This paper presents a technique that allows us to detect similarities among prosodic labels used to describe pitch accents within the ToBI framework. The inter-label proximity is determined empirically as a result of the evidence obtained in contingency tables of inter-transcriber agreement tests and in the confusion matrices used in automatic prosodic labeling experiments. This tool may be useful to decide which labels can be grouped together when a simplified representation is required.
Index Terms: Prosodic labeling, ToBI pitch accents, tonal representation