Investigating the connection between free word association and demographics

Free word association (FWA) has been used to analyze cultures, thoughts, beliefs, and demographics across various fields. FWAs are a widely used scientific tool to quickly view a subject's beliefs, biases, and opinions that are often repressed or difficult to detect in short interviews or surveys. Here, we explored the relationship between FWA and demographics through neural network analysis. We hypothesized that neural network analysis of FWA could accurately predict a participant's age, gender, first language, and current country based on their FWA responses to a random cue word from a set of 12,292 cues selected from prior FWA studies. Using the "Small World of Words'' dataset containing over 1.2 million FWAs, we created a prediction model and evaluated for accuracy across the four demographic variables. The study employed an existing linguistic neural network, Large Language Model Meta AI 2 (LLaMA 2), which was fine-tuned to predict demographics from FWAs. The trained model demonstrated noteworthy accuracy predicting first language (63.6%), current country (58.4%), and age (median distance of nine years from predicted to actual age), but demonstrated a fluctuating accuracy across generation parameters when predicting gender. Our findings suggest a correlation between FWAs and demographics, aligning with previous research on FWA reflecting geographical differences, cultural beliefs, and age-related patterns. The study demonstrates the potential of using FWA and neural networks to identify demographic information more efficiently than other large scale data collection methods such as surveys.