In this paper we propose that prosodic structures in spontancous discourse exhibit both linear and superpositional characterictics, and that these reflect the different scopes of multi-tiered emotional and cognitive processes. The multi-tiered structure encompasses the syllable and word level, interphrase movement, and extended pitch level baseline rise and fall. Analysis of the data suggests that integration of the 3 different prosodic levels within an overall prosodic model provides a critical link for the generation of natural-sounding interactive speech systems.