Computational Analysis of Semantic and Sentiment Features in Natural Language for Schizophrenia

Area of Science:

• Psychiatry
• Computational Linguistics
• Affective Computing

Background:

• Schizophrenia (SZ) is associated with impaired social functioning due to language abnormalities like semantic confusion and apathy.
• Previous research on SZ speech primarily focused on semantic/grammatical features, with limited analysis of emotional dynamics and linguistic fluency.
• This study addresses the gap by systematically analyzing emotional expression abnormalities in SZ patients using natural language processing (NLP) models.

Purpose of the Study:

• To systematically analyze abnormal emotional expression in schizophrenia patients.
• To investigate emotional dynamics and linguistic fluency in SZ using NLP.
• To identify distinct emotional expression patterns in SZ compared to healthy controls (HCs).

Main Methods:

• Recruited 78 SZ patients and 85 HCs for a free speech task.
• Participants described recent positive and negative events for one minute each.
• Employed NLP models analyzing sentence similarity, emotional conditional entropy, and emotional transition matrices.

Main Results:

• SZ patients exhibited less task-congruent sentiment polarity, failing to generate intense positive emotions for positive events and showing inappropriate positivity for negative events.
• SZ patients had significantly lower mean sentiment scores and emotional conditional entropy compared to HCs.
• SZ patients demonstrated higher neutral self-transition probability, indicating distinct abnormalities in emotional coherence, intensity, and dynamics.

Conclusions:

• A novel NLP-based computational strategy was developed to analyze linguistic patterns and emotional expression in SZ.
• Schizophrenia patients display distinct abnormalities in emotional coherence, expression intensity, and dynamics.
• These findings suggest an innovative approach for screening schizophrenic disorders.