Influential generative theories of prosody rely on phrasing to generate all sentence-level prominence: prominence can come from the edges or heads of phrases, but not from any other source. For English, one result of this approach is that pitch accent is treated as the head of a level of prosodic phrase just above the word. However, the evidence for this level of phrasing is inconsistent, and there is no evidence of a linkage between this level of phrasing and pitch accent placement. The rationale for linking the two together is generally that there are pitch accents (heads), so there must be phrases for them to head.

This dissertation investigates the question of whether there is support for including a level of prosodic phrase in the analysis of English prosody that is headed by pitch accent—that is, whether English has Accent Domains. To this end, two production experiments were conducted, both in American English and using read speech. Pitch accent position and syntactic structure were varied to create sentences that Accent Domain theories predict to have three different levels of prosodic unit: word, Accent Domain, and intermediate phrase. The segmental material around the boundaries of these units is kept identical, allowing for measurement of fine phonetic detail around the predicted boundary locations. The recorded items were measured for known phonetic markers of prosodic boundaries, including final lengthening.

The findings of these experiments are inconsistent with the proposals that the experiments were designed to test: final lengthening of the appropriate degree is found, but the distribution of the boundary marking is not consistent with these phrases being headed by pitch accent. Therefore, the use of this level of phrasing to generate pitch accent in English prosody is rejected and an alternative proposal—one that generates pitch accent independent of phrasing—is outlined. This proposal is formalized in the generative phonological framework of Optimality Theory; the crucial addition to the grammar that allows it to generate pitch accents without Accent Domains is the X > Y constraint type, which directly imposes conditions on the relative prominence of words in the output. The general properties of these constraints when incorporated into larger grammars for English and their possible uses in grammars for other languages are also discussed.