Publication Date

Spring 2015

Document Type

Thesis

Degree Name

Master of Science

Department

Computer Science

First Advisor

(Clare) Xueqing Tang, Ph.D.

Second Advisor

Kong-Cheng Wong, Ph.D.

Third Advisor

Chris Tweddle, Ph.D.

Abstract

Several models have been proposed attempting to describe the mental lexicon-the abstract organization of words in the human mind. Numerous studies have shown that by representing the mental lexicon as a network, where nodes represent words and edges connect similar words using a metric based on some word feature, a small-world structure is formed. This property, pervasive in many real-world networks, implies processing efficiency and resiliency to node deletion within the system, explaining the need for such a robust network as the mental lexicon. However, each model considered a single word feature at a time, such as semantic or phonological information. Moreover, these studies modeled the mental lexicon as an unweighted graph. In this thesis, I expand upon these works by proposing a model that incorporates several word features into a weighted network. Analyses on this model applied to the English lexicon show that while this model does not exhibit the same small-world characteristics as a weighted graph, by setting a minimum threshold on the weights (reminiscent of action potential thresholds in neural networks), the resulting unweighted counterpart is a small-world network. These results suggest that a more integrated model of the mental lexicon can be adopted while affording the same computational benefits of a small-world network. An increased understanding of the structure of the mental lexicon can provide a stronger foundation for more accurate computational models of speech and text processing and word-learning.

Share

COinS