Abstract
BACKGROUND: Language impairments are common in affective and psychotic disorders, yet their patterns and underlying pathomechanisms remain insufficiently understood. A transdiagnostic perspective provides a framework for identifying shared and disorder-specific language alterations across diagnostic boundaries. Combining natural language processing (NLP) with network analysis enables the investigation of complex associations between linguistic, cognitive, and psychopathological features. METHODS: Spontaneous speech from N = 372 participants (119 MDD, 27 BD, 48 SSD and 178 HC) was elicited using four Thematic Apperception Test pictures (~12 min per participant). NLP models were applied to extract latent linguistic variables across various levels, including lexical diversity, syntactic complexity, semantic coherence, and disfluencies. Network analysis was used to relate linguistic variables, psychopathology (SAPS, SANS, HAM-A, HAM-D, YMRS, TLI, GAF), and cognitive performance (attention, verbal memory, recognition, and verbal fluency). RESULTS: Linguistic variables formed the densest network cluster, with type-token ratio, mean length of utterance, and syntactic complexity emerging as central nodes. Psychopathology variables were less cohesive, while TLI "Impoverishment", coherence mean, and executive functioning bridged linguistic, cognitive, and psychopathological domains. Network comparison tests revealed no significant differences in linguistic-cognitive network structure across HC, MDD, BD, and SSD. CONCLUSIONS: Linguistic networks show high structural consistency across healthy individuals and patients, whereas psychopathological symptom networks reflect transdiagnostic profiles. These findings support a dimensional and transdiagnostic framework underscore shared language-cognition mechanisms, and highlight executive functioning as key cross-domain connection, which opens up new avenues for dimensional research into the pathophysiological and etiological mechanisms underlying language dysfunctions.