The emphasis on syntagmatic criteria in analysis, especially as interpreted by J. R. Firth (from his 1948 paper onwards) in the phonological aspect of language, has led many scholars, in Great Britain at least, to a disbelief in the adequacy of phonemes alone – or of phonemes aided by such suprasegmental features as intensity, duration and tone – for the tasks of accurate description and powerful theoretical explanation of the operation of a spoken language. Among the fruits of the new emphasis, the studies which are characterized as ‘prosodic’ and ‘polysystemic’, is that treatment of Sanskrit discontinuous retroflexion within words which handles this feature as a prosody operating over a considerable stretch (chain of phonematic units) and being phonetically realized wherever possible, which may be more often than is graphically hinted (Allen, 1951). In this view, apparent historical ‘assimilation’ is trenchantly dismissed: we are not to imagine a shift being induced in one phoneme by another which stands at a distance from it, nor is the first to be spoken of as changing from ‘something it never was’. This, of course, begs the whole historical question; and even if the nature of Indo-European reconstructions is so skeletophonemic that we cannot say with certainty that Skt. ṇ is not the unshifted allophonic reflex of IE /n/ in the relevant positions in the relevant words, nevertheless a phonetic continuation is scarcely likely (e.g. in *plH-nó-). In many cases a different earlier form can be pointed to, attested and unmistakable (e.g. Latin octo before Ital. otto). One cannot lightly oppose a prima facie possibility that ṇ was first one thing and then became another.

In the first issue of this journal, Fred W. Householder discussed two papers of ours which he found defective in various respects. We feel that the issues involved are important and deserve the fullest clarification. We will therefore discuss Householder's objections and the underlying issues in some detail, reiterating points that have been made in the aforementioned papers and elsewhere and making no attempt to avoid redundancy if this can contribute to clarity.

In 1962, on the occasion of the Ninth International Congress of Linguists, an article by Morris Halle (1962) appeared in the special issue of Word, and Noam Chomsky (1964) appeared before a plenary session of the Congress to defend some remarks which had previously been distributed in the Preprints.1 These two papers were perhaps intended as a sort of one-two punch to knock old-fashioned linguistics out of the ring; at any rate, on matters of phonology, their claims and assertions, if all wholly true, would tend to make all phonological work impossible on any known lines (curiously, they fail to reveal their own methods).

The term word-and-paradigm grammar (Hockett, 1954; Robins, 1959) may be used of any grammar which preserves the traditional distinction between morphology and syntax. This distinction is usually drawn as follows. Following the classical formulation (Dionysius Thrax in Uhlig, 1883:22, line 4; Priscian in Keil, 1855: 53, line 8), any sentence is said to have as its minimal elements a set of [grammatical] words. Such words are of two main types. Some are invariable: examples are the Latin conjunctions et and atque or the prepositions per and in. Others, for example the verb amo or the noun dominus, are variable: they appear in a variety of forms (amo, amabas, amatus, ….; dominus, domino, dominorum, ….) with different syntactic properties. The ‘function’ and distribution of these variants, in relation to each other and to the invariables, is handled by the syntactic sections of the grammar; their phonological or graphological ‘shape’, on the other hand, is stated by the morphology. It is, for instance, a syntactic rule which states that the preposition per must govern a noun in the ACC [usative]2 case, but it is a subject of morphology that dominum, terram, etc. are ACC forms of their respective nouns. The field of grammar, as opposed to phonology, semantics, etc., is thus divided into two separate but interconnected sub-fields: one dealing with what may be called the ‘external’, the other with the ‘internal’ characteristics of grammatical words.

This paper is concerned with two related problems. Briefly: how can we best handle, in a transformational grammar

(i) Restrictions on the collocability (Simon, 1953; Firth, 1957; Mitchell, 1958) of one word with another.

The proposal that a grammar should be considered as a device which generates all and only the well-formed sentences of a language, which has already had a profound effect upon linguistics in general, has begun to exert some influence on the more particular (if less well defined) subject of stylistics. To understand this, it is necessary to consider the way in which one tests such a grammar. This includes checking that for any grammatical sentence we observe there exists in the grammar a partially ordered sequence of formulae which would generale it. It is an inevitable outcome of any prolonged examination of a theory of this kind that there will arise cases which can be generated but with regard to which a decision as to whether or not they constitute a genuine example of the phenomena the theory is supposed to cover cannot be arrived at merely by reference to existing data: cf. Weyl (1949: 150). In linguistics this is called the problem of GRAMMATICALNESS. (Chomsky, 1957: 15)