Near-Zipfian Distribution is Prevalent in Infant Input
Skip to main content
eScholarship
Open Access Publications from the University of California

Near-Zipfian Distribution is Prevalent in Infant Input

Creative Commons 'BY' version 4.0 license
Abstract

Understanding infants' natural input is essential for advancing theories of cognitive development and learning. Recent research indicates that across modalities, infant input approximates a near-Zipfian distribution, with a large amount of input about a few items and substantially less about the rest. However, prior work has only examined aggregated distributions across subjects, focused on a single modality in isolation, and considered the input available to infants rather than what they actively select. We show that at both the corpus and individual levels, infant attention selection and the verbal input infants receive from parents follows a near-Zipfian distribution. Moreover, when integrating across modalities, the verbal input infants hear while attending to the same object becomes even more skewed than verbal input alone. Findings suggest that Zipfian-like structure is not only a property of infant environments but emerges through active selection, highlighting its potential role in shaping early learning.