Research Challenges 70-Year Assumptions About Human Language

UVM Study Finds Safety Bias in Language

The new UVM analysis shows safety-related words dominate language, appearing far more often than danger words across billions of uses. Published in Science Advances on Oct 24 2026, the work introduces 'ousiometrics' to map meaning. Instead of the old valence-arousal-dominance model, they identified power, danger and structure as the three core axes. Those three axes explain over ninety percent of word-meaning variation, versus roughly seventy-two percent for the VAD model. The study was funded by the NSF, Google, MassMutual and other partners. The surprise is that this safety bias rewrites the Pollyanna principle, showing positivity is really a proxy for security. Peter Dodds summed it up: 'Expressions of safety are crucial to all language.'