Longitudinal Analysis of Early Semantic and Phonological Networks

• In Event: Poster Session 11
  In Poster Session: PS 11 Topic - Language, Communication

Abstract

Language consists of multiple levels of representation of meaning in the form of words (Mani & Borovsky, 2018). These words are organized in the lexicon as network structures, where different types of representations interact. A number of language processing models argue that both semantic and phonological representations are fundamental for the retrieval of meaning and word selection (Dell, 1986; McClelland & Elman, 1986). Experimental evidence also shows differential developmental trajectories for the presence of phonological and semantic links between words: toddlers develop phonological links at 18 months of age (Mani & Plunkett, 2011), but semantic links are developed at 24 months (Arias-Trejo & Plunkett, 2009). Network analyses based on parental reports have been used to explore the vocabulary structure of the early lexicon (Hills et al., 2009). Studies using normative data approaches have found that some properties of word acquisition can be explained by phonological, co-occurrence, feature, and associative relationships in the structure of vocabulary (Stella et al., 2017). However, it is still necessary to explore the development of vocabulary networks in individual toddlers. This study explored and compared the development of phonological and semantic vocabulary structure in a longitudinal design.
Vocabulary data from 21 toddlers was evaluated at 18, 21, and 24 months of age, using the McArthur Communicative Developmental Inventory (CDI; Jackson-Maldonado et al., 1993). A semantic and phonological network was created for each toddler, and a word was included in the network as a node if it was marked as produced in the CDI. In semantic networks, relationships between nodes were taken from feature norms (McRae et al., 2005), while in phonological networks these relationships were computed using the normalized Levenshtein similarity of the phonological transcription (Yujian & Bo, 2007). Three measures were calculated to analyze the network structure: density, clustering coefficient, and closeness centrality (Newman, 2010). A comparison among ages and between networks was performed using non-parametric statistics. The results showed that the semantic networks had higher density than the phonological networks at 24 (z = 2.72, p = 0.006) and 21 months (z = 2.94, p = 0.003). The semantic networks had higher clustering coefficients than the phonological networks at 24 (z = 3.84, p < 0.001), 21 (z = 3.28, p = 0.001), and 18 months (z = 2.025, p = 0.04). The phonological networks showed higher close centrality at 24 than at 18 months (z = 2.13, p = 0.03). Overall, the semantic networks had greater global and local access than the phonological networks, indicating that the early lexicon tends to organize word clusters using semantic information. This organization, which may facilitate the recovery of meaning using shared features of words, remains stable with age. In contrast, phonological networks tend to shorten the distance between nodes with age, perhaps to favor lexical selection as the lexicon grows.

Authors

• Armando Quetzalcóatl Angulo-Chavira, Universidad Nacional Autónoma de México (UNAM)
  Presenting Author

• Fernanda Chua, Universidad Nacional Autónoma de México (UNAM)
  Non-Presenting Author

• Daniela Avila, Georg-August University
  Non-Presenting Author

• Natalia Arias-Trejo, Universidad Nacional Autónoma de México (UNAM)
  Non-Presenting Author