From sound change to phonological rule
Standard Generative Phonology and Lexical Phonology
We have now built up a picture of the present-day SVLR, detailing its input, environment and ordering, and justifying its separation from postlexical LLL. However appropriate this separate characterization may be, it misses the intuition that the two rules are in some sense related, as evidenced by the inclusion of SVLR inputs and environments in the set of operational contexts for LLL. I shall argue that this relationship can be accounted for in diachronic terms, and that SVLR has been `derived' historically from LLL, and become a lexical rule fairly recently.

This development of SVLR will be shown to exemplify a probably rather common `life cycle' of sound changes, which may begin as low-level rules, then move into the lexicon, and eventually become opaque and promote restructuring at the underlying level, producing dialect and ultimately language variation. We shall see that LP reveals connections of synchrony and diachrony which were impossible to capture in SGP.

To recap, the Standard Generative approach to historical linguistics assumed that each sound change, once implemented, is incorporated directly into the adult speaker's phonological rule system as the final rule, moving gradually up into the grammar as subsequent changes are introduced (King 1969). Restructuring of the underlying representations during acquisition by later generations of speakers is theoretically permitted, but infrequently invoked, so that the historical phonology of a language will be almost directly mirrored in the order of its synchronic phonological rules. The only extractable generalizations are that the `highest' rules will correspond to the oldest changes, and that a sound change and the phonological rule into which it is converted will be identical or markedly similar ± although we have already encountered several rules, including the Vowel Shift Rules and SVLR itself, which differ significantly in their optimal synchronic statement from their historical source.
The SGP approach casts no light whatsoever on the implementation of sound change in a speech community, on which there are two, apparently diametrically opposed, views. The Neogrammarian position holds that sound change is phonetically gradual but lexically abrupt, while the lexical diffusionists (Wang 1969, 1977; Chen and Wang 1975) argue that many sound changes are, conversely, phonetically abrupt and lexically gradual. Labov (1981) aims to resolve this controversy by considering evidence from language change in progress; but the data include cases of Neogrammarian and diffusing changes, leading to an apparent impasse where we are `faced with the massive opposition of two bodies of evidence: both are right, but both cannot be right' (Labov 1981: 269). Labov's solution is to recognize two distinct types of sound change, differentiated by the characteristics in (1).
(1) 

Labov adds that Neogrammarian changes involve modifications to low-level output rules, while lexical diffusion causes a redistribution of some abstract class into other classes. Finally, he tentatively proposes that certain features are associated with certain types of change: for vowels, low-level, Neogrammarian sound changes will manipulate features of fronting, backing, raising, rounding and so on, while the more abstract diffusing changes will involve tensing and laxing, lengthening and shortening, and monophthongization and diphthongization.
In SGP, Labov's two types of sound change have no clear analogues. However, Kiparsky (1988) points out that the sets of properties characteristic of diffusing and Neogrammarian changes overlap to a considerable extent with the properties of lexical and postlexical rules shown in (2).
(2) 

Some of the criteria in the relevant columns of (1) and (2) match exactly: for instance, both lexical rules and diffusing changes have discrete, categorizable effects observable by speakers, may have lexical exceptions, and are sensitive to morphological information. Kiparsky argues that a number of less obviously connected properties are also related: for instance, a diffusing change may extend beyond its original conditioning context, producing lexical selectivity and therefore lexical exceptions; an incomplete diffusing change will also retain a residue of lexical exceptions. Kiparsky also relates the necessity for two dictionary entries, which Labov cites as a property of diffusing changes, to Structure Preservation, which states that no lexical rule may introduce or operate on a feature which is not underlyingly distinctive.

However, a complete identification of diffusing and Neogrammarian changes with lexical and postlexical rules respectively may be too inflexible, as not all lexical rules necessarily start out as diffusing changes; they may begin as low-level, automatic and phonetically motivated Neogrammarian changes, but subsequently percolate into the more abstract regions of the grammar, becoming synchronically lexical rules. Harris (1989) discusses one such example, the rule of æ-Tensing.

æ-Tensing applies in a number of varieties of Modern English, including the New York City, Philadelphia and Belfast dialects, producing tense [Æ] (which is typically realized as long, diphthongized and relatively centralized) before a variable class of tautosyllabic consonants.

In Philadelphia, tensing occurs only before anterior nasals and anterior voiceless fricatives; in New York, it applies additionally before voiced stops; and in Belfast, tense [Æ] also surfaces before /l/ and voiced fricatives. The examples in (3) would hold for all three dialects.
(3) 

Harris (1989: 48) proposes that ñ-Tensing was historically an automatic, phonetically motivated change, operating in the hierarchy of environments in (4).
(4) 

However, in the varieties mentioned above, æ-Tensing is now a lexical rule: it may be lexically selective, as in Philadelphia, where mad, bad, glad have tensed [Æ] although /d/ is not generally a tensing context in this dialect; and Labov (1981) reports that lax [æ] and tense [Æ] are subject to categorial discrimination by New York and Philadelphia speakers. æ-Tensing is also sensitive to morphological information, since it applies before heterosyllabic consonants followed by ], as in manning.

The only factor which might argue against the characterization of æ-Tensing as lexical is its contravention of Structure Preservation, since Harris follows Halle and Mohanan (1985) in assuming that [± tense] is not part of the underlying feature inventory for English. However, Harris tentatively suggests that newly lexicalized rules may violate Structure Preservation temporarily, with the reassertion of Structure Preservation perhaps determining the direction of future change, although he produces no clear evidence of this determinative role of Structure Preservation.

Harris's discussion of æ-Tensing suggests that sound changes may be incorporated into the synchronic grammar by passing through a number of stages. Changes may be phonologized as postlexical rules, but may subsequently acquire properties from the lexical syndrome, notably sensitivity to morphological structure, and become lexical phonological rules; these may initially violate Structure Preservation, but might be predicted to attain conformity with this principle over time. Newly lexical rules may also begin to diffuse, as is the case with æ-Tensing in Philadelphia, where the tense reflex is now appearing before /d/ in certain lexical items. Ultimately, a lexical rule may cease to be transparent and productive: for instance, the number of lexical exceptions may increase to a point where the rule is no longer readily learnable. The rule itself will then be lost, but its effects will be incorporated into the underlying representations, as is the case for æ-Tensing in RP. Here, the historical short /æ/ class has split, with the tense reflex merging with /A/ from other sources in path and laugh.

If these suggestions are substantiable, LP gains considerably in a number of domains. Labov's two types of sound change can be matched with credible synchronic counterparts, and his notion of more and less abstract changes linked with the lexical±postlexical division (although, as Harris notes, æ-Tensing shows that Labov's correlation of particular features with only one type of change or rule cannot be maintained). The lexicalization of rules and their eventual loss also provides a mechanism for altering underlying representations and for the introduction of surface and underlying variation between dialects. I shall show that SVLR provides further evidence for these proposals, and constitutes an arguably even clearer illustration of the life-cycle suggested above, albeit with some interesting differences from æ-Tensing.