A hidden bias in human language challenges 70 years of science

A University of Vermont study proposes that word meaning is organized less by the emotional dimensions used in psychology for decades and more by power, danger, and structure: whether something is weak or powerful, safe or dangerous, and ordered or chaotic. Researchers analyzed billions of real-world uses of more than 20,000 words, accounting for how often each word actually appears. Their framework explained over 90% of variation in measured meaning, compared with about 72% for the traditional valence-arousal-dominance model, whose dimensions the researchers found were not independent.

The most striking result was a broad preference for safety-related language across books, news, social media, and speech. The team argues that the familiar tendency toward positive words, known as the Pollyanna principle, may largely reflect a deeper bias toward safety rather than positivity alone. They suggest this pattern could relate to survival needs such as recognizing threats, building trust, and coordinating action, but the findings do not prove that evolution caused it. The researchers also developed “ousiometrics” and an ousiometer to measure meaning in large texts and track how it changes through a narrative. The approach could reshape research in linguistics and psychology and help AI systems better interpret risk, trust, and other distinctions that sentiment analysis may miss.