BabyReasoningBench: Generating Developmentally-Inspired Reasoning Tasks for Evaluating Baby Language Models
Kaustubh D. Dhole
TL;DR
The paper addresses the mismatch between adult-centric benchmarks and the reasoning abilities of baby language models trained on developmentally plausible input. It introduces BabyReasoningBench, a 19-task, developmentally grounded suite drawn from classic developmental psychology paradigms, generated via GPT-5.2 and assessed on two BabyLM baselines trained on child-directed corpora. Findings show that while scaling pretraining data meaningfully improves several causal and physical reasoning tasks, explicit belief attribution and pragmatics-sensitive reasoning remain challenging, revealing dissociable patterns across task families. The work offers a mechanism-sensitive diagnostic framework for evaluating what reasoning capabilities can emerge from child-directed input distributions and motivates future multimodal and intervention-based extensions.
Abstract
Traditional evaluations of reasoning capabilities of language models are dominated by adult-centric benchmarks that presuppose broad world knowledge, complex instruction following, and mature pragmatic competence. These assumptions are mismatched to baby language models trained on developmentally plausible input such as child-directed speech and early-childhood narratives, and they obscure which reasoning abilities (if any) emerge under such constraints. We introduce BabyReasoningBench, a GPT-5.2 generated benchmark of 19 reasoning tasks grounded in classic paradigms from developmental psychology, spanning theory of mind, analogical and relational reasoning, causal inference and intervention selection, and core reasoning primitives that are known to be confounded by memory and pragmatics. We find that two GPT-2 based baby language models (pretrained on 10M and 100M of child-directed speech text) show overall low but uneven performance, with dissociations across task families: scaling improves several causal and physical reasoning tasks, while belief attribution and pragmatics-sensitive tasks remain challenging. BabyReasoningBench provides a developmentally grounded lens for analyzing what kinds of reasoning are supported by child-like training distributions, and for testing mechanistic hypotheses about how such abilities emerge.