Akademik Veri Yönetim Sistemi
2026 Congress of the Schizophrenia International Research Society (SIRS) , Florence, İtalya, 25 - 29 Mart 2026, ss.980-981, (Özet Bildiri)
Background: Schizophrenia spectrum disorders (SSD) are characterized not only by core symptoms such as delusions, hallucinations, and formal thought disorder but also by profound emotional disturbances. Emotional abnormalities, commonly reflected as negative symptoms including blunted affect, anhedonia, and avolition, play a crucial role in functional impairment and reduced quality of life. Thus, identifying emotional expressions in the speech of patients with SSD is important for understanding their affective disturbances. Nevertheless, assessing these emotional deficits remains challenging due to the subjective nature of traditional clinical evaluations.
State-of-the art natural language processing (NLP) offers promising tools for the objective, rapid, and replicable analysis of emotional expression in schizophrenia. In this study, we aimed to investigate emotional abnormalities in the speech of patients with SSD through NLP-based emotion analysis.
Methods: Speech samples were collected from 40 patients with SSD and 30 age- and sex-matched healthy controls during a free conversational speech task. All recordings were manually transcribed, and the resulting texts were preprocessed using the Natural Language Toolkit (NLTK) to segment them into sentences.
A pretrained BERT model was then employed to classify emotions within the text according to the GoEmotions taxonomy, encompassing 28 categories: admiration, amusement, anger, annoyance, approval, caring, confusion, curiosity, desire, disappointment, disapproval, disgust, embarrassment, excitement, fear, gratitude, grief, joy, love, nervousness, neutral, optimism, pride, realization, relief, remorse, sadness, and surprise. For each transcript, the probability of each emotion was computed at the sentence level and subsequently averaged to obtain overall emotional profiles.
Results: Patients with SSD showed reduced probabilities of amusement (p = 0.037, Cohen’s d = 0.52), desire (p = 0.035, Cohen’s d = 0.52), joy (p = 0.013, Cohen’s d = 0.62), love (p = 0.022, Cohen’s d = 0.57), optimism (p = 0.017, Cohen’s d = 0.59), and realization (p < .001, Cohen’s d = 0.91) during free conversation. In contrast, anger (p = 0.009, Cohen’s d = -0.65), annoyance (p = 0.024, Cohen’s d = -0.56), and curiosity (p = 0.035; Cohen’s d = -0.52) were significantly higher in the speech of patients with SSD compared to healthy controls.
Discussion: In this study, we investigated emotional abnormalities in the speech of patients with SSD using an objective NLP-based emotion classification model covering a broad range of emotional categories.Our findings revealed that positive emotions such as amusement, desire, joy, love, and optimism were reduced in patients’ speech, when they talked about themselves or their daily activities and hobbies. This result may indicate negative symptoms, such as anhedonia, can be detected through automated language analysis. Regarding positive symptoms, decreased realization may reflect altered sense of reality, while increased curiosity could be associated with paranoid delusion in schizophrenia. Future studies should investigate the relationship between these automated emotion features and clinical scales. Overall, these findings highlight the potential of NLP-based approaches as objective, efficient, and scalable tools for assessing emotional expression and psychopathology in patients with psychosis.