ABSTRACT
Introduction
Overregularization (OR) errors and U-shaped learning during past tense acquisition have been used to support the model of a symbolic, two process system of learning requiring both rule use and memorization. Beginning with the work of Rumelhart and McClelland (1986), connectionist networks have attempted to model these phenomena with varying degrees of success. We examine the ability of Kanerva's (1988) sparse distributed memory (SDM) to simulate Palermo and Howe's (1970; hereafter P&H) experimental analogy to past tense acquisition. SDM, like the connectionist networks, does not explicitly encode rules, therefore, if it is able to simulate P&H's behavioral data, the plausibility of a sub-symbolic, one process account will be strengthened.
