Interacting rules

If VSR applies on Level 1, so must any rules which are crucially ordered earlier. Halle and Mohanan (1985: 103±4) list various tensing and laxing rules, q-Lengthening, ɨ-Rounding, Velar Softening and their ablaut rules for strong verbs, as necessarily preceding VSR. Since they order all but a subset of the tensing and laxing rules on Level 2, this poses an obvious problem for a model restricting VSR to Level 1.

Some of Halle and Mohanan's rules can immediately be discounted. Since [jū] and related vowels can be derived without recourse to / ɨ ɨ̄ /, q-Lengthening and Rounding are not required. I shall also argue below that most strong verbs should be dealt with allomorphically rather than derived through a set of ostensibly regular phonological rules; I therefore reject Halle and Mohanan's ablaut rules. This leaves tensing, laxing, and Velar Softening.

It should be clear from 3.2 above that the interaction of tensing and laxing with VSR is even more crucial in this account than in Halle and Mohanan's. I assume that Trisyllabic, Suffix (before /-ɪk/, /-ɪd/ and /-ɪʃ/) and Pre-Cluster Laxing are all Level 1 processes, so that the interaction of V̌SR and laxing will be unproblematic. Not all tensing rules can be similarly ordered: Stem-Final Tensing must apparently operate on Level 2, since it affects underived vary, city. However, since Stem-Final Tensing does not feed V̄SR, this is irrelevant. CiV and Prevocalic Tensing, which do feed V̄SR (as in comedian and algebraic respectively), can both be regarded as Level 1 rules. Halle and Mohanan restrict Prevocalic Tensing to Level 2 on the basis of its alleged operation in underived ammonia, but since this never alternates with any form containing a lax vowel, I assume tense /ī/ underlyingly. Prevocalic Tensing will then be restricted to cases like algebraic, where tensed, shifted [ē] alternates with lax final [æ] in algebra.

We have now exhausted Halle and Mohanan's list of rules, with the exception of Velar Softening of /k g/ to [s ʤ], which must apparently be ordered on Level 2 since it applies in underived reduce, oblige; some further examples are given in (3.41).

Velar Softening is often cited as part of the internal evidence for the synchronic status of VSR, on the grounds that Velar Softening is hard to formalize unless it precedes Vowel Shift. If VSR applies first, the context for Velar Softening will consist of a following front high tense [ī], the lax monophthongs [ɪ] or [ε], and the diphthong [aɪ]. However, if Velar Softening applies to pre-VSR representations, the context is the far more natural class of /ī ē ɪ ε/ ± any non-low, non-back following vowel.

The facts of Velar Softening seem irreconcilable with a Level 1 Vowel Shift. However, Jaeger (1986: 76-7), reviewing the use of evidence from rule interaction in establishing the order and reality of rules, argues that `... before an internal claim of this sort can be convincing, the synchronic psychological reality and the phonetic accuracy of each rule must be substantiated'. Such substantiation seems unlikely for Velar Softening, which is not fully productive, and therefore applies to a lexically specified class of inputs, as shown by the contrasting softened and non-softened forms in (3.42).

In SPE, velar segments which are to undergo Velar Softening are lexically /k^d g^d/, where the superscript ^d corresponds to a diacritic [+derived], to distinguish them from non-softening /k g/. Rubach (1984: 27) supports this lexical marking of the relevant `subclass of Greek and Latin words'. If speakers learn the specific morphemes which undergo Velar Softening, it is questionable whether achieving greater naturalness in the statement of the conditioning context of the rule, by ordering VSR after it, is of particular relevance or help (McCawley 1986: 30). It follows that the consequences which moving VSR to Level 1 will have for Velar Softening do not constitute a strong enough argument for revoking this step and retaining VSR on Level 2.