Textual forma mentis networks bridge language structure, emotional content and psychopathology levels in adolescents

Alexis Carrillo, Simon Friedrich Roske, Rebeca Ianov-Vitanov, Enrico Perinelli, Alessandro Grecucci, Massimo Stella · 2025

Read Textual forma mentis networks bridge language structure, emotional content and psychopathology levels in adolescents by Alexis Carrillo, Simon Friedrich Roske, Rebeca Ianov-Vitanov, Enrico Perinelli, Alessandro Grecucci, Massimo Stella (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-06387.txt.

Introduction

Alexis Carrillo, Simon Friedrich Roske, Rebeca Ianov-Vitanov, and three co-authors link teen speech to mental health in this two thousand twenty-five paper. They read talks with two hundred thirty-two teens in the Healthy Brain Network data set, each speaking on a set topic meant to stir feelings. Test scores spanned social maladjustment, internalizing acts, and neurodevelopmental risk. The team cast each talk as a textual forma mentis network. It joins word links of syntax, sense, and feeling.

Network traits and feeling profiles went into random forest and gradient boosting models that forecast the clinical scores. Speech traits strongly forecast all three score types. Explainable AI showed a tight, core-periphery form of ideas went with higher social maladjustment. Internalizing scores tracked with more disgust and key bridge words. The team links that mark to rumination. The frame hints speech form itself holds clues for diagnosis.

Catalog record

CREATORAlexis Carrillo, Simon Friedrich Roske, Rebeca Ianov-Vitanov, Enrico Perinelli, Alessandro Grecucci, Massimo Stella
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE74 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-06387

