In a much cited and little appreciated paper, Y. R. Chao more than thirty years ago anticipated the logic-inspired position of Noam Chomsky that mechanical discovery procedures are not a realistic goal for linguistic analysis. Here I look briefly at a familiar problem in the analysis of English in an attempt to underline some relations of unique and nonunique analyses.

The problem is that of the relation of traditional long and short vowels in modern American English, and for paradigm I shall consider only the vocalic nucleus of beat vs. that of bit (with a sidelong glance at the word dear). But first it is necessary to dispose of a straw man, without which the matter cannot even be discussed as a problem. In currently fashionable terms, this straw man might be named the ‘strong non-uniqueness position’, and taken to hold that no criteria can POSSIBLY be adduced to prefer one solution over another: it is all a matter of taste.

The Algonkian family of languages, with its great geographical spread from the Atlantic seaboard to the Rocky Mountains and from Hudson's Bay to the Carolinas, together with the inevitable proliferation of dialects (daughter languages) which accompanies such a spread, provides a goldmine of still largely un-exploited source material to exemplify a wide variety of principles in comparative linguistics. Unfortunately, only a part of the family has been intensively studied from the comparative point of view; as a consequence, many points of interest are likely to be represented in a fragmentary or indecisive fashion in the literature. Bloomfield's Algonkian studies were limited to four of the Central languages—Fox, Cree, Menomini, and Ojibwa; and Hockett in his follow-up work stayed within the same boundary, except for the addition of Potawatomi. Excellent results have been achieved by this limitation, but proper caution requires that the limitation be kept fully in mind whenever there is a temptation to generalize, or to compare developments in this family with those in other families.

In 1919 Dixon and Kroeber proposed that five language groups in California (Wintun, Maiduan, Yokutsan, Miwokan, and Costanoan) were a genetic unit, which they called Penutian. A study of the available data over the past several years has verified this postulation. Penutian, so defined, is a single family of languages.

The Hare Indians, according to Jenness, lived in northwestern Canada, ‘west and northwest of Great Bear lake, extending in the east to a little beyond the Anderson river, and in the west to the first line of mountains west of the Mackenzie river’. Their language, spoken by about four or five hundred people, belongs to the Athapaskan stock.

In spite of the influential Saussurean dichotomy of synchronic and dia-chronic studies, it has in fact not been possible to keep these two main aspects of linguistics in hermetically sealed compartments. The aim of this paper is to explore the theoretical relations between phenomena in the two fields in the context of the current interest in universals. The examples cited are not intended to be conclusive and are meant only as illustrations.

Over the past thirty years or so a number of scholars, mostly but not entirely American, have sought to elucidate problems of genetic linguistics by employing, explicitly or not, some form of probability mathematics. As I reread these efforts, my own included, I wonder that no one has ever seriously questioned the pertinence or, indeed, the legitimacy of such employment in elucidating historical problems. Mathematics takes vaguely defined terms of ordinary speech like ‘probability’ and ‘chance’ and gives them precise meanings, but when we use such terms in linguistics, together with their associated mathematical procedures, do we always keep the precise meaning? Or do we fall back unwittingly into the popular use and thus vitiate our results? And even if we are careful to keep the precise meanings, are we justified in supposing that mathematical probability can furnish a theory to account for the phenomena of history in languages, or, if not a theory, then a model? Such questions as these deserve answers, and by now enough experience has accumulated to warrant an attempt to answer them. The present paper is offered as such an attempt with the hope that, if it does no more, it will at least ask the right questions. But first a word must be said about theories and models.