A connectionist model of human short-term memory is presented that extends the "phonological loop" (A. D. Baddeley, 1986) to encompass serial order and learning. Psychological and neuropsychological data motivate separate layers of lexical, timing, and input and output phonemic information. Connection weights between layers show Hebbian learning and decay over short and long time scales. At recall, the timing signal is rerun, phonemic information feeds back from output to input, and lexical nodes compete to be selected. The selected node then receives decaying inhibition. The model provides an explanatory mechanism for the phonological loop and for the effects of serial position, presentation modality, lexicality, grouping, and Hebb repetition. It makes new psychological and neuropsychological predictions and is a starting point for understanding the role of the phonological loop in vocabulary acquisition and for interpreting data from functional neuroimaging.Temporally modulated information-processing and sequential behavior are essential to people's everyday functioning, such as in interpreting auditory stimuli and in using language. In a classic article, Lashley (1951) argued that serially ordered behavior cannot be explained in terms of "associative chaining," whereby each element in a sequence is associatively linked to its neighbors. However, subsequent attempts to model serial order computationally have continued to use associative chaining (e.g., Lewandowsky & Murdock, 1989), and there remains a paucity of adequate alternative accounts of people's ability to recognize, remember, reproduce, and become familiar with intricately timed and rhythmic sequences such as those in speech.In contrast, there has been substantial progress in identifying the cognitive subsystems mediating the immediate serial recall of sequences of verbal items. There is a large body of evidence from psychological experiments suggesting that this form of memory is mediated by a "phonological loop" comprising a speech input store and a control process of subvocal articulation (Baddeley, 1986;Baddeley & Hitch, 1974). However, this simple model is incomplete. For example, it does not specify the processing of serial order information. This is a serious omission given the inherent
The episodic buffer component of working memory is assumed to play a role in the binding of features into chunks. A series of experiments compared memory for arrays of colors or shapes with memory for bound combinations of these features. Demanding concurrent verbal tasks were used to investigate the role of general attentional processes, producing load effects that were no greater on memory for feature combinations than for the features themselves. However, the binding condition was significantly less accurate with sequential rather than simultaneous presentation, especially for items earlier in the sequence. The findings are interpreted as evidence of a relatively automatic but fragile visual feature binding mechanism in working memory. Implications for the concept of an episodic buffer are discussed.
The authors summarize developments in the concept of working memory as a multicomponent system, beginning by contrasting this approach with alternative uses of the term working memory. According to a 3-component model, working memory comprises a phonological loop for manipulating and storing speech-based information and a visuospatial sketchpad that performs a similar function for visual and spatial information. Both are supervised by a central executive, which functions as an attentional control system. A simple trace-decay model of the phonological loop provides a coherent account of the effects of word length, phonemic similarity, irrelevant speech, and articulatory suppression in verbal short-term memory tasks. This model of the loop has also proved useful in the analysis of neuropsychological, developmental and, cross-cultural data. The notion of the sketchpad is supported by selective interference with imagery in normal adults and by specific neuropsychological impairment. Analysis of the central executive is illustrated by work on deficits in the ability to coordinate subproccesses in Alzheimer’s disease.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.