Against the underlying specification of an 'exceptional' English stress pattern

This paper is on English nouns, such as Mississíppi, that defy the general rule that a penultimate syllable receives stress only if it is heavy. Hayes (1980), and Halle and Vergnaud (1987) give these words a special underlying specification so that they can be subjected to the normal stress rules, including the rule of syllable extrametricality. The opposite tack is taken in Selkirk (1984), where extrametricality is lexically marked on the more common antepenultimately stressed words like Ameríca, and the Mississíppi pattern is generated by a grammar that contains no rule of extrametricality. In what follows, I argue that neither of these 'rule + underlying specification' approaches can explain the fact that both patterns are quite productive. I then apply Itô and Mester's (1993a) model of lexical constraint domains to construct a grammar that generates the two competing patterns, but still captures the core and peripheral status of antepenultimate and penultimate stress respectively. Because the proposal here is that the difference between the core and the periphery lies in whether anextrametricality-likeconstraint is active or not, I conclude with a reply to some criticisms of Selkirk's (1984) 'zero extrametricality' account of Mississíppi-type words that apply equally to the present analysis. The main thrust of the criticisms is that the consonant following the penultimate syllable of these words behaves as a coda, which is predicted if this consonant is underlyingly a geminate, as claimed by Chomsky and Halle (1968). The reply consists of a demonstration that independently needed constraints in the English grammar force the post-penultimate consonant into coda position, thus obviating the need for underlying specification that appears to have no contrastive function