Saussure’s emphasis on language as a structure, rather than on its individual elements, led to the adoption of the term structuralism, the importance of which in modern linguistic thinking is difficult to overemphasize. Robins (1997: 225) has gone so far as to say that ‘the structural approach to language underlies virtually the whole of modern linguistics’. His concept of distinctive oppositions, and notably that of the phoneme as a distinctive speech unit, was later developed and refined in the 1920s and 1930s by the Prague School Linguists, including Nikolai Trubetzkoy and Roman Jakobson. They also explored the causes of sound change, on which the Neogrammarians had been largely silent, from a structural point of view: phonemic mergers might, for example, be seen to decrease complexity within the system and thereby reduce effort from the speaker’s point of view (though the question of why a change happens at a particular time would not be addressed until the advent of variationist sociolinguistics).

It was, however, in the United States, between the 1920s and 1950s, that linguistics became established as an autonomous academic discipline. The direction for the subject was set by a group which came to be known as the North American Descriptivists, whose major figures include Leonard Bloomfield, Martin Joos, Henry Gleason, Charles Hockett and Zellig Harris; two important contemporaries, Edward Sapir and Franz Boas, came to linguistics from an anthropological background. Bloomfield above all is credited with achieving respectability for linguistics as a science, a central Descriptivist concern. His seminal 1933 work Language, which remains a highly readable introduction to the subject even today, attempts to bring the scientific rigour of the natural sciences to linguistics through detailed description of methodology and discovery procedures, and reflections on corpora and sample sizes.

The need to justify linguistics as a science resulted in an emphasis on those aspects of language which could readily be described and presented in terms of rules (notably at the phonological and morphological levels), and a consequent downgrading of those which could not, notably semantics, in which Bloomfield saw no imminent prospect of scientific progress:

"The study of language can be conducted without special assumptions so long as we pay no attention to the meaning of what is spoken."

Where nineteenth-century linguists had taken their inspiration from advances in natural history, the Descriptivists looked to the logical rigour of mathematics in the description of rule-governed linguistic behavior:

But of all the sciences and near-sciences which deal with human behavior, linguistics is the only one which is in a fair way to becoming completely mathematical, and the other social scientists are already beginning to imitate the strict methods of the linguists.

For the Descriptivists, a scientific approach demanded reliance on publicly observable linguistic data, the goal at this stage being to describe rather than to explain.

We do not answer ‘why’ questions about the design of a language… we try to describe precisely; we do not try to explain. Anything in our description that sounds like explanation is simply loose talk… and is not part of current linguistic theory.

The emphasis on description reflected a desire to shed some decidedly unscientific prejudices from the past, notably the assumption that all languages were, in essence, structured along similar lines to the Classical languages (or, worse, that they ought to be). Such beliefs were flatly contradicted by the material with which the Descriptivists generally worked: an aspiring doctoral student in the interwar years would typically be required to provide a description of the grammar and phonology of an (often obsolescent) native American language, for which a Classical or European language model was of little help. The Descriptivists’ approach, which stems directly from Saussure’s conception of language as a system of relations, was to establish observable regularities of form within their data sets or corpora (singular: corpus), and describe the distribution of each element.