The Strict Cyclicity Condition

Unquestionably the most important constraint of LP is the SCC, which restricts cyclic rules to derived environments, and therefore blocks free rides. However, the history of the SCC, and various problems with its formulation and operation, indicate that it may best be replaced by a modified version, in the form of the Derived Environment Condition, and perhaps supplemented by other, associated restrictions in LP.

The SCC originates in Kiparsky's (1973) Alternation Condition, which prohibited obligatory neutralization rules from applying to all occurrences of a morpheme, hence limiting such rules to alternating forms. This condition, however, was not statable as a formal condition on grammars, so that the entire output of a grammar would have to be checked for coherence with it. Other, equally stipulative conditions with approximately the same intended force were proposed at around the same period. For instance, Lass and Anderson (1975: 231) attempt to `define the constraints on permissible lexical abstractness and extra paradigmatic rule extension' using the formal statement in (2.42) (which they say is based on a suggestion by Bob Stockwell).

(2.42) a. Let there be a class P of formatives containing the phones pi,pj (where pi ≠ pj), of which it is true that pi ~ pj; and the rule R which produces the alternation pi ~ pj is well motivated.

b. Let there be potential classes, Pi, Pj, in which any morpheme mi ϵ Pi contains Pi, and any morpheme mj ϵ Pj contains pj, and it is not true that pi ~ pj in either class.

c. If the classes Pi, Pj are empty, i.e. it is always the case that pi ~ pj, then there is only one lexical representation /p/ occurring in all formatives that are members of P, and the rule R is a function R(p → pi) for some contexts, and a function R(p → pj) for the others.

d. If either Pi or Pj or both are non-empty, i.e. there is at least one class where pi or pj appears, but it is not true that pi ~ pj, then the non-empty class or classes must have separate lexical representations /pi/or/pj/.

And this is so even if the rule R which is well motivated for the alternation-class P could map some segment /p/ into pi or pj with no complication of the grammar.

As Lass and Anderson (1975: 231) paraphrase it, this condition means that `if any phone appears in a non-alternating form, it must be lexical in that form. No segment then which appears in non-alternating forms may not be lexical.' This, they point out, is the inverse of Kiparsky's Strong Alternation Condition: whereas the Strong Alternation Condition prohibits segments which do not appear phonetically from underlying representations, Lass and Anderson's condition requires segments which appear in non-alternating surface forms to be present underlyingly in those forms, even if theoretically derivable by a free ride through a rule formulated for alternating forms.

Like the Alternation Condition, however, Lass and Anderson's proposal permits the single underlier in alternating forms, /p/, to be differentially derived to [pi]and[pj] in the two shapes of an alternating pair like, say, divine and divinity, where the underlier consequently need not be identical to either surface form. Kiparsky (1982) took a step towards more surface-true representations even in alternating forms by formulating the Revised Alternation Condition, which restricts neutralization rules to derived environments; this is then further recast as the Strict Cyclicity Condition.

Kiparsky's main argument for the SCC is that it `does not have to be stipulated in the theory. A version of it is deducible from the Elsewhere Condition' (Kiparsky 1982: 46). This deduction rests on the assumption that each lexical entry, as well as the output of every morphological process, is an identity rule. If we accept this, then, in the case of Trisyllabic Laxing, the rules /nītVngæl/ and TSL will be disjunctive by the EC: the structural description of the identity rule properly includes that of TSL, and the result of applying them would be different, since TSL would give a lax vowel where /nītVngæl / specifies a tense one.

Kiparsky therefore argues that these two constraints are subsumed by `the essentially trivial Elsewhere Condition, which may conceivably be reducible to a more general cognitive principle' (Kiparsky 1982: 58). Mohanan and Mohanan (1984), however, challenge this conclusion, on the grounds that Kiparsky's identity rules lack independent motivation. Furthermore, Kiparsky gives no reason why his identity rules should be limited to Level 1; unless this restriction applies, we would expect SCC to operate throughout the lexicon, while in fact it seems to be violated freely on Level 2.

A stratagem for deriving SCC from EC without badly motivated identity rules has now been promoted by Giegerich (1988, in press). Giegerich adopts Selkirk's (1982a) hypothesis that Stratum 1 affixation operates on roots, while Stratum 2 processes require words, and proposes a general Root → Word rule (2.43) which performs the conversion necessary to allow Stratum 1 forms to be input to Stratum 2.

(2.43) [ ]r ? [[ ]r]L (where L = N, V, A)

Since roots are acceptable only on Level 1, Root → Word rules can apply only here; Giegerich argues that (2.43) will be the final process on Level 1. If Level 1 is the sole cyclic stratum, then the link of application in derived environments and cyclicity remains. However, this connection may be purely fortuitous: SCC may not be a property of cyclic rules at all, but of non-final strata. We might then choose to rechristen it the Derived Environment Condition (DEC).

