This paper presents the early stages of a growing corpus of psychiatric interviews from help seeking patients referred to an early detection and intervention center for psychosis. In order to contribute to the practitioner's diagnostic, we focus on a new method of automatic comorbidity detection in the corpus. Among the novelties of this method is the fact that it is based on syntactic features of paralinguistic data (interjections and pauses). We use the formalism of dependency syntax, a brief description of which we provide in the paper. Considering the (currently) small size of the corpus, our intention is to prove the applicability of the method rather than to obtain general results about the relevance of syntactic indicators.