The Vowel Shift Rule and the Derived Environment Condition

Chomsky and Halle (1968) first proposed that the phonology of Present Day English incorporates a synchronic analogue of the Middle English Great Vowel Shift, namely the VSR. Although VSR has subsequently been the focus of much theoretical argument (Goyvaerts and Pullum 1975), and various changes in its formulation have evolved over the years (see Halle 1977, Rubach 1984, Halle and Mohanan 1985), the core of the original SPE rule remains in much post-SPE generative phonology.

In the light of increasingly serious attempts at constraining phono logical rules, two major objections must be raised against the SPE version of VSR and its successors in the more recent literature, both involving allegations of excessive abstractness. First, non-surfacing vowels and rules of absolute neutralization are frequently proposed to ensure the proper application of VSR; for instance, Halle and Mohanan use VSR to produce surface [jū] from back unrounded . Secondly, VSR applies to non-alternating forms, which are given free rides through the rule. Thus divine, which alternates with divinity, will be listed with a remote underlying vowel, but so will non-alternating forms like bee, house, pine, road, pain and cube. Consequently, in SPE, all tense or long vowels are underlyingly distinct from their surface realizations. The plausibility of this assumption, which entails the hypothesis that children learning Modern English internalize what is basically a Middle English vowel system (with the addition of various underliers which equally did not surface in Middle English) has been questioned elsewhere (Goyvaerts and Pullum 1975, Zwicky 1970, 1974).

Although this version of VSR applies to all tense, stressed vowels, it is motivated only in alternating morphemes. So, the supposed output of VSR is observable in divine because of the existence of related divinity, where no shift has taken place. Similarly, the alleged operation of VSR in sane, verbose, comedian and variety is evidenced by the absence of its results in sanity, verbosity, comedy and various. There can be no analogous direct evidence of Vowel Shift in non-alternating forms like bee, pain and road, so there is no motivation for assigning them abstract underliers, and deriving the surface vowels via VSR.

If the problem of free rides is to be solved, then, we must crucially find some way of restricting VSR to members of alternating pairs of words like those in (3.4).

As discussed, this restriction might be effected via the Alternation Condition; this will limit VSR to alternating morphemes, but will still allow it to operate in underived forms like divine, sane, verbose. To restrict VSR maximally, we must turn to the DEC (3.5), which can be imposed on the grammar as a formal condition on the proper application of Level 1 rules, and is potentially derivable, as we have seen, from the more general Elsewhere Condition. DEC must be the obvious candidate for a suitable constraint on VSR.

(3.5) DEC: Cyclic rules apply in derived environments. An environment is derived for rule A in cycle (i) iff the structural description of rule A is met due to a concatenation of morphemes at cycle (i) or the operation of a phonological rule feeding rule A on cycle (i).

Whatever the hypothetical desirability of constraining VSR using the DEC, however, this seems impracticable. Halle and Mohanan (1985) classify VSR as a non-cyclic, Level 2 process, precisely in order to exempt it from DEC, since the majority of forms traditionally supposed to undergo VSR constitute underived environments for it: they show no concatenation of morphemes, and no phonological rule feeding VSR has applied. But this is again to ignore the fact that VSR is only motivated in alternating pairs of words; if VSR could be restricted to the derived members of these pairs, it could be ordered on Level 1 within the domain of DEC, and the problem of free rides would disappear. Indeed, Borowsky's (1990) Principle of Domain Assignment, which allows free application of any rule not explicitly restricted to Level 1, will prevent the ordering of VSR solely on Level 2 on which Halle and Mohanan's analysis relies.

The restriction of VSR to derived environments is unproblematic for the forms in (3.4a). If VSR applies to tense, stressed vowels, the capitalized vowels in varIous, comEdy, courAge, stUdy and harmOny will be ineligible for shifting. However, in the right-hand forms in (3.4a), each of the corresponding vowels has undergone one of the tensing rules, which are triggered by affixation and in turn feed VSR (see (3.6)).

In the alternating pairs in (3.4b), however, the underived forms contain tense, stressed vowels, while the derived forms have short or lax vowels. Relocation of VSR on Level 1, subject to DEC, therefore commits us to a fundamental revision of the Vowel Shift Rule: the single rule shifting tense vowels will be replaced by two rules, one for tense vowels (V̄SR) and the other for lax vowels (V̄SR). V̄SR will be fed by the tensing rules; similarly, derived environments for V̄SR will be created by the laxing rules ± TSL in divinity, Suffix Laxing in satiric, and so on.

