Modeling Early Lexico-Semantic Network Development: Perceptual Features Matter Most

• In Event: 1-100 - The Role of Semantic Networks in Early Word-Learning

Integrative Statement

This project explores how the composition of productive vocabulary between 16 to 30 months can be predicted by aspects of early word meaning (encoded in semantic features), and by connections between word meanings (in developing lexico-semantic networks).

Adult lexico-semantic systems flexibly encode multiple types of semantic features, including: functional features that describe ways people interact with objects, perceptual features that refer to information accessible through the senses, and taxonomic features that describe relations in a categorical hierarchy. However, less is understood regarding whether children encode all of these kinds of semantic properties in their early lexico-semantic networks. Some theories argue that perceptual features play a primary role in early lexico-semantic development, while others emphasize functional relations (Rakison & Oakes, 2003). We approach this issue by exploring whether and how these different feature types contribute to variability in the order of early word learning.

Using a database of semantic features of early-acquired nouns (Peters, McRae, & Borovsky, in prep), we constructed graph-theoretic noun-feature networks that represent links between words per 4 feature types (i.e. different networks with perceptual, functional, taxonomic and encyclopedic connections), with nodes representing nouns and links between nodes representing shared features. The early-acquired nouns consisted of 359 nouns appearing on the words and sentences form of the MacArthur-Bates Communicative Developmental Inventory (Fenson et al., 2007). We estimated age-of-acquisition (AoA) for the nouns using a publicly-available dataset of word-level vocabulary data in 16- to 30-month-old children (Wordbank; Frank et al., 2016). In study 1, we asked whether semantic properties of early-acquired words relate to the order in which they are typically learned, and study 2 modeled normative lexico-semantic noun-feature network development compared to random network growth.

Study 1 explored how the features comprising individual words related to AoA. Words with greater numbers of perceptual (β=-0.26, p<.001) and taxonomic (β=-0.16, p<.01) features were learned earlier than those with fewer of these same features. However, after controlling for word frequency in Child-Directed Speech, only perceptual features contributed significant variance in AoA (β=-0.10, p<.05).

Study 2 asked how different featural connections between words contributed to normative growth of early lexico-semantic networks. We compared normative lexical growth networks where words were added in order of AOA, to metrics characterizing random lexical growth networks where words were selected randomly. Results (Fig.1) indicated normative networks contain significantly more perceptual and taxonomic lexical-links between words and have significantly higher density than would be expected to happen randomly. Furthermore, perceptual networks show significantly shorter distances between words than would be expected to happen randomly.

These experiments provide converging evidence that perceptual and, to a lesser extent, taxonomic properties of word meanings, associate with early lexical development. Broadly, the findings suggest that perceptual features and linkages among words may support early word learning, and that concepts with more perceptual features are easier to learn relative to those that have fewer such features. This finding lends support to accounts that posit that perceptual features of objects support early word learning, and that more complex/abstract linkages among concepts develop later.

Authors

• Ryan E. Peters, Purdue University
  Presenting Author

• Arielle Borovsky, Purdue University
  Non-Presenting Author