Abstract

A classical problem in the philosophy of science is the characterization of the purposes served by theoretical terms in scientific theories. Closely associated with this is the question whether theoretical terms can always be eliminated from scientific theories without loss of the essential purposes served by these terms. If they can, then the empiricist has gained a potential argument against scientific realism. I consider the success of elimination strategies with respect to three minimal constraints: the replacement theory must be axiomatizable , the empirical content of the replacement must coincide with the empirical content of the original, and the replacement must be at least as well tested as the original theory. If elimination strategies prove unsuccessful with respect to some constraint, then not only is the potential anti-realist argument blocked, but a positive function of theoretical terms is thereby identified. ;The dialectic of the dissertation proceeds by taking successively more plausible representations of empirical content. If the empirical content of a theory is taken to be the set of its consequences in the observation language, then Craig's reaxiomatization procedure generates an elimination strategy that trivially satisfies and . I argue, despite counter claims, that it satisfies as well. ;From a semantic viewpoint, however, the empirical content of a theory should coincide with the class of observable configurations permitted by the theory. A first approximation to this idea is obtained by taking the class of reducts with respect to the observation language of the models of the theory. This class need not be elementary in the wider sense, and hence cannot be defined in terms of the observation vocabulary. This possibility raises two questions. Are any known scientific theories of this sort? In this vein I defeat Sneed's conjecture that "mass" is not semantically eliminable from classical mechanics. Can methodological constraints that ensure semantic eliminability be invoked? Simon and Groen have proposed such a constraint. I argue, however, that it is an unreasonable one. ;Regardless, this representation of empirical content is inadequate since a given reduct of a model of a theory may still contain unobservable individuals and states of affairs. This problem is rectified by identifying empirical content with the class obtained by taking as elements the set of observable facts of each model of the theory. These sets may or may not yield structures for the observation language. Even if they do, then the obtained class need not be elementary in the wider sense in the observation language, the set of sentences true in it may be distinct from the set of the theory's consequences in the observation language, and the set of sentences true in it cannot be shown to be axiomatizable. In any event, it is clear that there can be no general elimination strategy that produces a replacement theory in the observation languages and yields only the class of sets of observable facts permitted by the theory