ABSTRACT
In recent years there has been a considerable growth of interest in distributed memory models, and a variety of models have been developed. Particular instantiations of distributed models differ in specific and interesting ways, but as a class they share a number of important a priori advantages over discrete or localized storage models: (a) As there is only one memory store, no memory search process is required. (b) Memory is reconstructive (redintegrative); information can be retrieved even when the input is noisy or incomplete. (c) Any distributed memory system has the ability to store information about individual items and information about associations between items in the same fashion. This ability also provides one way of accounting for memory for serial order (Lewandowsky & Murdock, 1989). Thus distributed memory models can provide a parsimonious account of the storage and retrieval of item, associative, and serial order information. (d) As storage is distributed, local disturbance or damage can be sustained without complete loss or failure of the system (Wood, 1978). Thus memory fails gradually, not suddenly. (e) Finally, distributed memory models offer a neurologically plausible description of memory (Anderson, Silverstein, Ritz, & Jones, 1977; Squire, 1987).
