A new study challenges the widely accepted idea that word meanings are organized around emotion. After analyzing billions of words, scientists found that language may be shaped by something more basic: the need for safety.
Researchers at the University of Vermont have found a new way to understand language, challenging a major assumption in psychology, linguistics, and artificial intelligence that has guided research for more than 70 years.
Their study, published in Science Advances, presents “ousiometrics,” a quantitative approach to studying essential meaning. The work suggests that language is not organized mainly around emotion, but around a deeper pattern shaped by power, danger, and order.
The central finding is striking: across language, humans consistently lean toward safety.
A Hidden Bias in Language
For decades, many researchers have described meaning through three emotional dimensions: valence (positive vs. negative), arousal (excited vs. calm), and dominance (controlling vs. submissive), a model known as the VAD framework. The approach grew from influential work in the 1950s by Charles Osgood and others, and it has been widely used in psychology, linguistics, and artificial intelligence systems that analyze sentiment.
The new analysis, based on billions of uses of more than 20,000 words across varied real-world texts, shows that this long-used framework has major weaknesses. With support from the US National Science Foundation, Google, MassMutual, and other funders, the researchers used modern computational methods to identify a different set of basic meaning dimensions. They found that the VAD dimensions are not truly independent and can hide a more fundamental structure in language.
The researchers argue that meaning is better captured by three independent dimensions: power (weak vs. powerful), danger (safe vs. dangerous), and structure (ordered vs. chaotic).
University of Vermont researchers developed an “ousiometer,” a tool for measuring meaning in large texts, and identified three key dimensions of meaning: power, danger, and structure. Using Les Misérables as an example, they showed how a story’s language shifts across these dimensions over time, demonstrating the tool’s ability to map meaning in large-scale texts. Credit: University of Vermont
This matters now because language technologies are increasingly shaping how people communicate, from large language models to automated content moderation. Understanding the structure of meaning has become more urgent. The new framework explains more than 90% of the variation in meaning, compared with about 72% for the traditional VAD model.
When the researchers studied word use across books, news, social media, and spoken language, one pattern appeared again and again. Language strongly favors words linked with safety over words linked with danger.
This safety bias offers a new interpretation of the Pollyanna principle, a long-known idea in linguistics that language tends to skew positive. The new work suggests that the pattern is not simply about positive emotion. Instead, it reflects a deeper bias toward safety. “The Pollyanna principle’s positivity bias,” the study concludes, “is, in fact, a one-dimensional projection of an underlying safety bias.”
“This is a big observation that comes out of this work,” said Peter Dodds, director of the UVM’s Complex Systems Institute and senior author of the study. “Expressions of safety are crucial to all language.”
Beyond Positivity: Language as a Survival System
The findings carry broad implications. If language is tilted toward safety, then communication may have been shaped by evolutionary pressures connected to survival. Words do more than express emotions. They help people judge risk, identify threats, and coordinate behavior when the world is uncertain.
This view helps explain why people so often communicate whether something feels safe or dangerous. Across cultures and situations, humans regularly signal the risk level of places, actions, people, and events. The study suggests that this safety dimension is not secondary to emotion. It may be one of the foundations of meaning.
From this perspective, positivity in language is not only about happiness, approval, or optimism. It can also signal predictability and safety in a shared environment. Julia Zimmerman, a postdoctoral researcher in UVM’s Computational Story Lab and study coauthor, says the framework points to a basic feature of human experience. “Power, danger, and structure,” she said, “are relevant to every person that’s ever lived.”
Linguists have also long noted that language favors expressions of goodness and low aggression. “We now understand,” the team writes, that these are “shadows of an underlying linguistic safety bias.”
Rethinking Meaning Across Disciplines
The results challenge assumptions in several fields.
For artificial intelligence, the implications are direct. Many natural language processing systems depend on sentiment analysis shaped by frameworks similar to VAD. If those models do not capture the deeper structure of meaning, AI systems may be misreading human language in systematic ways. Adding power, danger, and structure to these systems could make them more accurate and easier to interpret, especially in tasks involving risk, trust, and decisions.
