, of interaction, etc. The second kind of entity, physical systems, mediates between the theory and the phenomena. They are phenomena idealised. The aim of the theory is to correctly identify the causally possible physical systems (that it be the same set as the theory-induces ones). The third kind of entity is what we have empirical access to, what the theory aims at explaining, etc. With regard to them, we can say that the theory is “quasi-true”.

The relation between the phenomenon and the physical system can be made using a theory of the experiment. The idea is that a lot of background knowledge is assumed when applying a theory, which is sufficient to warrant the robustness and relevance of our observations. We cannot ask all theoretical terms to have an operational definition: this would amount to either refuse to theorise with non-operational terms, which isn’t acceptable, or have theories relating theoretical and operational terms, which creates a regress (how to we test these theories?). In the end, the aim of having a contact between experiment and theory can be fullfilled without a division of our vocabulary, but rather by relying on this understanding of observation:

Our ability to see that φ consists in our having a visual sensory experience such that, given what we already know about the objects referred to in ‘φ’ we could not be having that experience unless φ were the case

Assuming this, a theory can be partly justified by induction (but another chapter puts some caveats on induction, this should be taken as a mere philosophical reconstruction): from the fact that some phenomena are indeed captured by the theory (they would behave as the theory says in ideal conditions), we can infer that all phenomena in the intended scope of the theory are such. (I have defended similar views)

There is something close to recent fictionalist trends in this picture (see here and here for instance), particularly the idea that physical systems are counterfactual (non-actual) systems that play a mediation role.

According to Suppe, this picture is more faithful to scientific practice. We can see what the problem is with the positivist view: there are two stages for interpreting the theory, one from the theory to the physical system, and one from the physical system to the actual phenomenon. The correspondence rules of logical empiricism lumps all this, mixing semantic, theoretical and practical aspects, such as calibration of instruments or use of auxilliary theories in their interpretation. The semantic view (according to Suppe) allows to separate these aspects more neatly.

This criticism is summarised on p. 64:

when one reflects that the theory’s reliance on the results and procedures of related branches of science, the design of experiments, the interpretation of theories, calibration procedures, and so on, are all being lumped into the correspondence rules, there is reason to suspect that a number of epistemologically important and revealing aspects of scientific theorizing are being obscured.

An unfortunate consequence of this lumping is that we should assume that the theory changes everytime an experimental technique is perfected or revised, which seems obviously false. The semantic view does not have this implication.

Follows, in different chapters, various considerations against instrumentalism, defending what I interpret as epistemic and semantic externalism associated with notions of observation and measurement, discussions of explanations, laws, scientific taxonomies, induction, etc. However, these aspects are less relevant for my purpose.

There are many interesting aspects in this book, but also unsatisfactory ones.

As with the structuralist programme, and as discussed in my previous post, I found this idea of divorcing the content of the theory from its linguistic formulation overstated, particularly in the prologue of the book. Of course, various statements can describe the same structure, just as sentences in different language can express the same proposition, but that does not make of the content of the theory something non-propositional (a structure living in a platonic realm?). There is always, with this way of thinking, the problem of intended interpretations, which is largely left unanalysed in this book, only implicitly assumed (whereas at least the structuralists attempted some kind of analysis, albeit an implausible one). But we know all too well that intended interpretations are identified by means of language, and there are good reasons to think that it fully belongs to the theory (structuralists agree). It is true that changing measurement techniques do not change the theory, but changing its intended scope while only keeping the abstract structure does make for a new theory, for example, a theory of electricity instead of a mechanical theory, both having the structure of the harmonic oscillator.

Another (related) problem that the structuralists had not is in the understanding of models as either being represented by statements, or representing a wordly state of affairs, also already discussed in my previous post. This issue is not addressed in the book, but it seems quite clear to me that models are taken to “describe”, “characterise” or represent systems, and so, they are definitly representational entities, on a par with propositional entities, rather than mind-independent ones. This is good, in a sense (remember my criticism of the other view in the previous post), but then again, the difference between the semantic conception so conceived of and the syntactic view seems overstated. In the end, we could well identify a theory with a propositional content that is about a set of (idealised) physical systems. And if we do so, we will realise that most, if not all of the purported advantages of the semantic view are actually completely independent of the alledged focus on structure rather than language. They have more to do with some general points that could have been made using a syntactic view: that theoretical vocabulary should be interpreted realistically and not partially, that the theory only describes idealised systems, etc.

The main difference with the view of logical empiricists appear to be a kind of “two-stage picture”, with models or idealised systems playing a mediating role between the theory and the phenomena given in experience. But this is also misleading. There is no more stage than in the logical empiricists view if the relation between theory and idealised system is, according to Suppe, entirely computational, just a matter of applying the theory to particular theoretically described cirumstances. Then the two stages follow exactly the distinction from logical empiricism between theoretical postulates and correspondence rules. So, the point, it seems, is that theories should only be identified with the former and not include the latter (which, as just said, would be problematic if it means removing intended scope) ? Or that the latter are just a reconstruction of a messy practice guided by auxiliarry theories, but also by informal knowledge and abilities that have nothing to do with definitions and cannot be given a clear formulation? Perhaps these are legitimate criticisms (I completly agree with the second one at least), but they have nothing to do with focusing on abstract structures instead of statements. The approach brings unnecessary complication, such as the idea that the same statement can describe a model or a system or a phenomena. As for the other aspects of the book that I haven’t fully summarised, such as the account of explanations or the realist arguments, I found them also completely independent of the general gist of the semantic view, and easily transposed to a statement view.

Having said this, I must recognise that this idea of “model as mediator”, as is well known, as been exploited in practice-oriented philosophy of science, so it eventually proved fruitful. We can see that an important motivation for the semantic view, perhaps the main motivation, is to account for idealisations, and models are important in this respect. However, first, I would insist that theoretical models at least are still representations that can be associated with a propositional content of sort, and need not be divorced from linguistic analysis, and second, what is missing in the book, it seems to me, and what really made these further approaches interesting, is to give more autonomy to models. Models are not mere extensional structures satisfying theoretical laws. They have built-in idealisations and postulates, etc. Only then can we see that indeed, there are two stages that are not accounted for in the logical empiricist picture, because building a model is not just applying theoretical statements to a particular case, but it is also genuine theorising. We can also understand that modelling (which involves claims about a particular system or type of system) has a particular role to play in science that is not captured by purely universal theoretical statements (but then we can also recognise that these statements have an importance of their own beyond models!) But I admit that the ideas put forth by Suppe in this book set the stage for these further developmentsm, and so they have their importance.