Modeling personality language use with small semantic vector subspaces

Abstract

In this proof of concept, we aim to expand the use of semantic vector subspaces by proposing a cost-effective method for generating observable semantic indicators that capture relevant variability of the Big Five model from language. Basically, we illustrate the formalization of the semantics of personality language use of the Big Five, by measuring general and specific semantic meanings through the coordinates of such subspaces. Two studies were conducted to validate the resulting semantic vector subspaces in two different prompted-based self-descriptions using standardized self-report inventories as criteria (N=643 Spanish native speakers). In the first study, we found convergent and discriminant validity of different latent profiles of personality language use. In the second study, different linear relations were observed between the language indicators and the personality traits showing the differences between the profiles found in the first study. Our findings partially support the well-known lexical hypothesis of personality. Results suggest that these semantic vector subspaces can be used to extract personality trait-relevant semantic properties from language, which has methodological and theoretical implications for the study of language and personality relations, and personality screening in texts.