The Semantic Conception of Theories

Philosophy of Science

Giere's text gives a fairly straightforward account of what is often called "the semantic conception of theories."

1. The "received view," "classical view," or "H-D view" of theories. (Giere uses the second term, Lloyd uses the third, both of them mention that the view has often been called the first)

context: relatively early days of symbolic logic, when various theories were "axiomatized." Great successes included arithmetic (Peano axioms -- but note that these aren't complete, nor is there a complete axiomatization, as Gödel showed!) and geometry (Euclid, formalized by Hilbert. Of course, these are both mathematical rather than scientific theories, though. Lloyd discusses an attempt to do something similar for evolutionary theory.

In such an axiomatization, some of the axioms will be thought of as definitions and others as laws. (One question about this approach is whether there is a principled way to distinguish which are which!)

Lloyd associates the classical view with the H-D model of confirmation. This terminology is due to Hempel, and is closely linked with his D-N model of explanation. The idea is that you confirm a theory by using it to make predictions about observable phenomena, and then checking to see whether the predictions are correct. For Hempel, predictions, as we remember, are structurally identical with explanations: you need premises that include at least one law and some initial conditions; you deduce a statement about something observable.

Roughly speaking, the classical view thinks of theories as collections of sentences; the semantic view thinks of theories as models or sets of models.

2. The "semantic conception" of theories.

On the semantic view, a theory is a model or family of models (perhaps together with theoretical hypotheses linking the model with some real-world phenomenon). What's a model? There are scale models, maps and diagrams, and analogies. More commonly in the sciences, we have mathematical models: sets of equations that are thought to capture real-world phenomena in an idealized form. Increasingly in psychology and the neurosciences, we are also getting computational models: computational architectures or programs that are meant to capture some aspect of human cognition.

Relatedly, we can think of a theoretical model as a kind of definition. In particular, a theoretical model defines a certain kind of system. (Examples: a classical particle system or a Mendelian system.)

That means that in a sense a model can't be false. It's just a big definition of a kind of system. (So we don't have to figure out which parts of the theory are definitional and which are empirical. This may be a good thing. Consider Newton's second law, F = ma. Is this an empirical claim, or a definition of force? (Or perhaps a partial definition of mass or acceleration?) It's not entirely clear.

However, to view a theoretical model as a definition is not to say that science has no empirical content! According to Giere, the content enters by way of theoretical hypotheses. A theoretical hypothesis is the hypothesis that a certain real physical system is a system of the type defined by a certain model. We might put this by saying that a theoretical hypothesis says that some part of the real world is similar to the model.

Another potential advantage of this conception of theories is that a theory need not be axiomatized, or possibly even axiomatizable. One way to characterize a set of models is by way of axioms: this is how Giere presents Newtonian physics and Mendelian genetics. The models are any systems for which the axioms are true. However, other sorts of model might not lend themselves to this sort of axiomatization. Models of chemical structure, for instance, don't seem easily characterizable in terms of axioms, and yet can be clearly presented by way of literal scale models.

(You can see how this connects up to issues about realism. van Fraassen says that realism regards theories as literally true. Giere considers whether this means exactly or completely true, and suggests that a model is never either exactly or completely true: the resemblance between model and reality is never perfect. But Giere also thinks that we shouldn't interpret realism in such a way as to require anything this strong.)



Last update: February 7, 2011
Curtis Brown | Philosophy of Science | Philosophy Department | Trinity University
cbrown@trinity.edu