Comparing annotations for the prosodic segmentation of spontaneous speech: Focus on reference units

This chapter reports a quantitative and qualitative comparison of seven annotations performed on the same two American English texts: a monologue and a dialogue. The analysis of these data is complex, since the annotations have been made independently by each research group on the basis of their own theoretical frameworks. Despite this difference, the fundamental role of prosody in the analysis of speech emerges clearly in every annotation. Prosodic breaks can be then viewed as theory independent entities. After summarizing the key features of theoretical models, we derived a unified tagset and developed a web application (SLAC) to compare different annotations. Finally, agreement on prosodic breaks has been measured in different ways, reporting promising results in terminal break identification.