The Delta of Language

Mental life fluctuates, changing from moment to moment as the incessant and turbulent flow of thought rages. In 3% of the worlds population, the fragile equilibrium that we all hope to maintain gives way to dynamical changes resulting in psychotic episodes, which after remission tend to recur over time. A capacity to predict them at a clinically relevant temporal resolution, similar to our capacity to forecast a thunderstorm, would be a major advance in public health, with the important difference that, unlike thunderstorms, psychosis can be prevented. We test a previously untested and untestable hypothesis: that meaning encoded in spontaneous speech, translated into digitalized quantitative features, computationally analyzed with natural language processing (NLP) tools, can serve as a key personalized and interpretable predictor of phase transitions from remission to psychotic relapse. Addressing this hypothesis requires conceptual and methodological breakthroughs in our understanding of language and what signals its variability can carry for a pathophysiological process. We pursue these with a synergetic combination of linguistic, neuroimaging, psychiatric, and e-health insights. Using a hypothesis-driven approach we will (i) define generalizable language metrics relating to symptom variability cross-sectionally; (ii) identify individually-specific neural signatures of psychosis and remission related to changes in these linguistic metrics, using a dense-sampling approach; (iii) test the metrics retrospectively as predictors of relapse in an independent longitudinal cohort; and (iv) take the paradigm from the lab to the patients home in a prospective clinical study testing whether we can predict state change before it is catastrophic, at the temporal resolutions clinically required. DELTA-LANG (LANG) thereby identifies the delta of language a linguistic change that enables the prediction of clinically significant change, before it occurs.