Researchers identify a hidden safety bias in human language — ScienceDaily

Researchers at the University of Vermont have found that in everyday speech, words associated with safety are used significantly more often than words associated with danger. The team also proposed a new mathematical approach to analyzing linguistic meaning — usiometrics, ScienceDaily reports.

A new model of word meaning

The study results were published in the journal Science Advances. The scientists analyzed billions of instances of use of more than 20,000 words in books, news, social media and spoken language.

For a long time, psychology, linguistics and text-analysis systems used the VAD model, which describes word meaning through valence — positive or negative connotation, arousal and dominance. The authors concluded that these three parameters are not independent of one another.

Instead, the researchers propose describing meaning through three dimensions: power — from weakness to might, danger — from safety to threat, and structure — from order to chaos. According to their data, the new model explained more than 90% of the variation in measured word meanings, while the VAD model explained about 72%.

More current news is available on the UA.News Telegram channel Telegram.

Safety bias in language use

When analyzing actual word frequencies, the team found that in books, media, social networks and spoken language, words associated with safety occur significantly more often than words associated with danger.

This may offer a new explanation for the Pollyanna principle — the documented tendency of language to use positive rather than negative words more often. According to the authors, the tendency toward positivity is largely a manifestation of a deeper linguistic bias toward safety.

For the study, the team created a tool called the usiometer, which assesses the averaged underlying meaning of large text corpora. It can track changes in such meanings throughout a literary work or across different types of media materials.

Potential applications for AI

The authors believe the new approach may be useful for artificial intelligence systems that analyze text sentiment and assess risks, trust and decision-making processes. At the same time, the researchers emphasize that the findings are consistent with a possible influence of survival needs on the organization of language, but do not prove that evolutionary pressure itself caused the identified safety bias.