Scientists analyzed billions of words and found human language is not mainly organized around emotion, but around one survival instinct
Since the 1950s, scientists have described the meaning of words mostly in terms of feeling: is a word pleasant or unpleasant, exciting or calm, dominant or submissive. That framework still sits inside much of today’s psychology and AI. A team at the University of Vermont went back to the question with billions of real word uses, from Jane Austen novels and Sherlock Holmes stories to the New York Times, Wikipedia, Twitter and talk radio. Emotion turned out not to be the main thing.
The researchers found that the meaning of more than 20,000 words is better described by three different dimensions: power (weak or powerful), danger (safe or dangerous) and structure (orderly or chaotic). This new framework accounted for more than 90% of the variation in measured word meaning, compared with about 72% for the old emotion-based model.
And one pattern stood out above the rest. In the way people actually use words, safe words show up far more often than dangerous ones. Across books, news, social media and speech, human language leans strongly toward safety.
“This is a big observation that comes out of this work,” senior author Peter Dodds, director of the University of Vermont’s Complex Systems Institute, said in a statement. “Expressions of safety are crucial to all language.”
That reframes an old idea. In 1969, psychologists proposed the Pollyanna principle: people use positive words more than negative ones. In 2015, a study in PNAS by the same Vermont group confirmed it across 10 languages and billions of words, finding a universal positivity bias in human language. The new paper argues that this positivity is a side effect of something deeper. In the researchers’ words, the positivity bias “is, in fact, a one-dimensional projection of an underlying safety bias.”
The key methodological change was counting how often each word is actually used. Earlier studies usually gave every word in a dictionary equal weight. Weighting words by their real-world frequency revealed the tilt toward safety that a simple word list hides.
“Power, danger, and structure are relevant to every person that’s ever lived,” said co-author Julia Zimmerman, a postdoctoral researcher in the university’s Computational Story Lab.
The second finding is a tool. The team built what they call an ousiometer, from an Ancient Greek root for essence, which measures the average meaning of any body of text along these dimensions and can track how it shifts as a story unfolds. They used it to map the changing meaning of Les Misérables from section to section.
The researchers suggest that a language built around power, danger and order may reflect what has always mattered most for survival: knowing who is strong, what is dangerous and what is predictable. They are careful to say their results are consistent with that idea but do not prove it. The approach could also help AI systems that judge the tone of text, which often still rely on the older emotion-based model. Language and art can also shift over time, as an analysis of 38,467 song lyrics recently showed.
There are limits to what this shows. The texts described in the press materials are in English, and it is not yet clear how far the safety bias extends to other languages. The evolutionary explanation is a hypothesis, not a finding. The study was supported by the US National Science Foundation, Google, MassMutual and other funders.
Source
Dodds PS, Alshaabi T, Fudolig MI, et al. “Ousiometrics: The essence of meaning aligns with a power-danger-structure framework instead of valence-arousal-dominance.” Science Advances, 2026.
DOI: 10.1126/sciadv.adr4039
Also cited
Dodds PS, Clark EM, Desu S, et al. “Human language reveals a universal positivity bias.” Proceedings of the National Academy of Sciences, 2015.
DOI: 10.1073/pnas.1411678112
