To formulate a diagnosis of Schizophrenia, clinicians perform time-intensive interviews and analyze the patient’s thoughts, language, and communication disorders, often requiring subjective interpretation. This paper presents DeepTald, an explainable AI system developed to assist clinicians in automatically scoring selected symptoms from the Thought and Language Disorder (Tald) scale. In close collaboration with psychiatrists, we focus on four core symptoms: Logorrhea, Slowed Thinking, Perseveration, and Rumination, chosen for their diagnostic relevance and compatibility with speech-based analysis. DeepTald integrates Natural Language Processing (NLP) techniques, including BERT-based semantic modeling, sentiment analysis, and temporal speech metrics, to compute interpretable on biological characteristics like veins, gaits, iris, fingerprints, signaturesmetrics and generate detailed reports. A key feature of DeepTald is its human-in-the-loop design, which enables clinicians to modify thresholds and parameters based on their expertise. In a preliminary exploratory study involving three experienced clinicians, DeepTald was perceived as usable, understandable, and aligned with clinical reasoning. Future research will involve real-world clinical data and a more diverse group of practitioners to evaluate DeepTald’s broader applicability and robustness. Nevertheless, this work highlights the potential of explainable NLP tools to enhance the objectivity, efficiency, and transparency of psychiatric assessments.
DeepTald: a System for supporting schizophrenia-related language and thought disorders detection with NLP models and explanations
Rita Francese
;
2025
Abstract
To formulate a diagnosis of Schizophrenia, clinicians perform time-intensive interviews and analyze the patient’s thoughts, language, and communication disorders, often requiring subjective interpretation. This paper presents DeepTald, an explainable AI system developed to assist clinicians in automatically scoring selected symptoms from the Thought and Language Disorder (Tald) scale. In close collaboration with psychiatrists, we focus on four core symptoms: Logorrhea, Slowed Thinking, Perseveration, and Rumination, chosen for their diagnostic relevance and compatibility with speech-based analysis. DeepTald integrates Natural Language Processing (NLP) techniques, including BERT-based semantic modeling, sentiment analysis, and temporal speech metrics, to compute interpretable on biological characteristics like veins, gaits, iris, fingerprints, signaturesmetrics and generate detailed reports. A key feature of DeepTald is its human-in-the-loop design, which enables clinicians to modify thresholds and parameters based on their expertise. In a preliminary exploratory study involving three experienced clinicians, DeepTald was perceived as usable, understandable, and aligned with clinical reasoning. Future research will involve real-world clinical data and a more diverse group of practitioners to evaluate DeepTald’s broader applicability and robustness. Nevertheless, this work highlights the potential of explainable NLP tools to enhance the objectivity, efficiency, and transparency of psychiatric assessments.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.