For linguistics, the study changes how researchers might think about the basic organization of meaning. Instead of treating emotional tone as the main structure behind words, the work points to survival-related distinctions, including what is powerful, what is dangerous, and what is orderly.
For psychology, the findings raise questions about decades of research built around the VAD model. If the core dimensions of meaning are different from what many researchers assumed, then some interpretations of emotion, perception, and behavior may need to be reconsidered.
For neurobiology, the results connect with what is already known about the brain’s strong sensitivity to threat and safety. A safety bias in language may reflect biological priorities in symbolic communication, helping link neural processes with the way humans use words.
A New Scientific Framework: Ousiometrics
To identify these patterns, the researchers built new tools for measuring meaning at large scale. One key tool is the “ousiometer,” an instrument designed to quickly measure the essential meaning of large bodies of text and produce an average meaning score. (The word “ouisa” comes from Ancient Greek and is a root for the English word “essence.”) Building on the team’s earlier “hedonometer” (a happiness meter), the new tool can detect broad meaning patterns in texts ranging from Jane Austen novels and Arthur Conan Doyle’s Sherlock Holmes stories to the New York Times, Wikipedia, talk radio transcripts, and Twitter.
One example in the study follows the “ousiometric trajectory” of an English translation of Victor Hugo’s Les Misérables. Like a multicolored protein, the book’s tangled path winds its way over a grid defined by four opposing pairs: dangerous and safe, weak and powerful, gentle and aggressive, and bad and good. This approach condenses the essential meaning of different sections of the novel as the story unfolds.
The study also makes an important distinction between words as categories, known as “types,” and words as they are actually used, known as “tokens.” (For example, as a category, “apple” is a type, and every time the word “apple” is used in a sentence is a token.) Earlier work often treated words as if they mattered equally, no matter how often they appeared.
By accounting for frequency of use, the 10 scientist team, led by Peter Dodds and Chris Danforth, professors in UVM’s College of Engineering and Mathematical Sciences, along with colleagues from the Santa Fe Institute, the Complexity Science Hub in Austria, Howard Hughes Medical Center, University of California, Berkeley, University of Adelaide, and MassMutual Data Science, was able to uncover patterns that appear only in real language use, including the safety bias.
Why This Matters Now
If language consistently leans toward safety, the finding may affect how researchers understand the spread of information, the building of narratives, and the way people interpret the world. It could matter for political discourse, mental health communication, and the design of AI systems that respond to human language.
More broadly, the study suggests a shift in how meaning should be understood. Meaning is not only a matter of emotion or sentiment. It is also rooted in the need to navigate risks, relationships, and social order. By revealing a deeper geometry of meaning, the team offers a new way to see language, not only as a system of symbols, but as a record of what humans need to survive in a social and dangerous world.
Reference: “Ousiometrics: The essence of meaning aligns with a power-danger-structure framework instead of valence-arousal-dominance” by Peter Sheridan Dodds, Thayer Alshaabi, Mikaela Irene Fudolig, Julia Witte Zimmerman, Juniper Lovato, Shawn Beaulieu, Joshua R. Minot, Michael V. Arnold, Andrew J. Reagan and Christopher M. Danforth, 6 May 2026, Science Advances.
DOI: 10.1126/sciadv.adr4039
Funding: Provided by the Vermont Advanced Computing Center, which was supported in part by NSF awards 1827314 and 2117345 (P.S.D. and C.M.D.); foundational support from MassMutual (P.S.D., J.L., and C.M.D.); National Science Foundation award no. 242829 (Science of Online Corpora, Knowledge, and Stories) (P.S.D., J.L., and C.M.D.); Google Open Source under the Open-Source Complex Ecosystems And Networks (OCEAN) project (J.L.); the Alfred P. Sloan Foundation (G-2024-22498) (J.L.); and an anonymous philanthropic gift (P.S.D.).
Never miss a breakthrough: Join the SciTechDaily newsletter.
Follow us on Google and Google News.
Full article can be found at: https://scitechdaily.com/scientists-challenge-a-70-year-old-theory-of-language-with-a-surprising-discovery/
