The understanding of complex human behaviors such as language and its variations in different conditions and contexts has been an important research aim for many decades. Naturalistic and quantitative approaches to precisely measure language variations from the structural and semantic points of view have recently emerged, allowing the measurement of variations manifested in free speech that reflect atypical cognitive decline in pathological situations such as psychoses, or typical cognitive development in healthy children during alphabetization, and even the processing of memories in different states of consciousness, such as waking and dreaming. In this work we will start discussing 1) the construction of tools for the analysis of speech structure inspired by the psychopathological descriptions of mental illnesses. 2) their application to the differential diagnosis of psychosis and dementias, and 3) the application of semantic tools to predict psychotic episodes. In the structural level it was possible to observe that subjects with Schizophrenia diagnosis report their dreams with word trajectories represented as graphs less connected than subjects without psychosis or with Bipolar Disorder diagnosis. In the semantic level it was observed a higher semantic distance between consecutive sentences on psychiatric interviews of patients during prodromal psychotic phase 2 years and a half before converting to a psychotic episode. We will proceed by widening this view away from pathology, so as to determine 4) how graph-theoretical measures of language structure vary across healthy cognitive development, and 5) how they relate to indices of academic achievement. We verified a correlation between graph connectedness and cognitive (such as fluid intelligence and theory of mind abilities), as well as academic performances (of reading). Next we will investigate 6) how speech structure varies within a large sample of healthy and psychotic subjects with large age and educational variation, to 7) evaluate the impact of years of education and 8) compare with the development of literature across 5,000 years. In summary, connectedness increases after the Bronze Age and the longer time of education the subject had, higher the connected components of his memory reports, values stabilized during high school period, and a developmental trajectory not found in the psychotic population. We will conclude by applying tools to calculate semantic similarity to 9) measure memory reverberation during dreams and their electrophysiological correlates in a sleep transition experiment. The results indicate that the structural and semantic tools used in this work can greatly improve the precision of naturalistic measurements of the complex behaviors expressed in speech.