¹ A discussion of these and related points of view is found in Uriel Weinreich, ‘Is a structural dialectology possible?‘, Word 10.388–400 (1954). For an application of the notions presented here, see Julia Sableski, A generative phonology of a Spanish dialect (unpublished M.A. thesis, University of Washington).

An earlier version of this paper was delivered at the meeting of the Linguistic Society of America, summer 1964, in Bloomington, Indiana. The author's research is supported by the National Science Foundation, the National Institute of Health, and the Office of Education.

² For example, William G. Moulton, ‘The short vowel systems of Northern Switzerland’, Word 16.155–82 (1960), and Edward Stankiewicz, ‘On discreteness and continuity in structural dialectology’, Word 13.44–59 (1957).

³ ‘Phonology in generative grammar’, Word 18.54–72 (1962).

⁴ I have retranscribed Halle's examples in a more familiar notation. The use of slant lines and square brackets is not intended to correspond to the now traditional distinction between phonemic and phonetic transcriptions. I use brackets only to make explicit when necessary the fact that a phonetic rule has been applied to yield the form in question. In a more precise presentation, slants might be used only to indicate the underlying phonological representation which serves as the input to the phonological component of the grammar. Note, however, the arguments proposed by Halle to support the view that segments have no systematic import and ‘are to be understood as circumlocutions introduced only to facilitate the exposition’ (56, fn. 2).

⁵ Review of Hans Kurath and Raven I. McDavid, The pronunciation of English in the Atlantic states, in Lg. 39.303–16 (1963). The relevant data are the pronunciations of five, twice, down, out, for which the underlying forms include diphthongs /ai/ and /au/. Rule 1 raises /a/ in the environment preceding a vowel plus voiceless consonant. Rule 2 is a fronting rule and applies to /a/ in the environment preceding /u/. The Charleston dialect requires rule 1, the New Bern dialect rule 2, the Winchester dialect rule 1 followed by rule 2, and the Roanoke dialect rule 2 followed by rule 1.

⁶ This position is equivalent to Halle's view of ordering (64): ‘Since ordered rules are all but unknown in present day synchronic descriptions, the impression has spread that the imposition of order on statements in a synchronic description is always due to an oversight, to an unjustifiable confusion of synchronic and diachronic. I must therefore stress that … order is determined by the simplicity criterion alone and that no historical considerations have entered in establishing it.’ In a paper read at the meeting of the Linguistic Society of America December 1964, in New York, ‘Realism in historical English phonology’, Robert P. Stockwell took a different view, namely that historical accuracy might be adduced as a criterion in evaluating competing sets of ordered rules.

⁷ The rules for Spanish are largely illustrative. In the absence of much of the relevant material, it is not clear what modifications might be required, for example, in the underlying representations. These changes would not alter the main line of the argument.

⁸ This rule illustrates one of the insights which result from using features instead of the more familiar notation involving segments as the basis for phonological statements. The rule might be something like this:

⁹ There are exceptions, e.g. cocer ‘to cook’, which do not have a velar before /o/ or /a/. These forms have to be listed in any case.

¹⁰ Cf. Robert P. Stockwell, J. Donald Bowen, and John Martin, The grammatical structures of English and Spanish (to appear), who point out the parallel between the use of the velars in the verbs discussed here and in verbs like venir ‘to come’, vengo ‘I come.‘ In both sets the velar occurs before /o/ and /a/, that is, the + grave, – diffuse vowels.

¹¹ The rule is more general, applying to the palatalized nasals as well, cf. desdeñar ‘to disdain’, with palatalized nasal medially but the corresponding noun desdén ‘disdain’ with nonpalatalized nasal finally. Here too the relevance of features in such descriptions becomes clear.

¹² The proposal differs somewhat from the more traditional positing of special morphophonemes to cover different alternations. For example, the difference between /θ/ and /s/ is not that they alternate with different phonological units, but rather that they have different conditioning effects. Looked at from the point of view of internal reconstruction, if one were faced only with the data from LA, one might be tempted to suppose that in an earlier stage, the source of the final [s] in [lápis] was different from that of the final [s] in lúnes], as indeed it was, and similarly that the medial [y] in [akéyos] was different from the medial [y] in [léyes]. This is not to claim that the methods of internal reconstruction are sufficiently developed to provide correct formulations in any systematic way. Indeed the methods are bound to fail in at least those cases where the simplest set of generative rules does not exactly correspond to the historical processes involved.

¹³ For example, by Wáshington Vásquez, ‘El fonema /s/ en el español del Uruguay’, Revista de la facultad de humanidades y ciencias 10.87–94 (1953).

¹⁴ Results of the conference of anthropology and linguists 18 (Indiana University publications in anthropology and linguistics, memoir 8; 1953).