The author argues for a purely naturalistic underpinning of the linguistic practice of reporting one's introspections. In doing so he avoids any commitments about the ontological status of entities referred to in introspective reports. He also presents evidence of the inadequacy of peripheralistic behaviorism as a naturalistic underpinning of introspective reports. The paper includes (a) a definition of ‘introspection’ and criticism of alternative definitions, (b) a classification scheme that sorts introspections into six different types, and (c) a presentation of evidence that the reporting of certain of these is based on such fundamental psychological phenomena as stimulus generalization, and possibly also conditioning to covert mediating responses.

This paper derives a mathematical expression giving the development of the probability of a scientific hypothesis with the number of confirming tests, as determined by Bayes's theorem, in a special case in which all the tests are “independent” of one another. The simple expression obtained shows clearly how the various factors influence the growth of the probability. The result is used to set a numerical lower bound on the probabilities representing the a priori beliefs of humans in generalizations that become accepted. By making a comparison with the predictions of a “logical atomic” model in the case of physical laws, it is argued that humans have significant a priori “knowledge” in a weak sense.

It is argued herein that there are two distinct ways in which all observation vocabularies are prejudiced with respect to theory. An argument based on the demands of adequate translation is invoked to show that even the simplest of our observation predicates must display the first and more obvious grade of bias—intensional bias. It is also argued that any observation vocabulary whose predicates are corrigibly applicable must manifest a second and equally serious grade of bias—extensional bias—independently of whatever intensional bias its predicates may or may not have.

Adolf Grünbaum, [1], and Philip Quinn, [7], have proposed two problems as sharpened versions of theses suggested by Pierre Duhem. (1) Can an hypothesis which in itself has no observational consequences ever be falsified by the evidence ? (2) When a theory has observational consequences only in conjunction with auxiliary hypotheses and some of these consequences fail, can the theory always be reasonably defended by constructing alternative auxiliary hypotheses ?