Data on person marking in Andalusian Spanish argue in favor of a revision of the functional hypothesis, which pertains to “the tendency for semantically relevant information to be retained in surface structure” (Kiparsky, 1982:87). The first part of this study shows that subject pronouns in Andalusian are not used to disambiguate verb forms, in direct contrast to Hochberg's (1986) findings on subject pronoun usage in Puerto Rican. In the second part of the study, an analysis of contextual person markers, not undertaken in previous studies, shows that person can be indicated by means outside the surface structure. When both linguistic and contextual person markers are taken into account, only 2% of the 1,035 verbs in this sample of Andalusian are unmarked for person. Person marking is thus maintained where it is important to the speaker's message, but it is not limited to surface structure realization as the strong form of the functional hypothesis suggests.

In Korean, the tendency to delete the labiovelar glide w when it appears postconsonantally is conditioned by linguistic, stylistic, and social factors. Chief among these is the place specification for the preceding consonant: labiovelar glides are more likely to be deleted when preceded by a velar than by a labial, alveolar, or alveopalatal. It is proposed that this observation can be explained in terms of current phonological theory, in particular, the notions of feature geometry (Clements 1985; Sagey 1986) and the Obligatory Contour Principle (McCarthy 1986). Under such a model, adjacent segments share common features in ways that promote variability; specifically, the labiovelar glide shares the feature [back] with an adjacent velar (but shares no place features with adjacent labials, alveolars, and alveopalatals), thereby providing structural cohesiveness that inhibits the application of the deletion rule.

-t,d deletion is a well-known variable phonological process subject to the influence of both external social factors and internal structural constraints, including phonetic environmental and morphosyntactic effects. Its profile of variation has been widely investigated in American English dialects. However, it interacts with another grammatical process – the regular affixation of final /-t, -d/ as a past-tense marker – that strongly distinguishes these dialects from English-related creoles, where past-marking by this mechanism is infrequent or non-occurrent. Investigation of -t,d deletion in mesolectal Jamaican Creole (JC) thus raises important questions about the intersection of variable processes, the generality of phonetic environmental constraints, and the degree of difference between English-related creoles and metropolitan standard and non-standard Englishes.

Although there is no consensus on whether French has any true modal auxiliary, devoir ‘must’ is one of the very few possible candidates. Most studies of devoir concentrate on criteria for distinguishing its epistemic and root semantic values, but some also explore the subtle semantic nuances it conveys depending on context. Here, we identify discrete meanings through the analysis of semantic overlaps between devoir and other French modal expressions, such as probablement ‘probably’, être supposé ‘to be supposed to’, or falloir ‘ought to’. The analysis of data from two corpora on Montreal French, collected in 1971 and 1984, shows that:

1. There is only one root semantic value exclusively associated with devoir, namely, the notion of desirability (as in the English ‘should’).

2. There are different patterns of use of the competing modal expressions within each semantic field analyzed, according to the social class of the speaker. Moreover, some patterns characterizing a specific social group are acquired relatively late.

The variable lexical phonology proposed in Guy (1991) predicts an exponential relationship among rates of retention in word classes of different derivational histories. A class of words that satisfies the structural description of a variable rule at an early lexical level of derivation will undergo multiple operations of the rule and, therefore, exponentially reduced rates of retention, compared to a class of forms that only satisfies the structural description of the rule at the end of the derivation and thus is subject to its operation only once. For the case of English -t,d deletion, it is postulated that monomorphemic words (e.g., mist) are exposed to the deletion rule three times in a derivation, whereas semiweak past tense forms (e.g., left) are exposed twice, and regular past tense forms (e.g., missed) undergo the rule but once.

The present article explores the consequences of this model for other variable constraints on a rule, such as the preceding and following segment constraints on -t,d deletion. Word-internal constraints, because they are present throughout the derivation, are shown to have quantitatively different patterns than external constraints, as the latter affect the rule only in its final, post-lexical operation. Four specific quantitative predictions are derived from the model to elucidate this distinction between internal and external constraints, and empirical data are presented to confirm the predictions.