What happens to the phonetic prominence of a constituent when it is expected to have semantic focus, but occurs in a repeated or given context? For semanticists, the answer crucially discriminates between grammatically mediated and pragmatic accounts of association with focus. Extending the methodology of Rooth (1996) and Beaver et al. (2007), a production experiment was run with three linguistically trained, English native speakers using three verb-noun homophone pairs (e.g. patches/patches) in pre- and post-nuclear position. In a syntagmatic and paradigmatic comparison of syllable duration, acoustic energy, and spectral balance, absolute differences in acoustic prominence did not consistently correspond to those expected on semantic grounds; in fact, a strong positional effect was found. This is attributed to low-level rhythmic effects, based on results from a second production experiment. In a forced-choice context-retrieval task with linguistically trained listeners, subjects performed poorly (57.5% success), and in a rating task the percept of prominence could be coerced by context for at least some listeners. Methodologically, the results caution against taking absolute acoustic measures as direct correlates of prominence. Theoretically, it is argued that second occurrence focus must be semantically marked abstractly, if at all; pragmatics must be involved in its computation.