Language-based assessments can predict psychological and subjective well-being.

Area of Science:

• Computational Linguistics
• Psychology
• Artificial Intelligence

Background:

• Well-being encompasses subjective (pleasant life) and psychological (meaningful life) aspects.
• Natural Language Processing (NLP) advances automated subjective well-being assessment.
• NLP's utility for psychological well-being assessment remains largely unexplored.

Purpose of the Study:

• To evaluate language-based model performance in predicting subjective and psychological well-being.
• To compare different NLP models (BERT, GPT-3.5, GPT-4) for well-being prediction.
• To analyze linguistic patterns associated with different well-being constructs.

Main Methods:

• Utilized contextual word embeddings from transformer-based NLP models.
• Trained models to predict self-reported well-being scores from verbal/written responses.
• Compared model predictions against standard questionnaire measures.

Main Results:

• Language models showed moderate correlations with well-being measures (rs = .16–.63).
• GPT-3.5 and GPT-4 outperformed BERT in predicting life satisfaction (r = .71, .75).
• Autonomy was less predictable than life satisfaction, with distinct linguistic patterns.

Conclusions:

• NLP tools effectively assess hedonic well-being (life satisfaction) but face challenges with eudaimonic constructs (autonomy).
• Further research needed to improve NLP for complex psychological states, balancing accuracy and interpretability.
• Advanced language models offer potential for nuanced well-being assessment.