This chapter provides an overview of the theoretical background that piloted prosodic processing in Rhapsodie. We explain what the components of an autonomous data-driven usage-based perceptual model are, as distinct from the standard approach in which data are used to validate phonological theories, and how the methodological principles applied for prosodic annotation clearly distinguish Rhapsodie annotation from standard approaches using the ToBi system. This specificity is first expressed by our view of the relationships between syntax and prosody: we highlight the role of syntax, or rather the absence thereof, in the generation of the prosodic structure, and, instead, the central function of prominences and disfluencies in this respect. Finally, we recall how important acoustic processing is in order to properly achieve the prosodic modeling of everyday speech, which is often full of noise and, consequently, unusable.