The study, published in the journal Science Advances, introduces an innovative computational approach called "ousiometrics," which utilizes advanced mathematical methods to measure the essential meaning of language. Its most striking discovery is that the words humans use in daily communication show a consistent, widespread, and previously unrecognized preference for safety over danger.
This hidden safety bias could fundamentally transform how researchers study linguistics, analyze human behavior, and develop next-generation artificial intelligence systems designed to interpret what people say, write, and feel.
Scientists Discover a Hidden Safety Bias in Human Language
Since the 1950s, much of the scientific community’s understanding of word meaning has relied on a foundational framework developed through the pioneering research of psychologist Charles Osgood and his contemporaries. Known as the VAD model, it describes meaning through three core emotional dimensions: valence, which contrasts positive versus negative feelings; arousal, which contrasts excited versus calm states; and dominance, which contrasts controlling versus submissive positions.
This long-standing approach became deeply influential across multiple academic fields, including psychology, cognitive linguistics, and artificial intelligence, particularly within systems designed to identify the emotional tone of written text, known as sentiment analysis.
However, the University of Vermont researchers found compelling evidence that this decades-old model overlooks vital aspects of how semantic meaning is actually organized in human communication.
Their rigorous analysis examined billions of individual uses of more than 20,000 words drawn from a remarkably diverse range of real-world texts. Using sophisticated computational techniques, the research team identified an entirely different underlying structure. They discovered that the three dimensions traditionally assumed in the VAD model are not actually independent of one another.
Instead, their empirical results indicate that word meaning can be much more effectively and accurately described through three distinct dimensions: power, contrasting weak versus powerful; danger, contrasting safe versus dangerous; and structure, contrasting ordered versus chaotic.
The research, which was supported by the US National Science Foundation, Google, MassMutual, and several other funding organizations, revealed that this new framework accounted for more than 90 percent of the total variation in measured word meaning. By contrast, the traditional VAD approach explained approximately 72 percent.
That substantial improvement in explanatory power could prove especially critical as large language models, automated content moderation tools, and other artificial intelligence technologies increasingly rely on a nuanced understanding of human communication.
Why Human Language Favors Safety Over Danger
One of the study’s most unexpected breakthroughs emerged when the researchers investigated how words actually appear and function in everyday communication.
Across an enormous corpus of books, news reporting, social media feeds, and spoken language, words associated with safety appeared with dramatically greater frequency than those associated with danger.
This finding offers a fresh explanation for a familiar linguistic phenomenon known as the Pollyanna principle, which describes the well-documented tendency for human language to lean toward positive rather than negative words.
For decades, researchers generally interpreted this recurring pattern as direct evidence that people naturally favor positive emotional expressions. The new computational analysis, however, suggests that something much more fundamental may be driving the phenomenon.
According to the researchers, the apparent preference for positivity largely reflects an underlying, systemic preference for safety.
"The Pollyanna principle’s positivity bias," the study concludes, "is, in fact, a one-dimensional projection of an underlying safety bias."
Peter Dodds, director of the University of Vermont’s Complex Systems Institute and senior author of the study, emphasized the broad significance of the insight. "This is a big observation that comes out of this work," Dodds said. "Expressions of safety are crucial to all language."
Could Human Language Be Shaped by Survival?
The discovery raises an intriguing and profound possibility: the very structure of human language may reflect ancient evolutionary pressures associated with staying alive.
From this evolutionary perspective, words do much more than simply communicate transient feelings or emotional states. They also help people recognize external threats, evaluate environmental uncertainty, establish social trust, and coordinate complex actions with others.
Being able to reliably distinguish between safety and danger is essential to biological survival. The researchers suggest that this basic human priority may also heavily influence how meaning is structured and prioritized in language.
Throughout daily life, people constantly exchange vital information regarding whether individuals, physical places, shifting situations, and behaviors are threatening or safe. The new findings suggest that these distinctions may be far more central to human communication than previously recognized.
They also offer a markedly different interpretation of why positive language is so common across cultures. Rather than simply expressing personal happiness or social approval, many positive words may actually function to convey a profound sense of security, stability, or predictability.
Julia Zimmerman, a postdoctoral researcher in the University of Vermont’s Computational Story Lab and a co-author of the study, noted that the three newly identified dimensions reflect experiences that are fundamentally universal to human life.
"Power, danger, and structure," she said, "are relevant to every person that’s ever lived."
Researchers have also long recognized that people tend to favor language associated with goodness and low aggression. The research team argues that these familiar linguistic tendencies may ultimately arise from the exact same deeper pattern.
"We now understand," the team writes, that these familiar tendencies are merely "shadows of an underlying linguistic safety bias."
Although the results are consistent with a possible connection between language and survival, the authors emphasize that their study does not definitively establish that evolutionary pressures directly caused the safety bias.
A Discovery That Could Change How AI Understands Language
The implications of these findings extend well beyond theoretical linguistics, carrying direct consequences for modern technology.
Many artificial intelligence systems rely heavily on sentiment analysis to determine whether language expresses positive, negative, or neutral feelings. Some of these technological approaches also draw directly on the traditional VAD framework to measure emotional meaning in text.
