Many contemporary linguists describe their data in terms of a system of rules which state that one string of symbols is to be rewritten as another string of symbols, perhaps with the condition that this only occur when the symbols occur in a particular context. When the linguist wants to use the grammar to produce a particular utterance, he starts with an initial symbol and, one at a time in linear sequence, applies or attempts to apply the rules to the string of symbols he is building up. After applying all of the rules, or perhaps a subset of them, he may have to reapply the rules of a subset of them again and again until no further changes can be made to the string. If at the end of this process the string consists solely of terminal symbols, he has produced a grammatical utterance. Basically such systems consist of symbols and a few operations on these symbols, including match, copy, concatenate, and replace. This system grew out of a union of the item-and-process approach to linguistics with automata theory, an area of mathematics related to the understanding of computers.

It has been traditionally recognized by linguists that the rules of linking in French are syntactically constrained and that their application depends on the grammatical category of the constituents involved. The author of this paper agrees with the basic thesis that linking is not solely phonological. However, the traditional way of analyzing linking misses an important generalization. The purpose of this paper is thus to show that there exists a deeper explanation for the phenomenon. It will be suggested that linking is not constrained by particular grammatical categories but by the presence of a boundary. The grafting principle will be offered as an explanation for the presence of this boundary. It will also be shown that the present analysis accounts for the phenomenon of pause.

The Notion of persistent rule in historical phonology has been discussed by Chafe in his important asticle on the order of phonological rules (Chafe 1968). Two examples are the wellknown devoicing of final obstruants in German - a rule added to the grammar of German at a point in time in its history, and still operative so that words introduced after the adding of the rule are subject to it; and the loss of final /n/ in Livonian, so that a word like /nainen/ first becomes /naine/, then is subject to a rule that eliminates final vowels, thus becoming /nain/, which, since the n-dropping rule is still in force, becomes /nai/. Rules are said to persist when they apply either to forms resulting from rules added later than they to the grammar of the language, or to forms introduced into the language from a source other than a later rule, i.e. through borrowing.

It is, perhaps, not unjust to divide linguists between those who accept the basic Saussurean dichotomy between langue and parole, and those who reject it. Saussure advanced this theory at a time when linguistics was still essentially diachronic, and the link between a language and its user was seen in literary rather than in social terms. But though his book was a much needed corrective to the mechanistic cast of contemporary Darwinian linguistics, developments in language analysis outgrew his statement of the theory as it appears in the Cours de linguistique générale.