The facts of English phonology, where the majority of phonological rules apply on only one level, motivated Kiparsky's (1982) hypothesis that `the phonological rules at each level of the lexicon and in the postlexical component constitute essentially independent mini-phonologies' (1985: 86). Each rule is assigned to a particular level or component, and each level in turn is defined by the rules which are located there. Although this model is perhaps suitable for English phonology in the unmarked case, processes which must apply in more than one component, like Palatalization, would have to appear twice or more in the grammar, in this approach.

Mohanan (1982) and Mohanan and Mohanan (1984) argue that such a model is untenable for Malayalam, a language with much more overlap between lexical levels and between the postlexical and lexical components. Rather than multiply listing each rule, Mohanan (1982, 1986) proposes that the rules should each be listed once, but that each should carry a domain specification. Mohanan claims that this notion of phonological modularity parallels developments in syntax. In early transformational-generative syntax (Chomsky 1965), rules `belonged to' individual modules; in Government and Binding theory (Chomsky 1981), however, rules are essentially independent of modules, so that `the same set of rules is allowed to apply in multiple modules, with different consequences' (Mohanan 1986: 13). Kiparsky (1985) accepts this revision of domain assignment, and suggests that the marking of rules for application on particular levels may be more restricted than Mohanan's model implies, because the constraints of LP, which operate differently in different modules, may themselves restrict rule operation; consequently, apparently quite different processes may be recognized as lexical and postlexical applications of the same rule, with distinct inputs and outputs determined by the differential application of principles like SCC and SP in the two components.

Kiparsky tentatively concludes that `it may, in fact, be possible to restrict the marking of domains to specifications of the form “rule R does not apply after level n”’ (1985: 87). A more extreme statement of the same kind of view is Borowsky's (1990: 3) Strong Domain Hypothesis, which states that `all rules which are marked for a particular domain of application apply at Level 1 only'. All other rules are available throughout the phonology, and apparent restrictions to certain levels result from the principles of the theory, not from any rule-specific stipulation.

In Borowsky's model, the unmarked mode of application would involve operation both lexically and postlexically, and at all lexical levels. Note, however, that Borowsky's hypothesis refers to potential application, with actual application often severely restricted by the constraints of LP. Her proposal cannot therefore be invalidated simply by observing that there are apparently few, if any, rules which do apply on all levels and in both components. Mohanan (1986: 46±7) takes a different, and weaker view; his principles of domain assignment (given in (2.24)) make postlexical application only the unmarked option. Evidence on the relationship of sound changes and phonological rules will suggest that the postlexical level is the unmarked domain for newly introduced rules; lexicalization may then proceed.

(2.24) In the absence of counterevidence, choose the minimum number of strata as the domain of a rule.

In the absence of counterevidence, choose the lowest stratum as the domain of a rule.

The domain of a rule may not contain nonadjacent strata.