Análise de grafos de textos literários: investigação de traços psicóticos

Abstract

Recent studies in computational psychiatry (Mota et al., 2012, Mota et al., 2014, Mota et al., in preparation) indicate significant structural differences between the speech of psychotic and healthy subjects. Graph analysis of these subjects’ autobiographic reports allows the quantification of psychotic speech symptoms, marked by poor, disconnected and repetitive speech. Based on a literal reevaluation of ancient texts, Julian Jaynes proposed that psychosis was socially prevalent until around 1,000 BC (Jaynes, 1976). This conjecture was corroborated for the first time by Diuk et al. (2012), who used Latent Semantic Analysis to demonstrate that semantic similarity to the introspection concept increased over time in Judeo-Christian and Greco-Roman texts from the first millennium BC. Here we set out to test the hypothesis that ancient texts are structurally more similar to reports from psychotic subjects than to reports from healthy subjects. To that end we assessed graph attributes in 448 historical texts from 3,000 BC to 2,010 AC, comprising six different cultures of Afro-Eurasia (Syro-Mesopotamian, Egyptian, Hindu, Persian, Judeo-Christian, Greek-Roman) and three more recent literary categories (Medieval, Modern, and Contemporary). A total of 14 graph attributes was assessed, comprising lexical diversity (number of nodes), connectedness (edges, largest connected component, largest strongly connected component) and size (diameter, average shortest path), recurrence (repeated edges, parallel edges, loops of 1, 2 or 3 nodes), and global attributes (average total degree, density, clustering coefficient). The results revealed a clear pattern: While lexical diversity and connectivity increased asymptotically over time, recurrence decreased. These findings are compatible with the notion that psychosis is an early trait of our species.