Chomsky's "Galilean" Explanatory Style

Abstract

<p>Chomsky pursues a methodology in linguistics that abstracts from substantial amounts of data about actual language use in a way that has met considerable resistance from many other linguists. He thinks of this method as like that employed by Galileo and later physicists who proposed laws of motion in considerable abstraction from many of the motions we observe in daily life, focusing, for example, not on leaves in the wind, but on frictionless environments that virtually never occur on earth. Thus, Chomsky’s theoretical proposals are supported not by studies of the corpora of actual language use, but often by the intuitions of native speakers; and the relevant intuitions are not about what they think is <i>often</i> or is <i>likely</i> to be said, but rather about what “<i>can’t</i>” be said (so called “negative data”), and about what types of interpretation a sentence can or cannot have. But doesn't this fly in the face of good, commonsensical scientific methodology? Aren’t theories confirmed by greater data, and refuted by data that seem to conflict with them? With regard to this issue, Chomsky (1980) writes: <p>Substantial coverage of data is not a particularly significant result, it can be attained in many ways, and the result is not very informative as to the correctness of the principles employed. It will be more significant if we show that certain far-reaching principles interact to provide an explanation for crucial facts – the crucial nature of these facts deriving from their relation to proposed explanatory theories. (Chomsky 1980, 2) <p>We’ll argue below that Chomsky’s observation here in fact accords with good explanatory practice elsewhere in science, but it does conflict with a traditional methodology in linguistics. In the spirit of the positivism/empiricism of the 1930s, the ‘structuralist’ linguist Leonard Bloomfield (1933, 20) insisted that “the only useful generalizations about language are inductive generalizations”, and linguists for the next several decades tried to specify ‘discovery procedures’, or rules for using a collection of phonetically characterized utterances to induce phonemic, morphemic and – it was hoped – finally syntactic analyses of the target language (see Sampson, 1980, 76ff). Such discovery procedures have fallen by the wayside, but many contemporary linguists would still agree with Bloomfield that linguistics seeks generalizations that both emerge from, and provide good coverage of, the data of language use.