Representing the Toddler Lexicon: Do the Corpus and Semantics Matter?

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 3960–3968 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 3960 Representing the Toddler Lexicon: Do the Corpus and Semantics Matter? Jennifer M. Weber, Eliana Colunga Department of Psychology and Neuroscience 345 UCB Boulder, CO 80309 USA {jennifer.m.ellis, eliana.colunga}@colorado.edu Abstract 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. Keywords: Word2Vec, corpora, semantic networks, word-learning 1. Introduction In the field of first language learning, there has been an increasing interest in modeling lexical development as a way to both understand and predict language development. There is enormous variability in the vocabularies of toddlers in their second year of life, which makes it difficult to tease out possible patterns in vocabulary acquisition and resulting lexical knowledge. To model vocabulary growth, and to predict future growth, we must accurately represent children’s current word knowledge and how the words children know relate to each other. Most previous research attempting to model young children’s lexicons has relied on adult metrics to create semantic networks. But the knowledge adults have about words may not be an accurate proxy for child knowledge. Further, advances in natural language processing techniques and network science provide the opportunity to create richer models than have previously been investigated. Thus, the main research question of this paper is whether the language samples from which we derive lexical representations matter for accurately modeling toddler vocabulary growth. Of additional interest is whether different distributional models of semantics provide richer representations that lead to more accurately capturing child development. To accomplish our goals, we will build and analyze networks representing the relationships between the words a typical English-speaking, American child knows at different ages between 16 and 30 months, investigating whether networks created using a toddler corpus more accurately represent and predict language growth compared to those created from adult corpora. 1.1 Representing a Lexicon Networks are often used to model vocabularies. In these …

Content truncated to 500 words