Study Finds 6 AI Models Favor Latin- and French-Origin Words Over Germanic Alternatives
Updated
Updated · The Washington Post · Jun 2
Study Finds 6 AI Models Favor Latin- and French-Origin Words Over Germanic Alternatives
1 articles · Updated · The Washington Post · Jun 2
Six AI models tested by Florida State University researchers consistently preferred Latin- and French-derived words over Germanic alternatives, the study found, with the skew stronger than in typical English usage.
That bias appears to emerge during preference learning, where human annotators may subconsciously reward language that sounds more formal, confident or educated.
The finding offers one explanation for chatbot prose that leans on words such as “meticulous” and “commendable,” producing a polished but slightly uncanny style.
Researchers argue the bigger risk is not Romance vocabulary itself but systems optimized for trust signals over accuracy, potentially making weak answers sound authoritative.
Can AI's deep-rooted language bias ever be truly fixed, or is it a permanent feature of its design?
Is AI’s sophisticated vocabulary an illusion of intelligence, and are we falling for the trick?
How AI’s Romance-Language Bias Is Reshaping English: FSU Study Quantifies 23% Shift Since 2022
Overview
A new study from Florida State University reveals that AI models like ChatGPT show a clear preference for words from Romance languages, such as French and Spanish. This AI-driven lexical bias means that certain sophisticated-sounding words are used more often by AI, and these 'ChatGPT buzzwords' are now appearing in everyday speech. The FSU team discovered this pattern through careful analysis, showing that AI is not just reflecting language but actively shaping it. As a result, AI's vocabulary choices are influencing how people communicate, highlighting the growing impact of artificial intelligence on our language.