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Textual forma mentis networks bridge language structure, emotional content and psychopathology levels in adolescents Alexis Carrillo, Simon Friedrich Roske, Rebeca Ianov-Vitanov, Enrico Perinelli, Alessandro Grecucci, Massimo Stella Abstract We introduce a network-based AI framework for predicting dimensions of psychopathology in adolescents using natural language. We focused on data capturing psychometric scores of social maladjustment, internalising behaviors, and neurodevelopmental risk, assessed in 232 adolescents from the Healthy Brain Network. This dataset included structured interviews in which adolescents discussed a common emotion-inducing topic. To model conceptual associations within these interviews, we applied textual forma mentis networks (TFMNs)—a cognitive/AI approach integrating syntactic, semantic, and emotional word-word associations in language. From TFMNs, we extracted network features (semantic/syntactic structure) and emotional profiles to serve as predictors of latent psychopathology factor scores. Using Random Forest and XGBoost regression models, we found significant associations between language-derived features and clinical scores: social maladjustment (r = 0.37, p < .01), specific internalising behaviors (r = 0.33, p < .05), and neurodevelopmental risk (r = 0.34, p < .05). Explainable AI analysis using SHAP values revealed that higher modularity and a pronounced core–periphery network structure—reflecting clustered conceptual organisation in language—predicted increased social maladjustment. Internalising scores were positively associated with higher betweenness centrality and stronger expressions of disgust, suggesting a linguistic signature of rumination. In contrast, neurodevelopmental risk was inversely related to local efficiency in syntactic/semantic networks, indicating disrupted conceptual integration. These findings demonstrated the potential of cognitive network approaches to capture meaningful links between psychopathology and language use in adolescents. Keywords: Text analysis; Artificial Intelligence; Cognitive network science. Introduction The cognitive counterpart of language in the mind is a complex system in which ideas can be linked to convey meaning vitevitch2019network . Whereas language exists in its own physical terms (written or digital traces), the cognitive processes that power language can only be investigated through theoretical models and tested indirectly through laboratory experiments aitchison2012words ; harnad2017cognize . A powerful cognitive model is the mental lexicon aitchison2012words ; doczi2019overview , an idealised cognitive system apt at acquiring, storing and processing knowledge expressed via language. The mental lexicon is modelled as a highly structured and dynamical system, storing, updating and accessing cognitive representations of concepts over time hills2022mind ; vitevitch2019network ; harnad2017cognize . Over the last fifty years, a large body of empirical research has shown that the mental lexicon operates in associative ways doczi2019overview ; aitchison2012words : Concepts in the mental lexicon can be accessed via a variety of associations—semantic, syntactic, phonological, orthographic, emotional, and more. The structure of these associations influences how concepts are accessed and used doczi2019overview ; harnad2017cognize . For instance, the number of associations a given concept is involved into can influence its processing, memorisation and retrieval from memory vitevitch2022can . The interplay between networked associations and language processing positions complex networks as a naturally powerful tool for investigating the mental lexicon. The blooming field of cognitive network science stella2022cognitive ; Siew_2019_CogNetSci bridges the associative structure of knowledge in the mental lexicon with cognitive and psychological phenomena related to language, as well as with all knowledge expressible through language. Applications of cognitive network science include the use of network topology to explain phenomena such as lexical processing vitevitch2022can , creativity Kenett_2018_flexibility , individual perceptions aeschbach2025mapping , and social media perceptions abramski2024voices ; stella2020text , among others. A cognitive network can be defined as a complex network modelling associative knowledge in representational terms stella2024cognitive ; Siew_2019_CogNetSci ; castro2020contributions . In other words, cognitive networks can encode the layouts of several types of similarities between concepts in the mental lexicon, including semantic, syntactic, phonological, orthographic, emotional and several others Semeraro_2025_EmoAtlas . Multilayer cognitive networks can even encode multiple types of similarities at once among a single set of concepts stella2024cognitive . These models are representational in nature—they capture structure but do not, on their own, perform computations, unlike artificial neural networks (ANNs) stella2020text . Whereas in ANNs, nodes represent computational units used to approximate functions or solve regression and classification tasks, cognitive networks explicitly map the relationships between nodes that represent individual concepts—or, more simply, words. Beyond computations, can the representations encoded in cognitive networks also reflect psychological constructs such as psychopathology levels (as defined in prior research ( Holmes_2021_threefactor ) )? This research question warrants careful consideration in connection with the so-called Deep Lexical Hypothesis cutler2023deep . This hypothesis posits that mental states and psychological constructs can percolate through cognition to the extent that they alter an individual’s language. In recent years, a rapidly growing body of evidence has supported this hypothesis (see Literature Review), particularly within the field of mental health. Several studies—despite not employing a network science approach—have identified distinct linguistic markers associated with varying levels of psychopathology. Non-network linguistic markers of psychopathology Linguistic analysis reveals consistent associations between language patterns and psychopathology dimensions across various mental health conditions tausczik2010psychological . Research shows that people experiencing depression tend to use language that is more self-focused and negatively valenced. For example, they often use a disproportionately high number of first-person singular pronouns (e.g., “I,” “me”) alongside more negative emotion words (e.g., “sad,” “hopeless”) in their writing Edwards2017 . The frequency of these linguistic markers has also been used in clinical samples to develop successful machine learning pipelines for detecting levels of depression from single-word expressions stade2023depression . More specifically, both clinical interviews stade2023depression and online posts on mental health forums al2018absolute have shown that higher levels of depression are associated with increased use of first-person singular pronouns (e.g., “I”) and negative-emotion terms, suggesting that these verbal markers reliably reflect an individual’s depression state. In online forums, language from individuals experiencing depression was also found to contain a higher frequency of absolutist terms (e.g., “never,” “always,” “ever”). Anxiety disorders, such as generalized anxiety disorder, were found to manifest linguistically through elevated usage of threat-related words (e.g., "danger," "risk") and future-oriented language Geronimi2015 . These findings align with evidence that anxious individuals show hyper-associative thinking around threats: In cognitive experiments, people high in anxiety more readily interpret ambiguous words in a threatening way, essentially “locking on” to threat-related meanings Richards1992 . Always in the context of anxiety disorders, post-traumatic stress disorder (PTSD) is often accompanied by distinctive language features in trauma narratives. Studies of survivors’ written or spoken accounts find an overuse of visceral, sensory details (e.g., “scream,” “blood”) and disjointed or fragmented syntax when recounting a traumatic event ( Crespo2016 ) . This linguistic profile — rich in concrete imagery yet low in coherence — mirrors the intrusive memories and disorganised recollections characteristic of PTSD. More recent work ( pugach2025uncovering ) , identified a high rate of fluctuations of negative thoughs in PTSD patients, which is also related to PTSD severity (the higher the severity, the higher the fluctuations of negative thoughts). In the context of eating disorders like anorexia nervosa, language use often centers obsessively on body image and food-related terms. Analyses of online forums and support groups for eating disorders have revealed an excessive focus on body shape/size and weight (words like “thin,” “weight,” “fat”) as well as on restrictive eating behaviors (e.g., “fast,” “starve,” “calories”). In one content analysis of thousands of tweets referencing eating disorder behaviors, 65 \% expressed a preoccupation with body shape

Read the complete text