Psychosis is a severe, common, and disabling psychological disorder. An epidemiological study conducted in England reported an incidence of 34 new cases per 100,000 person-years, with a peak between 16 and 19 years of age. Following a first psychotic episode, two clinical evolutions are possible: thymic psychosis (17%) and non thymic psychosis (83%). The first includes bipolar disorders with a psychotic component and major depressive disorders with a psychotic component; the second, other psychotic disorders, mainly schizophrenia. One of the major difficulties encountered is the frequent impossibility of specifying the type of psychosis at the beginning of the psychotic episode. However, these disorders require different therapies, particularly medication. This leads to a delay in diagnosis with a high risk of relapse. The semiological study of these diseases being carried out within the framework of interviews, it seems interesting to be able to record these and to obtain a quantitative and objective measurement through the study of language. The use of machine learning has made it possible to distinguish patients with schizophrenia from those with bipolar disorder by graphical analysis of language in a more efficient way than with clinical scales.Moreover, it is possible to identify linguistic markers: thus, an alteration of syntactic structures and prosody would be more present in non-thymic than in thymic psychoses. Paraclinical markers are also emerging. In particular, the link between inflammation and mental disorders.For example, an increase in IL-8 has been found only in thymic psychoses. In this context, it seems essential to be able to distinguish these disorders as early as possible through the combined use of clinical and paraclinical markers, and to be able to better understand their pathophysiology.