Language-based AI modeling of personality traits and pathology from life narrative interviews

Area of Science:

• Computational linguistics
• Psychological assessment
• Artificial intelligence in behavioral science

Background:

• Personality and personality pathology models require refinement.
• Natural language processing (NLP) offers novel approaches to personality assessment.
• Existing methods for personality assessment can be time-consuming and subjective.

Purpose of the Study:

• To model personality and personality pathology using natural language from life narrative interviews.
• To develop and validate AI-driven language models for personality assessment.
• To explore the utility of AI in refining personality disorder (PD) frameworks.

Main Methods:

• Trained and tested language models (RoBERTa, BERTopic, Linguistic Inquiry and Word Count) on transcribed life narratives from 1,409 older adults.
• Validated language models against established personality measures (NEO-PI-R, Structured Interview for DSM-IV Personality) and multimethod criteria.
• Utilized fine-tuning of RoBERTa parameters for personality prediction.

Main Results:

• Fine-tuned RoBERTa models predicted personality scores with correlations above r = .40, indicating large effect sizes.
• Life narrative language mapped more effectively to the Five-Factor Model domains than to DSM PD categories.
• Moderate support was found for mapping language to borderline personality pathology.

Conclusions:

• Language-based AI holds significant potential for refining conceptual frameworks of personality and personality disorders.
• AI-driven personality assessment can provide automatic prediction in research and clinical settings.
• Multimethod validation supports the use of language models for personality assessment.