CENTER FOR RESEARCH IN
COGNITION & NEUROSCIENCES/ConnexConnex

Since the initial descriptions of psychotic syndromes, psychiatrists
showed the importance of identifying aberrant thought organization
through the language in order to predict clinical outcomes, such as
cognitive loss linked to Schizophrenia (initially named “dementia
praecox”). The description and qualification of formal thought disorder
remained complex and subjective until the advent of computational tools
that aim to quantify these symptoms on free speech. One strategy based
on graph theory studied non-semantic structural features by quantifying
spontaneous word trajectory topology. Initially, it was possible to
characterize that dream reports from patients with schizophrenia
diagnosis were represented by less connected graphs, and these aspects
of language were correlated with negative symptoms. These markers were
able to automatically identify reports from patients with schizophrenia
diagnosis with more than 90%accuracy, even during the first clinical
interview at the first psychotic episode. As these markers appear early
on, it was important to understand its typical development and cognitive
correlates. First, collecting memory reports at public schools in
Brazil, we were able to identify a correlation of graph connectedness
with IQ and theory of mind performances, and an independent correlation
with reading acquisition. More, studying memory reports from more than
200 subjects with different age (0 to 60 years old) and educational
level (0 to 20 years of education), we could identify in typical
subjects an asymptotic development of language graph attributes (as
lexical diversity, connectedness and graph length increased, short-range
recurrence decreased). These changes over time depended more on
education than age, and the stabilization of connectedness (language
marker of schizophrenia), required at least 13 years of education, or
high-school level. Importantly, for atypical development (subjects that
presented psychotic symptoms), there was no correlation of graph
attributes with age or education, presenting at adulthood a
children-like speech structure. Typical subject increased language
connectedness over school years, but subjects with psychosis seem to be
resistant to formal education, failing to mature in complexity and
remaining to a near-random language structure. These results point to
feasibility to search for early markers of cognitive risk in
naturalistic settings such as school environment, enabling early
identification and intervention to mitigate cognitive damages.