The Problem of Identifying Text Markers of Depression and Depressiveness in Automatic Text Analysis

The paper examines the interdisciplinary topic of the possibility of determining the psychological characteristics of authors from their texts, which may be useful for artificial intelligence methods. The aim of the study was to identify textual markers of depression and depressiveness. For this purpose, a study of two corpora of texts was carried out using a linguistic analyzer developed at the Federal Research Center “Computer Science and Control” of the Russian Academy of Sciences (FRC CSC RAS). One corpus consisted of 557 essays written by patients with clinical depression ( N = 110) and healthy subjects ( N = 447), and the other was formed by 224 social media posts written by people with high (89) and low (135) scores of depressiveness on the Beck Depression Inventory (BDI). In total, data on 108 text parameters were obtained for both corpora. The authors identified textual features common and specific to the texts of the depressed patients and the texts of those with a high level of depressiveness according to the questionnaire data, and provided their psychological and linguistic interpretations. At the same time, not only lexical features were taken into account, but also grammatical ones (in the broad sense), such as parts of speech, morphemes, grammemes, locative, temporal and causal noun phrases, indicators of text segmentation and text coherence, etc. Based on the results of the analyses, three complex indicators of depression were proposed, including a number of specific psycholinguistic, linguistic and psychological markers. For the texts of the subjects with signs of depression according to the BDI, markers were selected from social media messages, which were combined into two complex indicators. They are proposed to be considered in mass surveys as indicators of dissatisfaction (hostility) rather than depression. The authors also discuss the theoretically and experimentally identified problem of identifying text markers of depression and formulate proposals on the methodology of using AI tools in network psychodiagnostics.