We present our work on modeling the context of an interview during diagnostic sessions for patients with mental health problems. The results are to be exploited by translation system for telehealth services. More specifically, we plan to use the context of the psychiatric interview in order to set informative priors over the vocabulary of the speaker. Therefore we have modelled the context with a hierarchical ontology, and we use it to classify the current state of the interview. The state is extracted after the doctor asks a question, and allow us to select a non-uniform prior regarding the vocabulary of the patient.