Researchers at the University of Vermont have unveiled a significant finding that challenges longstanding theories in linguistics regarding the nature of word meanings. Their study, published in Science Advances, introduces the concept of “ousiometrics” and suggests that meanings of words may be more fundamentally influenced by concepts of power, danger, and order rather than merely emotional factors.
The research indicates that human language reflects a consistent preference for safety over danger, which the authors argue can reshape how language is studied, human behavior understood, and artificial intelligence developed. This hidden bias toward safety suggests that language is crucial in communicating not just emotions, but also in recognizing threats and establishing trust in social interactions.
The study builds on a framework established in the 1950s known as the VAD model, which categorizes word meanings based on three emotional dimensions: valence (positive or negative), arousal (calm or excited), and dominance (controlling or submissive). However, the Vermont team found that this model does not account for the interdependency among these dimensions and overlooks the significant influence of safety-related language.
Utilizing advanced computational methodologies, the researchers analyzed billions of word usages from over 20,000 terms across diverse texts. Their analysis revealed a different framework for understanding word meanings, characterized by three separate dimensions: power, danger, and structure. This new method explains over 90% of the variation in word meaning, surpassing the 72% explained by the traditional VAD model.
The findings revealed that words associated with safety are significantly more prevalent in everyday communication than those linked to danger. This observation offers a novel interpretation of the Pollyanna principle, suggesting that the historical bias toward positive language may in fact stem from an underlying safety bias. Peter Dodds, the study’s senior author, emphasized that expressions relating to safety are central to all language.
The implications of this discovery extend beyond linguistics and into artificial intelligence. Many AI systems currently rely on sentiment analysis based on the emotional framework established by the VAD model. The Vermont researchers argue that incorporating dimensions of power, danger, and structure could enhance AI’s ability to interpret language more accurately, particularly in contexts involving trust and risk.
To conduct their research, the team introduced the ousiometer, a new analytical tool designed to evaluate the fundamental meanings of large text bodies. This tool provides an average meaning score and can track changes in meaning throughout narratives, revealing how the context influences word interpretations.
The implications of this research may lead to a reevaluation of how language correlates with human behavior, potentially influencing studies in communication, mental health messaging, and AI interactions. The findings suggest that the structure of human language may be more reflective of evolutionary survival mechanisms than previously acknowledged.
The study was conducted by a ten-member team led by Professors Peter Dodds and Chris Danforth from UVM, alongside collaborators from various institutions including the Santa Fe Institute and the University of California, Berkeley. As research in this area continues, a deeper understanding of these linguistic structures could reshape not only language studies but also broader implications in technology and societal communication practices.
