The Content, Structure, and Evolution of English Trait Words

Traits are the currency of personality psychology and social cognition, helping us understand and represent both real and perceived differences between people and groups. Here, we advance our understanding of trait language by revealing new insights into the content, structure, and evolution of trait words using large language models (LLM) and human raters. Across four studies, we: (a) introduce one of the most comprehensive trait lists to-date (2,847 trait words); (b) establish normative human ratings of these traits (from N = 3,070 English speakers) along 24 dimensions from past literature; (c) detect a four-factor structure of Fitness, Agency, Communion, and Traditionalism (FACT) from these dimensions; and (d) reveal systematic differences in the coherence, valence, and historical evolution of these dimensions and factors. Specifically, we find that the FACT model covers 96% of trait words. Moreover, these four factors differ in their coherence (e.g., Communion is the most semantic coherent while Traditionalism the least), valence (e.g., Communion and Agency show more alignment between their semantic direction and valence than Traditionalism and Fitness; Communion has more negative words while Agency and Traditionalism more positive ones); and historical usage (e.g., Fitness and Communion are older than Traditionalism; Communion is converging in semantic space while Agency diverging). These findings showcase new LLM-based methods for discovering and analyzing trait words. We close by motivating a functional constructivist view in which trait concepts are tools for social evaluation, shaped by the recurrent social interaction problems they help solve, and organized accordingly in semantic structure and use.