The first part of the paper contrasts two different systems for representing the stress or accentuation of English words—that exemplified by Chomsky and Halle (1968) and that found in many English-language dictionaries. The main difference lies in the treatment of "full" (as opposed to "reduced") vowels: For Chomsky-Halle full vowels are always stressed; for the lexicographers they can be either accented or unaccented. Whereas the Chomsky-Halle notation concentrates on degrees of prominence among syllables, the dictionary representation corresponds to a surface representation of underlying metrical foot structure. We claim that the latter notation provides a simpler and more elegant characterization of English word accentuation. Adopting this perspective we show that English words typically are composed of binary and ternary feet that are left-headed, whereas unary feet and unfooted syllables are restricted to the right and left edges of words, respectively.
The main body of the paper provides an extensive analysis of English word accentuation. First, we consider morphologically simple (underived) words, and we present the rules for creating their foot structures and associated accents. Then we turn to morphologically complex (derived) forms.We adopt a noncyclic approach where the rules for deriving foot structures apply simultaneously to each of the constituents of a derived word. We discuss accent preservation, where the syllable that is accented in a contained word retains that accent in the derived word. Nonpreservation happens whenever there is overlap of syllables in the feet of two adjacent constituents. Next, we offer a nonderivational (OT-style) account of the data, and we compare the ranked constraints to the derivational rules for deriving foot structures. In support of the noncyclic approach, we show how the constraints evaluate simultaneously the morphological constituents of a derived form.
The final part of paper presents a set of correspondence rules for converting an accentual representation to a notation with stress levels. In this way we are able to extend the analysis to phrases, compounds, and sentences and to show that similar principles govern the stress levels of these higher-level constituent types. We reject the depth-of-embedding typical of many previous analyses, and instead we argue for a flat phonological phrase, one composed of unary, binary, or ternary constituents.