Semantic Change in Adults is Not Primarily a Generational Phenomenon

Gaurav Kamath, Michelle Yang, Siva Reddy, Morgan Sonderegger, Dallas Card

This repository contains code and data for the 2025 paper "Semantic Change in Adults is Not Primarily a Generational Phenomenon", published in Proceedings of the National Academy of Sciences (PNAS).

You can access the paper here: https://www.pnas.org/doi/10.1073/pnas.2426815122.

Contents

Below are the structure and contents of this repository.*

• a-optimization_results: Results of alpha-optimization for each sense studied of each word.
• assets: Extra files used for some of the data processing.
• clusters: Word-wise results of the replacement clustering algorithm, used to estimate word sense likelihood.
• datasets: As uploaded to this repository, only datasets containing GAMM predictions; but we recommend adding the preprocessed corpus files here if you plan to reproduce our experiments.
• plots: All plots generated as part of this study.
• scripts: Python and R scripts used to process the data, fit models, and generate plots.
• substitutions: The target word replacement data used to generate replacement clusters for word sense induction.
• collated_results.csv: CSV file containing the final results of our analysis by each word sense studied.
• meta-analysis.rds: RDS file that stores the Bayesian Meta-Analysis model we fit to estimate an average age effect.
• requirements.txt: Python libraries required for the Python scripts to run.
• r_env.yml: R libraries (using a Conda environment) required for the R scripts to run.

*Note that the original Stanford Congressional Record Dataset is not included in this repository, as it is far too large. The original data can be accessed here (download hein-bound.zip), and then processed using the Python scripts in our scripts folder. To do so, you will also need to download the legislators-current.yaml and legislators-historical.yaml files from here.

Furthermore, note that the following files (related to high-frequency words) are also not included in this repository on account of being too large:

• substitutions/call.jsonl
• substitutions/high.jsonl
• clusters/record_cluster_counts.csv
• substitutions/record.jsonl
• substitutions/security.jsonl