A sparse distributed memory model of overregularization.
Item
-
Title
-
A sparse distributed memory model of overregularization.
-
Identifier
-
AAI9417498
-
identifier
-
9417498
-
Creator
-
Nicols, Annemarie.
-
Contributor
-
Adviser: Martin Chodorow
-
Date
-
1994
-
Language
-
English
-
Publisher
-
City University of New York.
-
Subject
-
Psychology, Experimental | Computer Science
-
Abstract
-
As children acquire language they often make mistakes. One of the most notable of these errors is the overregularization (OR) of irregular past tense verbs since numerous studies have indicated that past tense acquisition is U-shaped (Cazden, 1968; Ervin, 1964; Kuzjac, 1977; Slobin, 1971). Children start out using the correct form of the verb, (e.g. broke, came) then go through a stage of overregularization, producing incorrect forms (e.g. breaked, comed) and finally, they gradually return to the use of the correct form. It may be argued that U-shaped learning and OR are general cognitive phenomena since a number of studies have shown that they occur when adults try to learn rule based systems with exceptions (Palermo & Eberhart, 1968; Palermo & Howe, 1970) and in other types of problem-solving tasks (Karmiloff-Smith & Inhelder, 1974/75).;Traditionally, OR and its U-shaped developmental sequence have been explained with a two process theory that involves both rote memory and rule use. In contrast, recent interest in connectionist theories of human memory have offered one process theories of OR which do not require the use of explicitly defined rules (MacWhinney & Leinbach, 1991; Plunkett & Marchman, 1991, 1993; Rumelhart & McClelland, 1986). The success of these connectionist models at simulating U-shaped learning and OR have led to recent modifications of the traditional theory (Pinker, 1991; Marcus et al., 1992).;In 1988 Kanerva introduced sparse distributed memory (SDM), a mathematical memory model, inspired by the biology of the human nervous system. This dissertation offers evidence that, although different from the connectionist models in a number of ways, SDM can account for OR and its U-shaped developmental sequence without built-in rules. This evidence includes: (1) a SDM simulation of Rumelhart and McClelland's (1986) "rule of 78", which is a simple demonstration of OR and U-shaped development, (2) a SDM simulation of Palermo and Howe's (1970) non-linguistic OR data, (3) a paired associates learning experiment with human subjects which was designed to investigate the effect of "friends" and "enemies" on OR and its learning curve, and lastly (4) a SDM simulation of this experiment.
-
Type
-
dissertation
-
Source
-
PQT Legacy CUNY.xlsx
-
degree
-
Ph.D.
