Researchers at the University of Vermont have identified a widespread safety bias in human language, challenging a scientific framework that has shaped research for more than 70 years. Their analysis suggests that word meaning is organized not only by emotion, but also by three deeper dimensions: power, danger and structure. The findings could influence linguistics, psychology and artificial intelligence systems that interpret human communication.
A new model for understanding word meaning
Since the 1950s, researchers have often used the VAD model, developed through pioneering work by psychologist Charles Osgood and others, to describe the emotional meaning of words. VAD measures three dimensions: valence, or positive versus negative; arousal, or excited versus calm; and dominance, or controlled versus submissive.
The model has influenced psychology, linguistics and artificial intelligence, including systems that analyze the emotional tone of written language. However, University of Vermont researchers say it does not fully capture how meaning is constructed.
The team examined billions of individual uses of more than 20,000 words across a range of real-world texts. Using computational analysis, they found that the three dimensions of the traditional VAD model are not independent of one another, as previously assumed.
Instead, the researchers found that word meaning can be described more effectively through three separate dimensions: power, ranging from weak to powerful; danger, ranging from safe to dangerous; and structure, ranging from orderly to chaotic.
The study was supported by the National Science Foundation, Google, MassMutual and other funders. The new framework accounted for more than 90% of the variation in measured word meanings, compared with approximately 72% for the traditional VAD model.
The difference could be important as large language models, automated content-moderation systems and other AI technologies increasingly rely on interpreting human communication.
Why safety-related language is so common
One of the study’s most unexpected findings emerged from an examination of how words are used in everyday communication. Words associated with safety appeared much more frequently than words associated with danger in books, news reports, social media, and spoken-word material.
The finding offers a possible explanation for the Pollyanna principle, the well-known tendency for human language to contain more positive than negative words. Researchers have commonly interpreted that pattern as evidence that people prefer expressing positive emotions. The new analysis suggests that it may instead reflect a more fundamental orientation toward safety.
According to the researchers, apparent positivity in language is largely a reflection of an underlying safety bias.
“Pollyanna principle positivity bias,” the study concludes. “In fact, this is a one-dimensional projection of an underlying safety bias.”
“This is a big observation from this study,” said Peter Dodds, director of UVM’s Complex Systems Laboratory and senior author of the study. “Representing safety is critical to any language.”
Could language reflect survival priorities?
The findings raise the possibility that the structure of human language reflects evolutionary pressures associated with survival. From this perspective, words do more than express emotion: they help people recognize threats, assess uncertainty, establish trust and coordinate actions.
Distinguishing between safety and danger is essential to survival. The researchers suggest that this basic priority may also influence how meaning is organized in language.
People constantly exchange information about whether individuals, places, situations and actions are threatening or safe. The study suggests that these distinctions may be more central to communication than previously recognized.
This interpretation also offers a different explanation for the prevalence of positive words. Rather than simply expressing happiness or approval, many positive words may communicate security, stability or predictability.
Julia Zimmerman, a postdoctoral fellow in UVM’s Computational Stories Institute and a co-author of the study, said the three identified dimensions reflect experiences fundamental to human life.
“Power, risk, and structure are things that concern every person who has ever lived,” she said.
Researchers have also long observed that people tend to favor words associated with goodness and reduced aggression. The Vermont team argues that these patterns may arise from the same deeper tendency.
“We now understand that these are ‘shadows of an underlying linguistic safety bias,’” the researchers wrote.
The result is consistent with a possible connection between language and survival, but it does not prove that evolutionary pressures directly caused the safety bias.
Implications for artificial intelligence
The findings could have implications well beyond linguistics. Many AI systems use sentiment analysis to determine whether language is positive, negative or neutral. Some also use traditional VAD frameworks to measure emotional meaning.
If a framework does not fully represent the underlying structure of a language, systems built on it may overlook important distinctions in what people are communicating.
The researchers suggest that incorporating power, danger and structure could eventually help AI models interpret language more accurately and explain their assessments more clearly. Such an approach could be especially useful when systems analyze trust, risk and decision-making.
In linguistics, the framework provides another way to study the basis of meaning. Instead of treating emotional tone as the main organizing principle, it emphasizes distinctions related to strength, security and predictability.
The findings could also affect psychology. Decades of research have used VAD models to investigate emotion, perception and behavior. If the underlying dimensions differ from those assumed by the model, researchers may need to reconsider how some previous results are interpreted.
The work may interest neuroscientists as well. The human brain is highly sensitive to signals of threat and safety, and the language patterns identified in the study may reflect those biological priorities. This possible association could help researchers examine relationships between brain activity and symbolic communication, although further research is needed to establish such connections.
Measuring the hidden meaning of words
To investigate these patterns, the researchers developed an analytical method for measuring meaning across large collections of written and spoken language. They describe the approach as “ausiometrics,” using mathematical methods to assess the essential meaning of language.
One of the approach’s central tools is the “osciometer,” which rapidly evaluates the essential meaning of large bodies of text and produces an average semantic score. The source describes the term “weisa” as coming from ancient Greek and as the origin of the English word “essence.”
The tool builds on an earlier instrument developed by the research team called the “hedonometer,” or “happiness meter.” The hedonometer measures the overall emotional tone of language, while the newer methods examine broader patterns of meaning.
The researchers have used the tools to analyze material from varied sources, including Jane Austen’s novels, Arthur Conan Doyle’s Sherlock Holmes stories, The New York Times, Wikipedia, Twitter and talk-radio transcripts.
The method can also track how meaning changes as a story unfolds. For example, the researchers analyzed the English translation of Victor Hugo’s Les Misérables and mapped changes in meaning throughout the novel.
They called the resulting pattern an “ausiometric trajectory.” The visualization resembles the twisted forms of multicolored proteins. It maps a story across a grid based on four contrasting pairs: dangerous and safe, weak and powerful, kind and aggressive, and bad and good.
Following the trajectory allows researchers to observe how the underlying meaning of a text changes from one section of a story to another.
Why word frequency matters
Another important part of the research was distinguishing between words themselves and how frequently they are used. In linguistics, individual words are classified as “types,” while each occurrence of a word in a text is called a “token.”
For example, “apple” is a type as a category of word, while every use of “apple” in a sentence is a token.
The distinction matters because words that appear millions of times contribute more to everyday communication than words that are rarely used. Previous studies have often treated every word as equally important, regardless of frequency.
The Vermont team instead incorporated how often words appear in real-world communication. This allowed the researchers to identify patterns that could otherwise remain hidden, including the widespread preference for safety-related language.
A broader view of what language reveals
The 10-member research team was led by Peter Dodds and Chris Danforth, professors in UVM’s College of Engineering and Mathematical Sciences. Collaborators included researchers from the Santa Fe Institute, the Austrian Complexity Science Hub, Howard Hughes Medical Center, the University of California, Berkeley, the University of Adelaide and MassMutual Data Science.
A consistent preference for safety could influence how stories are told, how information is distributed and how people interpret events and experiences. The findings could ultimately inform research into political communication, mental-health messaging and AI systems designed to interact with people.
More broadly, the study challenges the idea that human language can be understood primarily through emotions such as happiness, fear and excitement. The researchers propose that language has a deeper structure tied to power, danger and order.
The findings suggest that language may reflect the challenges people face when navigating uncertain environments, maintaining relationships and identifying threats. By highlighting this previously overlooked structure, the research offers a new way to examine what words convey and why particular meanings dominate everyday speech.
The results suggest that human language may be more closely connected to the basic need for safety than scientists previously realized, although the evolutionary explanation remains a possibility rather than a proven cause.
Source: www.sciencedaily.com