Abstract

The success of a scientific theory T is not an all-or-nothing matter; nor is a theory something one can usually accept or reject in toto (i.e. one may take T as being "approximately true", or take as true just certain "parts" of it, without necessarily affirming every posit and claim specific to T as being either completely right or completely wrong). This, however, raises questions about precisely which parts of T deserve to be taken as approximately true. on the basis of its success. A line of thinkers, particularly Kitcher, Leplin and Psillos, variously look for parts of a theory they can claim to have been "essentially" implicated in its distinctivesuccess, which they regard as primary candidates for realist truth ascription. But, how is one to determine which parts of any theory are "central" or "peripheral", "essential" or "idle" in the required sense? Attempts at spelling out relevant synchronic links between successful predictions and correct partial theorizing increasingly look like a misguided effort. As an alternative, this paper proposes a weaker, but arguably powerful enough, version of the realist relation between success and truth. Focusing on a pivotal case study in recent debates between realists and anti-realists (the conceptual changes undergone by theories oflight in the 19th century), a promising link between success and partial theoretical representation is located in the expansion and stabilization of approximately correct partial theoretical models of the theory's intended domain. The realist link is then formulated accordingly. In the resulting approach (a) predictive success is preserved as a marker of cumulative theoretical gain, but (b) specification of the latter is a diachronic rather than synchronic matter (i.e. gains become clear only after generations of theory change; specification of the particular loci of theoretical gain in connection with a given line of predictive success is not assumed to be generally possible at the time of the success in question). The truth ascriptions that get licensed are partial-of a piece-meal and retrospective sort, focused on methodologically specifiable theoretical subplots from past science.