AbstractsComputer Science

SEGMENTATION AND INTEGRATION IN TEXT COMPREHENSION: A MODEL OF CONCEPT NETWORK GROWTH

by Manas Sudhakar Hardas




Institution: Kent State University
Department: College of Arts and Sciences / Department of Computer Science
Degree: PhD
Year: 2012
Keywords: Artificial Intelligence; Cognitive Psychology; Computer Science; Experimental Psychology; Physiological Psychology; text comprehension; concept network; network analysis; Alzheimer's; Autism; clustering; giant component; phase transition; semantic network; semantic memory
Record ID: 1986576
Full text PDF: http://rave.ohiolink.edu/etdc/view?acc_num=kent1334593269


Abstract

Text comprehension (understanding by reading) is a fundamental way in which people will learn about the world around them. During text comprehension the reader segments the concepts from the text into recognized or unrecognized concepts and then integrates the recognized concepts into their present knowledge base which is represented as a concept network. Formation of this concept network is the central process in understanding from texts. In this thesis we present a mathematical framework for the segmentation (recognition) and integration of concepts. The model can explain how and why different readers construct different concept networks on reading the same text. It can also describe why some readers may understand a text easily as compared to other readers, and also why some texts are difficult to understand than other texts for the same reader. The model is also used to explain the effect of the age of acquisition of a concept on comprehension. It is seen that earlier a concept is acquired the more important it is for comprehension of other concepts. The model leads to an algorithm which is used to simulate concept network growth during text comprehension. These networks are then analyzed to investigate their structural properties. It is seen that these networks are small worlds with high local clustering and a normal degree distribution. These properties are indicative of the high connectivity and reachability not observed in similar random networks. It is also seen that although concept networks may start off with multiple disconnected components, the process of comprehension leads to most of the nodes getting connected to form a single giant component.