Analyzing Roman Personality Traits Using Large Language Models

The preprint "Roman Personality" attempts to answer an interdisciplinary question using computational methods: did ancient Romans exhibit a stable personality structure similar to the modern "Big Five" in their actual writings? The study does not rely on modern questionnaires but directly reconstructs how "personality" was described, categorized, and distinguished from historical texts.

Methodology and Data

According to @MohammadAtari90, the research team compiled 8,032 texts from 88 BCE to 800 CE and extracted 3,516 character descriptions. Methodologically, they first used Large Language Models (LLMs) to identify sentences genuinely describing specific individuals, and then applied topic modeling to observe which qualities repeatedly clustered around the same person, thereby converting classical literature into systematically comparable personality trait data.

Key Findings

The study not only examined whether aggregatable personality dimensions exist in Roman texts but also compared descriptive differences among authors from various traditions. Results indicate that even when describing the same types of figures, Christian authors more frequently emphasized traits such as piety, holiness, asceticism, and humility. Pagan authors, conversely, highlighted courage, justice, severity, loyalty, and clusters related to vices. The researchers suggest this difference reflects varying understandings of which personality dimensions are most important, rather than merely being a byproduct of describing different subjects.

Additional Information

Related posts mention that the authors provided a preprint link, an interactive data visualization website, and researcher profiles, allowing the public to further verify the methods and results.

2026-07-14 ~ 2026-07-14 · 8 related posts