This line of argument is also followed by Cole (1995), who points out that SCC has two components. First, there is the DEC. Secondly, there is what Cole (1995: 72) calls the Reaching Back Constraint, which `prevents a cyclic rule R applying on cycle j from reaching back inside an earlier cycle i to apply to a string contained wholly within cycle i'. Cole points out that very few analyses in LP have made use of the Reaching Back Constraint, and argues that those which do are subject to reanalysis, or have been pre-empted by more recent phonological developments. Cole further argues that the DEC, the remaining portion of the earlier SCC, should be replaced by the earlier Revised Alternation Condition. This would capture the connection with derived environments, but would be a stronger constraint as it would not be limited to Level 1, or even to the lexicon, but might be expected to hold of all neutralization rules throughout the phonology. As this condition seems too strong to deal with the behavior of many postlexical rules, Cole proposes to limit the Revised Alternation Condition to the lexicon by exploiting the lexical interaction of morphology and phonology, and restricting neutralization rules to morphologically derived environments. She accepts that this is stipulative, but points out that there is empirical support for the move, since `in the two decades of research since the proposal of the Revised Alternation Condition, many examples have been cited in which rules apply only across a morpheme boundary, yet there have been no additional examples in which a derived environment can be created morpheme-internally by the prior application of a phonological rule' (1995: 76). I shall show that the Vowel Shift Rules of Modern English constitute just such examples, so that the DEC will have to be stated independently of the Revised Alternation Condition (whether or not it is derivable from the EC). The question then is whether the DEC and some version of the Alternation Condition are both required, or whether we need only the former.

Giegerich (in press) argues that there is a place in LP for the Alternation Condition as well as the DEC. He follows Kiparsky (1982: 36), who sees the Alternation Condition as `a strategy of language acquisition which says that a learner analyzes a form ``at face value'' unless he has encountered variants of it which justify a more remote underlying representation'. Anderson (1981: 530) makes a similar point when he claims that `the language learner does not hypothesize an underlying form distinct from the observed surface form without some positive evidence to support doing so'. The Alternation Condition, in Giegerich's proposal, is still not a formal condition on grammars, but can be seen as an informal constraint imposed during acquisition. It will, however, be pre-empted by the stronger DEC on Level 1: the Alternation Condition limits neutralization rules to alternating forms, but permits application in either the derived or underived member of an alternation. On the other hand, DEC enforces application in derived environments only. Thus, the Alternation Condition would allow Vowel Shift in either divine or divinity, precisely because both exist. As we shall see, DEC would limit Vowel Shift to derived divinity, were it to constrain the rule. All of this would mean that the optimally learnable grammar would be one with no structure-changing rules on Level 2. Giegerich (in press) argues that such rules will exist on a temporary basis, prior to relocation on Level 1, and we shall adduce further evidence for this idea.

In what follows, I shall assume the following constraints, which replace the SCC, extend some of its effects onto Level 2, and control the shape of possible underlying representations.

● The Derived Environment Condition limits Level 1 rules to derived environments.

● The Alternation Condition, although more stipulative, means we shall limit lexical rules to Level 1 wherever possible, rather than allowing them to apply in non-alternating forms. Level 2 application will be permitted only when there is positive evidence, for instance interaction with the Level 2 morphology.

● Finally, I propose to adopt and strengthen one aspect of Lass and Anderson's suggestion given in (2.42) above. Lass and Anderson assume that, in a class P where [pi] alternates with [pj], there will be a single underlier /p/, from which [pi] and [pj] will both be derived by rule. Bearing in mind Anderson's (1981: 530) argument that `the language learner does not hypothesise an under lying form distinct from the observed surface form without some positive evidence to support doing so', and given that learners will generally encounter morphologically underived forms earlier, I assume as a working hypothesis that the underlying representation for an alternating form will be equivalent to the lexical representation of the underived member of the alternating pair in the first instance. That is, factoring out the effects of postlexical rules, the underlier will necessarily be either /pi/or/pj/ , whichever appears at the lexical level in the underived form. We will then explore the derivation of the surface forms from this underlier as an initial possibility, before entertaining the option that restructuring has taken place during acquisition to produce an underlier equivalent to the lexical representation of the derived form, or intermediate between the two.

The connection of this last restriction to the Alternation Condition follows from the invocation of learning strategies and the idea of an optimal grammar in acquisitional terms; its links with the DEC will be observed below, in that it will frequently result in the limitation of phonological rules to Level 1. To illustrate these connections, we turn to one of the best-known processes of Modern English segmental phonology, Vowel Shift.