"Science and Metaphysics" by Wolfrid Stalker Sellars,1965

Sellars' Philosophy of Science and Epistemology
Sellars' interpretation of the epistemology of natural science departed decisively from the received view according to which explanation was identified with derivation — singular matters of empirical fact being explained by deriving descriptions of them from (“inductively” established) empirical generalizations (along with appropriate statements of initial conditions), and these “empirical laws” in turn being explained by deriving them from theoretical postulates and correspondence rules. On this received Positivist view, theories (e.g., microtheories) explain empirical matters of fact only indirectly, by implying generalizations framed in an observation-language that explain them directly. In consequence, as Hempel pointed out in “The Theoretician's Dilemma”, such theories, although perhaps convenient aids to calculation and compact representation, are in principle utterly dispensable.

Sellars regarded this “layer-cake model” or “levels picture” of theories as fundamentally misguided. He argued that there is no autonomous stratum of empirical counterparts to theoretical laws. The empirical generalizations corresponding to theoretical laws become salient only from the theoretical perspective. Generalizations arrived at autonomously at the observational level, however reliable, are not laws of nature, and theories consequently cannot be in the business of explaining such lower-level generalizations by entailing them. Rather, “theories explain laws by explaining why the objects of the domain in question obey the laws that they do to the extent that they do” (LT, 123).

[That is,] they explain why individual objects of various kinds and in various circumstances in the observation framework behave in those ways in which it has been inductively established that they do behave. Roughly, it is because a gas is ... a cloud of molecules which are behaving in certain theoretically defined ways, that it obeys the empirical Boyle-Charles Law. (LT, 121)
On Sellars' view stories that postulate “theoretical entities” are not merely manageable second-class surrogates for more complicated and unwieldy stories about entities that we have good, i.e., observational, reasons to believe actually exist. Theoretical entities, rather, are those entities we warrantedly believe to exist for good and sufficient theoretical reasons. On this understanding, scientific theories explanatorily “save the appearances” precisely by characterizing the reality of which the appearances are appearances.

Like Quine, Sellars was deeply influenced by the work of Rudolf Carnap. Sellars' sophisticated account of the nature and import of theoretical reasoning in natural science, however, enabled him to develop a systematic naturalistic alternative to Quine's influential critique of Carnapian logical empiricism. In particular, the epistemological contrast between two sorts of empirical generalizations — those adopted on narrowly inductive grounds and those expressing constitutive principles of postulational theories adopted on broadly empirical, i.e., explanatory grounds — enabled Sellars to distinguish among three different grades of “observational involvement”: observations and general claims individually validated “inductively” by way of direct appeals to observational backing, the constitutive posits of postulational theories holistically validated by way of indirect, explanatory appeals to observational backing, and purely formal claims expressing necessary conditions for the formulation of scientific hypotheses in general. Consequently, where Quine rejected the classical Kantian analytic-synthetic dichotomy out of hand, Sellars argued that there were two quite different distinctions tangled up in the single dichotomy that Carnap had inherited from the Kantian tradition: the distinction between logical and empirical (matter-of-factual) claims (analytic2-synthetic2), and the distinction between claims whose revision requires abandonment or modification of the system of (theoretical) concepts in terms of which they are framed and claims revisable on the basis of observations formulated in terms of a system of (theoretical) concepts which remained fixed throughout (analytic1-synthetic1). Like Quine, then, Sellars moved decisively away from classical Kantian rationalism, but in the direction of a Kantian empiricism which preserved logical space for a theory of semantic meaning and the correlative distinctions between individual matter-of-factual truths and truths which, although belonging to theoretical systems themselves adopted on broadly empirical (synthetic2) grounds, were, relative to such a system, true ex vi terminorum (analytic1):

Kant's Rationalism

Grounded in experience
(“a posteriori”, simple induction) Not so grounded
(“a priori”)
Synthetic Analytic
Empirical Laws
(regularities) Arithmetic, Geometry, Mechanics
(“synthetic a priori”) Logic
“Our conceptual framework” (innate principles)

Kantian Empiricism

Grounded in experience (Empirical) Not so grounded
Synthetic2 Analytic2 (L-true)
Synthetic1 Analytic1
Observation, Simple Induction
(Operational geometry, mechanics) Postulation
(Physical geometry, idealizing scientific theories, mechanics, micro-physics) Logic, arithmetic, mathematical analysis
(Pure geometry qua calculus)
“Our conceptual framework”:
Material (empirical) categories Formal (ontological) categories