BabyBabelLM: A Multilingual Benchmark of Developmentally Plausible Training Data

2025-10-11

AI Summary

This paper addresses the critical lack of developmental plausibility and cognitive interpretability in multilingual pretraining. We introduce the first developmentally interpretable, multilingual benchmark grounded in empirical child language acquisition (CLA) trajectories. Methodologically, we integrate principles from developmental psychology with multilingual data compilation techniques to construct a staged linguistic input dataset covering 45 languages—each scaled to the informational equivalence of 100 million English word tokens—and modeling natural language exposure from infancy to native proficiency. We accompany this resource with a standardized evaluation suite and baseline models. Our contributions are threefold: (1) the first cross-lingual, developmentally grounded pretraining data construction paradigm; (2) a reproducible resource for cognitive modeling and neurolinguistic validation of multilingual models; and (3) empirical evidence that our data significantly improves model performance on cross-lingual transfer and cognitive alignment tasks.

Technology Category

Natural Language Processing: Machine Translation, Multilinguality, Cross-Lingual NLPMachine Learning: Large Multimodal Models (LMMs)Cognitive Modeling & Cognitive Systems: Computational Creativity

Application Category

Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web searchWeb Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web dataSemantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

We present BabyBabelLM, a multilingual collection of datasets modeling the language a person observes from birth until they acquire a native language. We curate developmentally plausible pretraining data aiming to cover the equivalent of 100M English words of content in each of 45 languages. We compile evaluation suites and train baseline models in each language. BabyBabelLM aims to facilitate multilingual pretraining and cognitive modeling.

Problem

• Models language acquisition from birth data
• Provides multilingual developmentally plausible training datasets
• Facilitates cognitive modeling and multilingual pretraining

Innovation

• Multilingual developmentally plausible pretraining data curation
• 45 languages with 100M English word equivalents
• Evaluation suites and baseline models for cognitive modeling