Natural language processing methods are sensitive to sub-clinical linguistic differences in schizophrenia spectrum...

Area of Science:

• Psycholinguistics
• Computational Linguistics
• Neuroscience

Background:

• Schizophrenia spectrum disorders (SSD) are characterized by speech disturbances.
• Objective and sensitive detection of these disturbances is crucial for diagnosis and monitoring.
• Computerized natural language processing (NLP) offers potential for analyzing speech patterns.

Purpose of the Study:

• To explore NLP methods for characterizing speech changes in individuals with SSD compared to healthy controls (HC).
• To compare NLP-derived linguistic features with a clinical standard (TLC).
• To investigate the utility of NLP in identifying subclinical language disturbances.

Main Methods:

• Employed Bidirectional Encoder Representations from Transformers (BERT) for linguistic analysis.
• Analyzed speech data on three levels: individual words, parts-of-speech (POS), and sentence-level coherence.
• Compared NLP features with the Scale for the Assessment of Thought, Language and Communication (TLC).

Main Results:

• Individuals with SSD used more pronouns but fewer adverbs, adjectives, and determiners compared to HC.
• SSD exhibited increased use of first-person singular pronouns and a striking increase in incomplete words.
• Sentence-level analysis with BERT indicated greater tangentiality in SSD, with larger sentence embedding distances.
• NLP measures discriminated between SSD and HC more effectively than clinical ratings alone, despite low average speech disturbance in the SSD sample.

Conclusions:

• NLP methods are sensitive to subclinical language disturbances in SSD.
• NLP-derived linguistic features may serve as clinically relevant biomarkers for SSD.
• Further research is warranted to explore NLP's potential in characterizing language disturbance in SSD.