Pragmatic communication, encompassing verbal and non-verbal cues, is a core aspect of social interaction. Schizophrenia is often associated with pragmatic deficits, impacting various communicative acts (direct/indirect, deceit, irony) and modalities (language, gestures, prosody). While previous studies have highlighted these deficits across different modalities, a comprehensive multimodal assessment aiming to identify the most reliable discriminative features is lacking. This study addresses this gap by performing a multimodal assessment of communicative-pragmatic abilities in individuals with schizophrenia and healthy controls. The primary goal is to identify the specific pragmatic features and expressive modalities that best differentiate these two groups. This information is crucial for improving diagnostic accuracy and targeting rehabilitative interventions. A machine learning approach, specifically Decision Tree analysis, is employed to identify the most informative features, offering advantages over traditional regression models in handling non-linear interactions and providing an interpretable representation of the data.

Extensive research demonstrates pervasive communicative-pragmatic difficulties in schizophrenia. Patients struggle with understanding non-literal language (irony, metaphors, indirect speech acts), recognizing violations of Gricean maxims (quantity, quality, relation, manner), and managing deceit. Non-verbal impairments also exist, affecting gesture comprehension/production, facial expression recognition, and prosodic patterns (flat intonation, pauses, atypical tone). While prior studies examined specific aspects of pragmatic abilities, few have conducted comprehensive multimodal assessments. Meta-analyses reveal significant differences between patients and controls across several domains, but identifying the most informative features for discrimination remains a challenge. The current study addresses this gap by employing a machine learning technique to analyze a wide range of pragmatic phenomena and expressive modalities, seeking to improve our understanding of the communicative profile in schizophrenia.

Thirty-two individuals with schizophrenia and 35 healthy controls, matched for age, gender, and education, participated. Participants underwent the Assessment Battery for Communication (ABaCo), a validated tool assessing various pragmatic phenomena across linguistic, extralinguistic, and paralinguistic modalities. The ABaCo consists of five scales: linguistic, extralinguistic, paralinguistic, context, and conversational. Each scale comprises several items in the form of live interactions and short video clips. Items are scored as correct or incorrect. A Decision Tree (J48 algorithm in Weka) was used to analyze the data, identifying the pragmatic features that best discriminate between the two groups. The model's performance was evaluated using 10-fold cross-validation, reporting accuracy, sensitivity, precision, specificity, and the area under the ROC curve (AUC).

The Decision Tree model demonstrated good overall performance, with a mean accuracy of 82%, sensitivity of 76%, and precision of 91%. Linguistic irony emerged as the strongest predictor for classifying participants, followed by violations of Gricean maxims of linguistic communication. Extralinguistic deceit and extralinguistic sincere acts were also significant predictors. The decision tree model visually represented how these features, combined with specific thresholds, best predicted group membership (schizophrenia vs. control). The results indicated a high probability of correctly classifying individuals with schizophrenia based on their performance in these specific pragmatic tasks. This highlights the significant impairment experienced by patients with schizophrenia in understanding and producing communication acts involving these features.

The findings confirm the presence of significant communicative-pragmatic deficits in schizophrenia, with linguistic irony and Gricean maxim violations being particularly informative for differentiating patients from controls. These results align with previous research demonstrating impairments in higher-level linguistic processes, theory of mind, and executive functions. The inclusion of extralinguistic measures underscores the importance of considering multimodal aspects of communication in schizophrenia. The Decision Tree model offers a valuable tool for identifying the most relevant features, which can be utilized for enhancing diagnostic procedures and informing the design of targeted rehabilitative interventions. Further research is needed to explore the neural correlates underlying these pragmatic deficits and to develop more effective interventions.

This study provides a comprehensive multimodal assessment of communicative-pragmatic abilities in schizophrenia, using a machine learning approach to identify the most discriminative features. Linguistic irony and Gricean maxim violations emerged as the most important factors for distinguishing patients from controls. The findings highlight the significance of multimodal assessment in understanding communicative deficits and emphasize the potential of machine learning for improving diagnostic and rehabilitative practices. Future research should focus on larger samples, diverse clinical profiles, and exploring longitudinal effects of interventions targeting these key pragmatic features.

The relatively small sample size is a limitation that could affect the generalizability of the findings. The heterogeneity of schizophrenia also needs to be considered, as clinical characteristics can influence performance. The specific pragmatic tasks employed might influence the results, necessitating replication with different task batteries. Further research with larger, more diverse samples and different assessment tools is warranted to validate the findings.