Current psychiatric classification struggles with the heterogeneity and comorbidity of affective and psychotic disorders like MDD and SSD. While biomarkers remain elusive, speech features show promise as objective, quantitative, and readily obtainable measures. SSD is characterized by reduced speech production, verbal fluency deficits, word retrieval problems, neologisms, and less complex sentences. MDD presents with longer response latencies, reduced spontaneous speech, and potentially more truncated sentences. Formal thought disorder (FTD), encompassing various speech aberrations, is not unique to SSD and occurs in other disorders, including MDD. While reduced syntactic complexity is well-documented in SSD, its distinctiveness from MDD remains unclear. This study aimed to compare syntactic complexity and diversity in SSD, MDD, and healthy controls, explore associations with neuropsychological and psychopathological measures, and identify clusters based on syntactic performance using network analyses. The researchers hypothesized that SSD patients would produce less complex speech than controls and MDD patients, with a negative relationship between syntax and positive/negative symptoms.

The introduction extensively reviews existing literature on language-related symptoms in SSD and MDD. It highlights studies showing reduced speech production, verbal fluency, and sentence complexity in SSD. For MDD, the literature suggests longer response latencies and reduced spontaneous speech, along with differences in pronoun use and self-focused language. The review emphasizes the prevalence of FTD in both disorders and the need for transdiagnostic approaches. The authors note that while the association between reduced syntactic complexity and SSD is established, the comparison with MDD and the relationship with neuropsychological and psychopathological factors requires further investigation. The use of natural language processing (NLP) techniques in distinguishing these disorders is also mentioned, but the current study focuses on a manual syntactic analysis for deeper insight into complex sentences.

The study included 112 German-speaking participants: 34 with SSD, 38 with MDD, and 40 healthy controls. Participants described four pictures from the Thematic Apperception Test (TAT). Speech samples were audio-recorded and transcribed. Syntactic analysis focused on complexity (number of main clauses embedding subordinate clauses) and diversity (number of different types of complex sentences). Thirteen types of complex sentences were manually coded. Neuropsychological assessment included measures of executive functioning, verbal fluency (semantic, phonemic, alternating), and verbal episodic memory. Psychopathological assessment used the Global Assessment of Functioning (GAF), Hamilton Rating Scale for Depression (HAM-D), Hamilton Anxiety Rating Scale (HAM-A), Scale for Assessment of Negative Symptoms (SANS), and Scale for Assessment of Positive Symptoms (SAPS). Group comparisons used ANOVA or Kruskal-Wallis tests. Classification analysis (SVM) assessed the diagnostic utility of syntactic measures. Cluster analysis (random forest) identified transdiagnostic clusters based on syntactic performance. Network analysis (Gaussian Graphical Model with EBICglasso) investigated relationships between syntactic, neuropsychological, and psychopathological measures within and across clusters.

SSD patients exhibited significantly less syntactic complexity and diversity than both MDD patients and healthy controls. The difference between MDD patients and healthy controls was not significant. Classification analysis showed moderate accuracy in distinguishing SSD from healthy controls and SSD from MDD. Cluster analysis revealed four transdiagnostic clusters ranging from extremely complex to slightly complex speech. The distribution of diagnoses varied across clusters, with the slightly complex cluster predominantly composed of SSD patients. Participants with less complex speech showed poorer neuropsychological performance (verbal fluency, episodic memory) and more severe positive and negative symptoms (SANS and SAPS). Network analysis revealed strong within-domain connections and weak cross-domain connections in the overall sample. Cluster-based networks showed varying degrees of interconnectivity. Syntactic complexity measures were closely related within the network, while syntactic diversity was a separate node. Cross-domain associations were more pronounced in clusters with more complex syntax.

The findings confirm reduced syntactic complexity and diversity in SSD compared to both MDD and healthy controls, supporting previous research. The lack of significant difference between MDD and controls suggests that reduced syntax is a more specific feature of SSD. The weak performance of classification analyses highlights the limitations of using syntax alone for diagnostic purposes. The transdiagnostic clusters based on syntactic complexity show that the severity of psychopathological symptoms is related to the level of syntactic complexity. Network analyses indicated that more complex speech production is associated with stronger connections between syntax, neuropsychology, and psychopathology, indicating that the breakdown of these integrated processes are more prominent in the lower syntax clusters. The study underscores the need for transdiagnostic and dimensional approaches to understand psychiatric disorders and highlights the usefulness of combining linguistic, neuropsychological, and psychopathological data.

Reduced syntactic complexity and diversity in SSD is associated with poorer neuropsychological performance and more severe psychopathology. While syntax alone is not sufficient for diagnosis, a dimensional approach that considers transdiagnostic clusters based on syntactic performance provides valuable insights. Future research could involve larger samples, longitudinal studies, and the analysis of written language to further explore these findings and potentially develop more robust diagnostic tools.

The relatively small sample size and cross-sectional design limit the generalizability and causal inferences of the study. Education level was a confounding factor that could have influenced syntactic performance. While medication effects were not observed, the potential influence of lifetime medication exposure cannot be entirely ruled out. The manual syntactic analysis, while thorough, is less efficient than NLP approaches and limits comparability across studies.