Working Memory and Language Processing ALAN ADDELEY University of Bristol In his classic work Principals of Psychology, William James (1890) proposed to distinguish between what he termed primary and secondary memory. The former was assumed to reflect material held in a temporary form so as to be readily accessible to conscious awareness, while the secondary system was assumed to reflect a much more durable system for the long-term storage of information. Many years later, Donald Hebb (1949) speculated that the tempo rary system might reflect the electrical activity of the brain, while the more durable seat of long-term memory was probably based on more permanent neurochemical links, a view that has been revived in recent years and imple mented in a range of neural network models, for which the term Hebbian's Learning represents one possible representation of the process of acquisition. At an empirical level, long term learning formed an active area of research throughout this century (Bartlett 1932; McGeoch and Irion 1952; Crowder 1976), whereas the more temporary or short-term system received compara tively little attention until the late 1950's, when Brown (1958) in England and Peterson and Peterson (1959) in the US first reported that even small amounts of information would show marked forgetting over a matter of seconds if the subject were prevented from active rehearsal by a simple task such as counting backwards. They proposed that their results reflected the operation of a short-term memory store (STM) which differed from long-term memory (LTM) in being based on a memory trace that would spontaneously fade within a matter of seconds unless maintained by rehearsal. The 1960's saw a period of intense controversy, with some arguing that the new results were entirely compatible with existing unitary theories of memory(e.g. Melton 1963), while others argued for a two component view, while accepting that any given experimental task may well show the simultaneous evidence of both components (Waugh and Norman 1965). By the late 1960s, the evidence seemed to be favouring a dichotomous view; many models were proposed, but the most influential and probably most characteristic was that of Atkinson and Shiffrin (1968), which consequently became known as the modal model. This proposed that information passed through a parallel series of sensory memory systerns that are essentially part of the processes of percep tion, before entering a limited capacity short-term store from which informa tion faded unless rehearsed. The store was capable of encoding and elaborating the information, and was responsible for feeding it into and out of the long term store. The probability of transferring an item from short to long-term storage was assumed to be a simple function of how long the item stayed in the short-term system. Probably the strongest evidence for a dichotomous view came from the study of patients with neuropsychological deficits. The classic amnesic syndrome is one in which patients appear to have lost the capacity to lay down new memory traces although they may retain the capacity to recall memories from the time before their amnesia. Such patients however typically have preserved immediate memory span as reflected in their capacity to repeat back a string of numbers such as a telephone number, or, provided they are intellec tually otherwise intact, they may well be able to perform normally on the Brown-Peterson Short Term Forgetting Task mentioned earlier (Baddeley and Warrington 1970). They appear in short, to have impaired LTM but preserved STM. The second type of patient was reported by Shallice and Warrington (1970), showing the opposite pattern of memory performance. The patients suffered from what would probably have previously been categorised as conduction aphasia, but showed a pattern of behaviour that was consistent with the hypothesis of a very specific STM deficit, with extremely rapid forgetting on the Brown-Peterson task, cou pled with a digit span limited to one or two items, while at the same time showing apparently normal LTM performance. Although the Atkinson and Shiffrin model initially appeared to give a good account of the data, problems rapidly began to appear. One concerned the assumption that maintaining items in STM was enough to guarantee learning. Evidence failed to support this (e.g. Bjork and Whitten 1972) resulting in an influential paper by Craik and Lockhart (1972) proposing their levels of proc essing hypothesis, whereby the probability of long term retention of an item was a function not of the frequency of rehearsal, but of the depth at which it was processed. Hence if a word were processed only in terms on its visual appearance, a very short duration trace would result, whereas processing the sound of the visually presented word would lead to somewhat more durable learning, while the deeper process of encoding the meaning and relating it to existing knowledge, would produce an even more durable trace.
