This paper presents a prosodic analysis of schizophrenic speech in an Italian corpus based on prominence labeling. Four recordings from a clinical setting (about 40 minutes total) were orthographically and phonetically transcribed. The transcriptions and annotations were aligned on the speech signal at various levels, including phonemes, VtoV, words, speakers, and overlaps. The aim is to compare an annotation conducted on this corpus in the functional framework of the Language into Act Theory (L-AcT) to the automatic detection of prominences using the data-driven model of ANALOR. First, we present the perceptual methodology used to conduct the manual annotation of the corpus in line with the principles of L-AcT. Second, we describe the implementation of the prosodic model for automatic prominence detection and its application to our data. Finally, we compare the manual and automatic output and discuss the advantages and limits of automatic data-driven labeling. Initial results show a close correspondence between the two approaches because ANALOR predicts the prominences annotated manually. It can, therefore, be a robust tool for the automatic annotation of prominences in Italian pathological speech.