This repository accompanies the paper:
Baby Scale: Investigating Models Trained on Individual Children's Language Input
Feng et al. (2026) Paper: https://arxiv.org/abs/2603.29522
The code and data for this original study have not been publicly released.
Our subsequent paper, Specialization, Diversity, and Scaling in Child-Scale Language Models, extends this research with new experiments examining specialization to individual children's language environments, training-data diversity, and scaling across naturalistic and synthetic language input.
The follow-up study uses an updated and improved experimental pipeline and evaluation methodology.
Code and reproducibility materials for the follow-up paper will be available in a separate repository soon.