For some thirty years, now, discussions of explanation in science have begun with the theory most clearly articulated by Hempel and Oppenheim (1948). It has been modified, extended, and defended by Hempel with perserverance and patience since then (1965). As a result, it has served as the standard against which competitors strived, and the clarity with which it has been set forth and defended has played a large role in the fecundity of the debate it has aroused.

It is not my intention here to attack or defend the theory in any of its versions, but to examine some aspects of the debate as they bear on the development of our concepts of what science is about. For the theory has been, taken by its friends and its enemies to play a dual role: it is, on the one hand, intended to be descriptive of what science does; while, at the same time, it is to serve as a model or ideal of what science ought to strive to be.

‘Bell's Theorem’ is the collective name for a family of arguments, which more accurately should be referred to as ‘Theorems of Bell's Type’, all being variants of the remarkable argument given by J. S. Bell (1964). All of these arguments have the format E & H → I, where E is a description of a type of experimental setup involving pairs of particles emitted from a common source, H is a physical hypothesis which typically expresses some version of “realism” and some version of “locality”, and I is an inequality concerning correlations between particles of a pair. There are differences among theorems of Bell's type regarding the details of the inequality I , which are important in actually carrying out experimental tests. There are differences regarding the arrow—i.e., the procedure whereby I is inferred from E & H.

Scriven, with his well-known bridge example, showed that explanations need not carry with them accurate or useful predictions. A bridge has collapsed, but there is no history of overload or natural shock. Metal samples taken near the break show severely reduced elastic capabilities of a sort usually associated with natural aging. Therefore, given that the bridge did fall, it fell because of this sort of metal fatigue. However, on the basis of current theoretical and measurement capabilities, it could not have been predicted that it would fall. (Scriven 1962, pp. 181-185).

The existence of explanations which lack true or even usefully approximate predictions is not restricted to complex situations in engineering. Nor are such explanations the exclusive property of the non-physical sciences. The existence of such explanations, I want to claim, is common in the physical sciences as well.

How do we collect data to test theories about syntax and semantics? For most of us, the process is very simple: We have a hypothesis, we construct sample sentences to test it, introspect about their grammaticality or meaning, and move on to modify our hypothesis as necessary. It would be fair to say that the standard method of data collection in syntax and semantics is personal introspection. This personal introspection is not totally uncontrolled, of course: Doubtful cases are checked informally with friends or colleagues; and there is an implicit, check when the referees accept a paper or an audience listens without objecting. What we have, in effect, is the introspection of the professionals, the people who know what the issues are.

This standard method has been enormously successful: Linguists have produced precise and interesting theories about syntax and semantics; and the bulk of the data on which these theories depend is undisputed and, we believe, undisputable.

It has become customary for linguists (especially generative linguists) to draw a distinction (initially rather unnatural for philosophers of science) between “internal” and “external” evidence. Usually classified as internal are data on the cooccurrence and alternation of linguistic elements in some language, as well as such systemic considerations as formal simplicity, economy, and the like. External evidence is everything else: the use of phonemes in rhyme schemes, patterns of acquisition, comparison to other languages, speech errors, dialect differences, historical change, and so on. The distinction is usually made invidiously--only internal evidence is probative--or defensively--external evidence, or at least some types of external evidence, are relevant and useful.

The distinction arises at two quite different points in the investigation of language. It arises first in the division of labor between linguistics and other fields.

Although philosophers of science have always been interested in the actual work of scientists, there has been a strong turn in the last generation away from prescribing how science ought ideally to proceed and toward studying more carefully how science has proceeded. In part this turn has been a reaction to previous work in philosophy of science, which to many seemed misguided and largely irrelevant to the sciences. In part this change reflects a general scepticism about the possibility of doing traditional foundationalist epistemology. Such scepticism is itself a reaction to the failure of the foundationalist program of the logical empiricists. The contemporary turn toward careful empirical study of the sciences constitutes a new program for the philosophy of science, which I shall call ‘empirical philosophy of science’ or ‘the empirical approach to the philosophy of science’.

Among the numerous criticisms of sociobiology is the criticism that it is not genuine science. The attempts to support this charge are based on the beliefs that sociobiology is in some sense untestable (Allen et al 1976 and 1977; Sahlins 1977; Lewontin 1977) and untestable hypotheses and theories are pseudoscientific. There are, however, a number of difficulties with both of these beliefs. First, the testability criterion of genuine science is far from clear or well accepted. It is now obvious to most philosophers of science that hypotheses and theories do not occur in isolation. Numerous assumptions and hypotheses together form part of a general framework within which any hypothesis is formed, applied and tested (see, Quine 1951; Quine and Ullian 1978, pp. 96-107; and Kuhn 1970). The refusal to give up a hypothesis in the face of a negative test result is not only not grounds for considering it pseudoscientific but is also a common occurrence in scientific research.

Questions about the nature of the appropriate data for linguistics may seem a mundane matter, but they reflect deep issues about'the nature of language and of the science of linguistics, if science it be. Consequently the methodological modifications suggested by Carden and Dieterich (1981) and Zwicky (1981) could have ramifications far beyond a slight broadening of the range of data relevant to syntactic theory. In section 1, I review the arguments for the currently standard theory against the suggestion to broaden the range of data, and in section 2. I try to give a brief historical account of the origins of the presuppositions of the orthodox theory.

The current official conception of the goal of syntactic theory is that it is to distinguish syntactically well-formed strings of a given language from those that are not syntactically well-formed, and to provide structural descriptions of the sentences.

By “scientific realism” philosophers ordinarily mean the doctrine that non-observational terms in scientific theories should typically be interpreted as putative referring expressions, and that when the semantics of theories is understood that way (“realistically“), scientific theories embody the sorts of propositions whose (approximate) truth can be confirmed by the ordinary experimental methods which scientists employ. There are as many possible versions of scientific realism as there are possible accounts of how “theoretical terms” refer and of how the actual methods of science function to produce knowledge.

What I will do in this paper is to explore the consequences of one such version of scientific realism, a version which embodies the implicatures as well as the implications of the realist slogan that reality is prior to thought.

The empiricism/realism debates have been going on now for so long, in so many philosophical arenas, that any simple statement of the issue is bound to be wrong or biased. Yet we need a simple formulation to begin, if only to know who is meant to pick up which gauntlet. I shall begin with a short sketch of how I see the enterprise of philosophy of science, and then I shall describe what I take to be good empiricist views on two subjects: theory construction and experimentation. At the end I shall sum up with a simple recasting of the main points at issue.

Within philosophy, the philosophy of science is comparable to philosophy of law, philosophy of art, of religion, of mathematics. These disciplines are comparable in that each has as subject matter a large-scale cultural phenomenon, with its associated activities, institutions, products, language, and internal standards of evaluation.