Poster #42 - Using Distributional Semantics to Understand Preschoolers' Depth of Vocabulary

• In Event: Poster Session 07

Abstract

Word knowledge networks, also known as vocabulary depth, have an explanatory role in understanding and processing human language (Jones & Mewhort, 2007; Ouellette, 2006; Tannenbaum et al., 2006). Vocabulary depth is conceptualized as the interconnected networks of semantic knowledge supporting word labels, with relevant concepts linked together by shared semantic relationships (Anderson & Freebody, 1985). Network science, a technique using graph theory to investigate complex systems like social networks, helps understand how knowledge is structured in the mental lexicon (Börner et al., 2007). Word knowledge may be represented as semantic networks employing network science algorithms, with each node representing a word in the vocabulary and each connection between two nodes representing a link between the corresponding pair of words (Hills et al., 2009; Steyvers & Tenenbaum, 2005). This study suggests that there are generic rules governing the structure of network representations for natural language semantics and that these structural rules provide substantial implications for adult-child language interaction analyses.
In particular, given that adults talk differently in different activity contexts and each context has affordances and constraints for fostering language and literacy, the following research questions are addressed: (1) Does adults’ use of semantically related conversation vary across activity settings (book reading, toy play, and mealtime)?; (2) Do children show a higher degree of language development, when adults engage in more semantically related conversation within each activity context?
This study used the Home-School Study of Language and Literacy Development (HSLLD) Corpus (Dickinson & Tabors, 2001) which studied the social prerequisites to language development of a group of racially diverse, English-speaking children from low-income families. In this research, 83 children were visited in their home at age 3 and transcripts were collected in the home for book reading, toy play, mealtime, narrative, and experimental tasks.
Using semantic network and word embedding algorithms, I explored (1) static features of words’ connectivity in networks in book reading, toy play, and mealtime and (2) the relation between word learning and semantically related conversation within each activity setting. Four statistical features were used to describe the structure of semantic networks: the average distance L, the diameter D, the sparsity S, and the clustering coefficient C. A type-token ratio (TTR) is a metric used to determine whether a child employs a variety of different words to communicate.
This study found the following: (1) The density for the semantic networks in toy play was significantly greater than mealtime and book reading, indicating that the semantic networks in toy play included fewer concepts than mealtime and book reading. (2) The clustering coefficient for the semantic networks is book reading was significantly greater than toy play and mealtime, indicating that the networks in book reading included word communities or groups of words that were densely connected internally (i.e., semantically related conversation). (3) Adults’ use of semantically related conversation within book reading related to children’s vocabulary development. Potential implications of these findings for research and practice will be discussed.

Authors

• Wonkyung Jang, The University of Oklahoma
  Presenting Author

• Boo Young Lim, The University of Oklahoma