In recent years linguists have looked to linguistic change for empirical support for various proposed constraints on the form of grammars. This orientation of diachronic studies is reflected, for example, in Kiparsky (1968a) where linguistic change is characterized as “a window on the form of linguistic competence.” Arguments based partly on diachronic evidence are advanced by Kiparsky (1968b) in support of his contention that the standard theory of generative phonology as outlined in Chomsky and Halle (1968) should be modified to include a condition which, in its strong form, would not permit rules of absolute neutralization in synchronic phonological descriptions. More recently, Hooper (1976) also appeals to evidence from linguistic change in support of a theory of phonology which differs significantly from the standard theory referred to as transformational generative phonology (TGP). Hooper’s theory of natural generative phonology (NGP) is supposed to be characterized by greater concreteness of descriptions than is achieved in the framework of TGP. This greater concreteness of descriptions is supposed to be guaranteed in Hooper’s theory by the inclusion of two constraints on rules. One constraint, the True Generalization Condition, requires that “all rules express transparent surface generalizations, generalizations that are true for all surface forms.” The second constraint, the No-Ordering Condition, does not permit a grammar to contain rules which must be extrinsically ordered. In addition to these constraints, Hooper adopts a model of lexical representation according to which alternants of morphemes must be directly represented in the lexicon in certain cases. In such cases, the lexically specified allomorphs are correctly distributed by morphophonemic (MP) rules. Hooper (1976:23-41) claims that evidence from some phonological developments in certain dialects of Spanish points to the correctness of NGP and, ipso facto, shows the inadequacy of a theory such as TGP which permits the violation of the two conditions mentioned above.

G. L. Piggott’s paper “Implications of linguistic change for concrete phonology” does not present diachronic evidence in favor of a theory of phonology with abstract representations and derivations, as it claims, but rather reconstructs a hypothetical sequence of events. This reconstruction is based on a theory of phonology; that is, Piggott reconstructs a sequence of events that is consistent with the theory that he holds. It is not legitimate, therefore, to turn around and use this reconstructed situation as an argument for his theory. We return to this point below.

A somewhat weaker point is in principle possible, and Piggott tries to make it also. This would be to show that some theory cannot provide an historical account of how a certain situation could have arisen. Note that all that is required here is that the theory provide sufficient machinery to produce a plausible reconstruction of a state of affairs. Piggott is quite correct in maintaining that this is a legitimate test of a theory. He argues that the theory he defends provides such an account of the examples he discusses, but that the theory of Hooper (1976), natural generative phonology, cannot provide such an account.

Recent changes in Hare, an Athapaskan language of the lower Mackenzie River Valley, require that a rule of epenthesis be ordered in two places in the grammar. The original rule is ordered before a rule of vowel raising. In the innovative dialect of Hare, part of the environment for this epenthesis rule is revised and it must be ordered after the raising rule.

Il est plus éeconomique et naturel de rendre compte de certaines régles phonologiques lorsque nous analysons les contextes dans lesquels elles s’appliquent en tenant compte de la structure syllabique. La première théorie qui a été élaborée sur cette base, dans le cadre génératif, est celle de Kahn (1976). Par la suite, d’autres ont suivi. Celle que je retiendrai pour nos fins est la théorie de Liberman et Prince (1977). C’est en me servant de leurs théories que je vais démontrer, dans la deuxième partie de cet article, certains avantages que comporte une telle analyse, L’analyse en structure syllabique fait très bien ressortir le rôle de l’accent dans la formation des syllabes. Je crois que la structure de ces dernières doit être prise en considération pour rendre compte d’une règle comme celle que l’on a baptisé “Canadian Raising.” Les principales données qui furent utilisées pour la formulation de la règle de Canadian Raising ont été rapportées, il y a plus de 35 ans, par Joos (1942). Chambers (1973) a repris la question et nous offre une réanalyse dans le cadre de la théorie de Chomsky et Halle (1968; voir aussi King 1972, Picard 1977). C’est la formulation de la règle de Canadian Raising que Chambers a présentée que je vais réviser dans la deuxième partie de l’article. La première partie servira à la présentation de cette règle.

On a toujours considéré, semble-t-il, que la nasalisation vocalique régressive procède habituellement par étapes, c’est-à-dire que certaines voyelles se font nasaliser avant d’autres, soit à cause de leur nature même ou à cause du contexte plus ou moins propice dans lequel elles se trouvent. Or, voici qu’après avoir effectué une étude exhaustive sur le développement des voyelles nasales, George Entenman en est arrivé récemment à la conclusion tout à fait opposée. Son hypothèse peut se résumer ainsi: “A language cannot nasalize some vowels before nasal consonants without nasalizing all vowels in this environment” (1977:1). C’est ce qu’il a baptisé le principe “tout ou rien” (the “all-or-nothing” principle).

Bien que les arguments qu’il présente en faveur de ce principe semblent en général solides et bien motivés, je vais tenter de démontrer qu’il y en a un d’entre eux qui s’effondre complètement. L’argument en question découle d’une tentative de rendre compte d’un phénomène de toute évidence phonologique, à savoir la dénasalisation vocalique en français, en des termes morphologiques. Avant d’aborder ce problème particulier, cependant, il est essentiel de le situer dans le contexte de sa théorie.

In alternative questions, the second clause is frequently incomplete. For example, sentence (2) is understood as synonymous with sentence (1):

1. (1) Did John read the book or did Mary read the book?

2. (2) Did John read the book or did Mary?

This paper addresses the problem of how the incomplete second conjuncts are to be assigned full semantic representations, and thus attempts to account for the synonymy of (1) and (2).