Study: ChatGPT organizes word associations differently from human speakers
A study in Language and Cognition compared word association networks for the category la casa (the house) from 50 native Spanish-speaking undergraduates, advanced learners of Spanish, and 50 separate runs of ChatGPT (GPT-5.2). The two human groups had very similar global network structures despite different learning histories. The model's network was more modular, had longer paths, converged more on the same words and used less concrete vocabulary.
This is a neat example of how network analysis can test whether an AI system organizes knowledge the way people do. Producing the right words is not the same as organizing them like a human mind. The study blends graph theory, psycholinguistics and statistics, showing how interdisciplinary the field has become.
Can language models teach us about human cognition, or are they too different to compare?

0 Comments
Sign in to join the discussion
No comments yet. Be the first to comment!