With recent advances in large language models (LLMs), the concept of automatically generating children’s educational materials has become increasingly realistic. Working toward the goal of age-appropriate simplicity in generated educational texts, we first examine the ability of several popular LLMs to generate stories with properly adjusted lexical and readability levels. We find that, in spite of the growing capabilities of LLMs, they do not yet possess the ability to limit their vocabulary to levels appropriate for younger age groups. As a second experiment, we explore the ability of state-of-the-art lexical simplification models to generalize to the domain of children’s stories and, thus, create an efficient pipeline for their automatic generation. In order to test these models, we develop a dataset of child-directed lexical simplification instances, with examples taken from the LLM-generated stories in our first experiment. We find that, while the strongest-performing current lexical simplification models do not perform as well on material designed for children due to their reliance on large language models behind the scenes, some models that still achieve fairly strong results on general data can mimic or even improve their performance on children-directed data with proper fine-tuning, which we conduct using our newly created child-directed simplification dataset.

Understanding child language development requires accurately representing children’s lexicons. However, much of the past work modeling children’s vocabulary development has utilized adult-based measures. The present investigation asks whether using corpora that captures the language input of young children more accurately represents children’s vocabulary knowledge. We present a newly-created toddler corpus that incorporates transcripts of child-directed conversations, the text of picture books written for preschoolers, and dialog from G-rated movies to approximate the language input a North American preschooler might hear. We evaluate the utility of the new corpus for modeling children’s vocabulary development by building and analyzing different semantic network models and comparing them to norms based on vocabulary norms for toddlers in this age range. More specifically, the relations between words in our semantic networks were derived from skip-gram neural networks (Word2Vec) trained on our toddler corpus or on Google news. Results revealed that the models built from the toddler corpus were more accurate at predicting toddler vocabulary growth than the adult-based corpus. These results speak to the importance of selecting a corpus that matches the population of interest.