If that traditional framework fails to fully capture the true underlying structure of language, systems built upon it could easily overlook critical distinctions in what people are actually communicating.
The researchers suggest that incorporating power, danger, and structure could eventually help artificial intelligence models interpret human language with greater accuracy and explain their assessments more clearly. This enhancement could prove particularly useful when analyzing complex scenarios involving human trust, financial risk, and strategic decision-making.
Within the field of linguistics, the findings offer an entirely different way to understand the foundational roots of meaning. Instead of treating emotional tone as the primary organizing principle, the new framework emphasizes distinctions related to strength, safety, and predictability.
Psychology could also be profoundly affected. Decades of behavioral studies have utilized the VAD model to investigate human emotions, perception, and action. A different set of fundamental dimensions could encourage researchers to reconsider how many of those historical findings are interpreted.
The discovery may likewise capture the attention of neuroscientists. The human brain is evolutionarily wired to be highly sensitive to environmental threats and signals of safety, and the newly identified linguistic bias suggests that these biological priorities are deeply mirrored in human language. This potential connection could help researchers explore complex relationships between brain activity and symbolic communication, although further empirical research would be required to establish direct links.
A New Tool Measures the Hidden Meaning of Words
To investigate these intricate patterns, the researchers developed sophisticated analytical methods designed to measure meaning across enormous collections of written and spoken language.
One of their central technical innovations is a new tool called the "ousiometer," which rapidly evaluates the essential meaning of large bodies of text and produces an aggregate average meaning score. The name is derived from the Ancient Greek word "ousia," which serves as a root for the English word "essence."
The ousiometer builds directly upon an earlier analytical tool developed by the research team called the "hedonometer," commonly known as a happiness meter. While that earlier instrument measures the overall emotional tone of language, the new ousiometer approach examines much broader patterns of core meaning.
The researchers demonstrated that the ousiometer can successfully analyze material drawn from remarkably diverse sources, including Jane Austen’s classic novels, Arthur Conan Doyle’s Sherlock Holmes stories, the New York Times, Wikipedia, Twitter, and transcripts of talk radio broadcasts.
The tool is also capable of tracking how meaning shifts dynamically as a narrative unfolds over time.
For example, the researchers analyzed an English translation of Victor Hugo’s epic novel Les Misérables and mapped its changing semantic meaning throughout the entire story.
They termed this changing pattern an "ousiometric trajectory." The resulting data visualization resembles the twisting shape of a multicolored protein, with the narrative moving across a multidimensional grid based on four contrasting pairs: dangerous versus safe, weak versus powerful, gentle versus aggressive, and bad versus good.
By following this analytical path, researchers can observe precisely how the underlying meaning of a text changes from one section of a story to another.
Why Counting Word Frequency Changes the Results
Another critical methodological innovation involved carefully distinguishing between words themselves as abstract categories and how frequently people actually use them in practice.
In linguistics, researchers describe individual words as categories known as "types," while each distinct appearance of a word in a text is referred to as a "token." For example, as a general category, the word "apple" is a type, whereas every time the word "apple" is actually deployed in a sentence, it counts as a token.
This distinction is vital because a word that appears millions of times contributes vastly more to everyday communication than one that is rarely used in conversation.
Earlier research in the field frequently gave each dictionary word equal analytical weight, regardless of its actual frequency in daily life. The Vermont team instead incorporated how often words appear in real-world communication, allowing vital patterns to emerge that might otherwise have remained completely hidden.
This frequency-aware approach was particularly instrumental in identifying the widespread, system-wide preference for safety-related language.
The ten-member research team was led by Peter Dodds and Chris Danforth, both professors in the University of Vermont’s College of Engineering and Mathematical Sciences. Their extensive collaboration included researchers from the Santa Fe Institute, the Complexity Science Hub in Austria, the Howard Hughes Medical Center, the University of California, Berkeley, the University of Adelaide, and MassMutual Data Science.
Rethinking What Human Language Reveals About Us
Understanding this hidden safety bias could have sweeping consequences for how researchers study communication across human society.
If words consistently reflect an underlying preference for safety, that psychological and linguistic tendency may significantly influence how stories are structured, how information spreads across populations, and how people interpret real-world events and personal experiences.
The findings could eventually inform research into political communication, public health and mental health messaging, and the design of artificial intelligence systems built to interact safely and effectively with humans.
More broadly, the study challenges the long-held assumption that human language can be understood primarily through basic emotional states such as happiness, fear, and excitement.
Instead, the researchers propose that semantic meaning possesses a much deeper structure intrinsically connected to power, danger, and order. Their findings suggest that human language may faithfully reflect the core challenges humans face when navigating uncertain environments, maintaining social relationships, and identifying potential threats.
By revealing this previously overlooked structure, the research team offers a fresh way to understand what words communicate and why certain kinds of meaning dominate everyday conversation. Human language, their findings suggest, may be far more closely connected to our fundamental need for safety than scientists have traditionally recognized.