Hidden bias in human language calls into question 70 years of science

• The study revealed hidden bias in language in favor of safety
• The new structure of word meanings is based on power, danger, and order
• The results could change approaches to language analysis and AI.

Hidden bias in language

A study conducted by researchers from the University of Vermont revealed unexpected patterns in human language that challenge traditional scientific theories that have existed for over 70 years. In particular, the study shows that the meaning of words is shaped not only by emotions but also by more fundamental concepts such as power, danger, and order. These findings could significantly change approaches to studying language and understanding human behavior.

As part of the study, a new methodology called "usimetry" was proposed, which uses mathematical methods to measure the underlying meaning of language. One of the most striking discoveries was that the words people use demonstrate a consistent bias in favor of safety. This finding could change not only linguistics but also approaches to developing artificial intelligence systems that interpret human speech and text.

Since the 1950s, the scientific understanding of the meaning of words has been based on the VAD model developed by psychologist Charles Osgood. This model describes meaning through three emotional categories: valence, arousal, and dominance. However, a study from the University of Vermont showed that this model does not account for important aspects of meaning organization, which calls its universality into question.

New discoveries in linguistics

The analysis conducted by researchers covered billions of usages of over 20,000 words in various texts. Using modern computational methods, the team found that the traditional three dimensions of the VAD model are not independent. Instead, they proposed a new structure based on three dimensions: power, danger, and structure, which allows for a more accurate description of word meanings.

The new model proved to be significantly more effective than the traditional one: it explained over 90% of the variations in word meanings, while the VAD model managed only 72%. This discovery has important implications for artificial intelligence systems that increasingly rely on understanding human communication.

One of the most unexpected discoveries was that words related to safety occur in everyday communication much more often than words related to danger. This finding offers a new explanation for the well-known phenomenon called the Pollyanna principle, which describes the tendency of language towards positive expressions.

Influence on artificial intelligence

Researchers argue that the bias towards positivity in language may be a result of a deeper bias towards safety. "The bias of positive language, as our research shows, is a projection of an underlying bias towards safety," notes one of the study's authors, Peter Dodds.

This discovery raises an interesting question: the structure of human language may reflect evolutionary pressures related to survival. From this perspective, words not only convey feelings but also help people recognize threats, assess uncertainty, and establish trust.

Researchers suggest that the ability to distinguish between safety and danger is key to survival. In everyday life, people constantly exchange information about whether individuals, places, and situations are threatening or safe.

Perspectives and uncertainties

New findings show that these distinctions may be more central to communication than previously thought. They also offer another explanation for why positive language is so prevalent: many positive words may convey a sense of safety and stability.

The research team also notes that the habit of using language associated with kindness and low aggression may be linked to the same underlying bias towards safety. "Now we understand that this is just the shadows of a hidden bias in language towards safety," they write.

Although the research findings align with a possible connection between language and survival, they do not establish a direct link between evolutionary pressures and bias towards safety. These discoveries may have implications that extend beyond linguistics, particularly in the field of artificial intelligence.

Source: ScienceDaily