In this paper we will discuss several cases of the phonology of loan words. We will show that loan words are adapted phonologically to meet certain constraints on possible words or morphemes imposed on them by the borrowing language. Most significantly, such constraints are defined on the level of lexical representation. To our knowledge so-called “surface phonetic constraints” (henceforth SPC’s) play no role in the phonology of loan words, nor indeed in any other part of phonology. Our examples show that deep phonotactic constraints, constraints not necessarily reflected on the surface, are the drums to which loan words must march. This is not to say that all aspects of loan word phonological behaviour are now understood. Many mysterious facts still remain unexplained. We do claim that substantial aspects of this phenomenon may be understood by assuming that the systematic phonemic level of generative phonology is the level on which the constraints controlling loan word phonological behaviour are defined.

Dans cet article nous présentons une nouvelle théorie pour rendre compte du maintien du e [ə] final en ancien français dans l’évolution historique des mots suivants:

Nous allons présenter d’abord les descriptions traditionnelles de ces évolutions, en précisant leurs limites, puis la solution de Fouché (1966), et enfin dans une troisième partie, notre propre solution.

In Natural Generative Phonology (NGP), the only phonological rules are those which describe alternations that take place in environments that can be specified in purely phonetic terms. As indicated by Hooper, these “‘phonetic terms’ refer to phonological features (that have intrinsic phonetic content) and phonological boundaries (that have a necessary and consistent phonetic manifestation)” (1976:14). Any rule which changes phonological features in an environment described in morphosyntactic or lexical terms is not phonological but morphophonemic. Since by definition a phonological rule cannot contain non-phonetic information, this entails, among other things, the exclusion of any boundary that is not determined by phonetic means. In other words, the syllable ($) and the pause (| |) are the only boundaries which can appear in a phonological rule in NGP. The word boundary (# and ##) and the morpheme boundary (+), which are determined by syntactic and semantic means, can only appear in a morphophonemic rule.

Drawing on evidence found in the Survey of English Dialects (SED), in Joseph Wright’s English Dialect Grammar (1905) and elsewhere, J. Lawrence Mitchell (1979) suggests that the development of Middle English [x] into [θ], [f], [s] and ø in such terms as trough, plough and sigh is the result of a systematic phonological process which he terms ‘sliding.’ This sliding, he maintains, takes place along a consonantal strength scale which determines the order of change as [x] → [θ] → [f] → [s]. The purpose of the present paper is to bring forward some evidence for. [f] → [θ], which runs counter to his theory.

In the Merriam-Webster file of transcriptions, which now numbers over 272,000 slips, attention was first drawn to the substitution of [θ] for [f] in 1955 by a citation for aphis consistently pronounced [eθis] by a resident of western Massachusetts. We also have the testimony of a college-educated native English speaker from eastern New England (Waltham, Mass.) that he grew up saying Philadel[θ]ia and epita[θ]. He reports, in addition, that this pronunciation of the Pennsylvania city was used by a college friend from northeastern New England, a usage which was the butt of remarks by other classmates.

All treatments of so-called to-contraction in English (Lakoff (1970), Bresnan (1971), Lightfoot (1976)) attempt to deal with this phenomenon entirely in terms of contiguity. The purpose of this squib is to point out that the major constraint on this phenomenon is rather one of subject coreference, and furthermore that this coreference condition is identical to that for a universal rule which I have called ‘equi-subject clause union’ (Frantz 1976).

To illustrate, I will reconsider typical examples cited in discussions of this phenomenon, pointing out the relevance of coreference.

Berlin and Kay (1969) present strong evidence that “all languages share a universal system of basic colour categorization” and that “the basic color-term inventories of most languages expand through time by lexicalizing these categories in a highly constrained, universal order” (Kay and McDaniel 1978:610). The validity of these conclusions has also been demonstrated by Collier et al. (1976) and by Kay and McDaniel (1978). The evolutionary process identified by Berlin and Kay can be represented schematically as follows:

where the arrow may be read ‘is encoded before’ (Kay and McDaniel 1978: 615). In other words, languages proceed through a series of stages in the development of their basic colour-term lexicon, with the most simplistic system containing only terms for white (light) and black (dark). Any expansion of this fundamental system will first result in the addition of a term for red and then in the addition of a term for green or yellow. If further expansion takes place, “yellow or green, whichever did not emerge at the previous stage, now emerges” (Berlin and Kay 1969:18), followed by terms for blue and brown. When a system develops beyond Stage VI, “there is a rapid expansion to the full roster of eleven basic color categories,” although no particular order of appearance among terms for purple, pink, orange, or gray has been ascertained (Berlin and Kay 1969:21-22).

Králík (1977) examined four Czech texts of 7000 words each, in which he marked every occurrence of the word meaning and (and separately, three other words). The interval between two successive occurrences of the marked word is called a gap, and the number of words in that gap is the gap length. When the quantity of gaps of each length was recorded, it was found that there were many more short gaps than long ones. Králík proposed an exponential decay model of this distribution of gap lengths: For N gaps in a text of T words, the proportion of those gaps that are of length x should tend to be:

f(x) = a exp (— ax) where a = N/T

That is, as the gap length increases, the number of such gaps decreases in a smooth, downwardly convex, “swooping” curve.