In the area of large speech corpora, there is a demonstrated need for a common prosodic notation system that would allow easy data exchange, comparison of annotations and analyses, and automatic processing. A single, simple scheme of prosodic transcription could also form the basis of guidelines for non-expert manual annotation, and be used for linguistic teachingand research. We present a corpus-based approach for prominence labeling in French continuous speech, taking into account the effects of grammatical constraints on prominences perception.