Estimating the Semantic Network of Vocabulary Created by Teacher Talk and Text Reading in a Fifth-Grade Humanities Class

• In Event: 34.077 - Moving Toward a Multicomponent Model of Vocabulary Learning in School

Abstract

Purpose
The long-term goal of this work is to better understand and model individual differences in children’s ability to add academic vocabulary into existing semantic networks as they learn about a topic. A first step in understanding how children learn vocabulary is to better understand the frequency with which the teacher and class text use the target words. To estimate the “semantic network of instruction,” we recorded every lesson by the teacher and collected the text read by students during a yearlong unit on a topic. The semantic networks of the target vocabulary words used in instruction (i.e., teacher transcripts, texts) are created via Neural Network Language Modeling (similar to Latent Semantic Analysis). The strength of the word relationships in these networks is used to predict word-learning difficulty using measures administered to children at the conclusion of the instructional unit.

Perspective

Theories of vocabulary acquisition rely on the assumption that the acquisition of semantic knowledge is a gradual process in which learners accumulate knowledge about words through multiple exposures. However, little is currently known about how children add academic vocabulary to their sematic networks through instruction and how much variation exists among the network connections that are formed. This study provides an opportunity to model the semantic network of instruction, which can lead to understanding models of individual children’s semantic networks in the response to instruction.

Data Sources and Methods

Data consists of the digitized transcriptions of recordings of a teacher’s talk and digitized copies of the text over a unit of instruction. Twenty-five target vocabulary words were selected as representing important academic vocabulary. Word2vec was used to process teacher and text data and to create the semantic network of each target word. Word2vec takes as its input a large corpus of text and produces a high-dimensional space (typically of several hundred dimensions), with each unique word in the corpus assigned a corresponding vector in the space. Word vectors are positioned in the space so that words that share common contexts in the corpus are located in close proximity to one another. Word vectors can be configured such that each node in a semantic network represents a word with weighted connections between nodes that indicate the degree of similarity (i.e., connectedness). This allows sub-networks around a target word to be generated with the size regulated by altering the network threshold for connection strength. The long-term question is whether the networks that are supported by classroom experience resemble the networks that are generated by children as they experience instruction and if significant individual differences exist in network building across children.

Significance

Modeling the instructional semantic network as a means of understanding the conceptual environment in which children are embedded is a first step in understanding how children learn and integrate new vocabulary into their existing sematic networks. This study should provide a rich explanation of the network connections promoted by instruction and lead to understanding how these connections support or deter the addition of target words to children’s semantic networks.

Authors

• Amy Elleman, Middle Tennessee State University

• Donald L. Compton

• Laura Steacy, Vanderbilt University

• Jeff Elmore, MetaMetrics

• Jill Fitzgerald, University of North Carolina - Chapel Hill