The possibility of shifting lax vowels is mentioned by McCawley (1986), who reports that Chomsky and Halle considered a lax-vowel VSR in the early 1960s, before replacing this with the tense-vowel VSR published in SPE. In their earlier version, `tense vowels retain their underlying heights and lax vowels shift their heights (in the opposite direction from the shift that tense vowels undergo in ... SPE)' (McCawley 1986: 30). The derivations predicted by this VSR are given in (3.7), but will be amended below.

Whereas Chomsky and Halle first proposed a vowel-shift rule for lax vowels, then adopted instead a rule shifting tense vowels, I assume that both V̄SR and V̄SR (formulated in (3.8)) are synchronic rules of Modern English; neither would be sufficient to account for the data in (3.4). The inevitable allegations of rule duplication and missed generalizations must be weighed against the solution to the problem of free rides which is supplied by splitting VSR and ordering both rules on Level 1, in the scope of DEC: some complication of the grammar is necessary in the interests of the principles. However, I believe that minor formal complications are far less important than the greater goal of producing a grammar which adheres to the principles and constraints of LP; in other words, the optimal grammar is not necessarily the simplest and most elegant, but the one which coheres best with both internal and external evidence, and in which the rules are bound by the constraints of the theory.

Interestingly, the DEC makes precisely the correct predictions here, accounting for the absence of V̆SR in damnable and solemnity and V̄SR in obesity and notify, although these forms initially look problematic. Consider solemn ~ solemnity. If the underlying representation is /sɒlεmn/, and if solemnity is derived from this by affixation on Level 1, it would be expected to be eligible for Level 1 rules, including VSR. However, if VSR did apply, the result would be *[sɒlæmnɪti]. Conversely, to produce [sɒlεmnɪti] after V̆SR, the underlier would have to be /sɒlɪmn/, which would give the wrong surface vowel in the underived form. The same applies to obesity, which might be expected to surface as [ōbaɪsɪti] by V̄SR.

This apparent exceptionality in fact follows from the failure of damnable, solemnity, obesity and notify to undergo any tensing or laxing rules in the course of the derivation (contrast, for instance, obese ~ obesity with obscene ~ obscenity, with TSL only in the last form). To clarify this assertion, we must return to the notion of derived environment embodied in the DEC. Although both VSRs appear to operate consistently in morphologically complex environments, it is not the addition of a morpheme per se which sanctions VSR, since neither Vowel Shift Rule demands a structural description which can be satisfied by morpheme concatenation. Both `ask for' a specific type of segment to apply to, but this environment is purely phonological - [+tense] vowels for V̄SR, and [-tense] ones for V̆SR. In contrast, TSL and CiV Tensing require certain combinations of segments to follow the focus vowel; since these configurations can be provided by adding a Class I affix, DEC can be satisfied morphologically. Level 1 tensing and laxing rules will then feed the appropriate VSR by supplying the derived features [+tense] or [7tense]. It follows that forms like obesity, notify, damnable and solemnity, which exceptionally fail to undergo tensing and laxing, will necessarily fail to meet the conditions for VSR.

The fact that both VSRs are clearly fed by preceding phonological rules makes these processes rather important theoretically: recall that Cole (1995: 76) suggested the replacement of DEC with the Revised Alternation Condition on the grounds that ‘there have been no ... examples in which a derived environment can be created morpheme-internally by the prior application of a phonological rule'. It is true that the tensing and laxing rules are triggered by affixation, and that the addition of a morpheme is crucial here, but the VSRs themselves are fed directly by tensing and laxing, and are co-morphemic with these rules. It follows that the DEC and any version of the Alternation Condition thought to be desirable must be stated independently, and that both clauses of the DEC must be retained.

In the rest, I shall examine some potential problems for the account of Vowel Shift sketched above. In 3.3, problematic aspects of the lax-vowel V̆SR are discussed; these include the derivation of the divine ~ divinity alternation, the generation of the high and low back vowels, and the analysis of [jū]. In 3.4, I shall consider difficulties for Level 1 VSR, concerning interacting rules and the Modern English irregular verbs, which present a test case on how far the formulation of Vowel Shift proposed here itself limits the adoption of abstract under lying representations.