Mapping the Positive Self-Bias Embedded in Human Languages

Whether humans inherently view themselves more positively than others underpins a fundamental question of psychology, yet it has eluded scientific consensus for decades. Overcoming behavioral research’s limitations, advancements in natural language processing facilitate exploring more nuanced societal-level biases through text analysis. This study employed word embeddings to investigate the positive self-bias in human languages and its universality across cultures and historical periods using multilingual corpora totaling approximately two trillion words. We first demonstrated the presence of positive self-bias in English, aligning with prior behavioral findings. Cross-cultural comparisons then revealed self-deprecation tendencies in trait-based self-evaluation in specific Eastern languages, contrasting with Western patterns. However, affective self-positivity—the tendency to associate the self with generally positive rather than negative words—was universally observed across 11 languages. Analysis of 200 years of English texts further substantiated this dichotomy: while trait self-positivity fluctuated historically, affective self-positivity remained stable. These findings support an integrative dual-dimensional model of positive self-bias, illuminating a pathway to reconcile the longstanding debates.