Language or Syndrome? Investigating the Origins of the General Psychopathology Factor through Large Language Model Embeddings

This study examines whether transdiagnostic factors, including the general psychopathology factor (p-factor), emerge primarily from linguistic properties or reflect underlying syndromic co-occurrence. By comparing text embeddings with self-reported responses from a sample of 985 individuals, both approaches yielded comparable clustering patterns and general factor structures. However, self-reported responses exhibited stronger associations with participants’ mental health histories, indicating a greater capacity to capture clinically relevant features. These findings highlight the potential for large language models (LLMs) to advance psychopathology assessment while emphasizing the continued importance of traditional diagnostic approaches.