Study finds safety bias in everyday language
Researchers propose measuring word meaning through power, danger and structure rather than relying on emotional dimensions.
Language associated with safety occurs more often than language associated with danger, University of Vermont researchers found after examining billions of uses of more than 20,000 words. The study, published in Science Advances, identified the pattern across books, journalism, social media and speech.
The team developed an approach called ousiometrics to assess meaning mathematically. Its framework organizes words along three dimensions: power, danger and structure. According to the researchers, it accounted for more than 90% of variation in measured meaning, compared with approximately 72% for the established VAD model.
That older model describes words in terms of valence, arousal and dominance—dimensions concerning positive or negative feelings, excitement or calm, and control or submission. The researchers found that these dimensions were not independent, contrary to a longstanding assumption.
Rather than assigning equal weight to every word, the analysis took account of how frequently words appeared. The team said the results suggest that language's familiar tendency toward positivity may largely reflect a preference for safety. The findings do not establish that evolutionary pressures caused that preference.
The researchers also created an analytical tool, the ousiometer, to score meaning across large text collections and track changes within narratives. They suggest the approach could eventually improve how artificial intelligence interprets communication, including expressions involving trust and risk.
