Working Memory ALAN BADDELEY ‘The term working memory refers to a brain system that provides temporary storage and manipulation of the information necessary for such complex cognitive tasks 2s, Janguage comprehension, learning, and reasoning. This definition has evolved from the concept of a unitary short-term memory system, Working memory has been found to require the simultaneous storage and processing, of information. It can be divided into the following three subcomponents: (i) the central executive, which is as- sumed to be an attentional-controlling system, is impor- tant in skills such as chess playing and is particularly susceptible to the effects of Alzheimer’s disease; and two slave systems, namely (ii) the visuospatial sketch pad, which manipulates visual images and (iii) the phonologi- cal loop, which stores and rehearses speech-based infor- ‘mation and is necessary for the acquisition of both native and second-language vocabulary. HE QUESTION OF WHETHER MEMORY SHOULD BE REGARD: ced as a single unitary system or whether it should be fractionated into two oF more subsystems formed one of the major controversies within cognitive psychology during the mid- 1960s. During that time, evidence began to accumulate in favor of a dichotomy (1). Some of the most convincing evidence care from the study of brain-damaged patients; those suffering from the cassie amnesic syndrome appeared to have gross disruption ofthe capacity to form new lasting memories but showed preserved performance ‘ona range of tasks that were assumed to test short-term memory (2) Conversely, a second type of patient was identified who appeared 10 show normal long-term learning but kad a shore-term memory span limited co one or ewo items (3). Je was suggested that such patents had a deficit in shore-term storage, in contrast to the long-term storage deficit thar occurs in the amnesic syndrome. This finding, together with considerable evidence from the study of normal subjects, appeared by the late 1960s t0 argue for a dichotomous view of memory, such a that proposed by Atkinson and Shifrin (4) By the ealy 1970s ie was becoming clear tha che two-component model was running into difficulties. One of its problems was inherent in che neuropsyehological evidence chae initially appeared 1 support itso strongly. Atkinson and Shifrin (4) suggested that the short-term store within their model acted as a workin memory, being necessary for lexrning, forthe retrieval of old material, and for the performance of many other cognitive tasks. If chat were the ease, fone wculd expect patients with 2 grossly defective shore-term store to show many other cognitive problems, inchiding impaired long- rerm learning. In fet, such patients appeared to have a normal Fhe author the ect of ie Medal Recah Cou, Appl Pocology Un. Cambridge C83 388, Unt Kino * . 86 long-term learning capacity and surprisingly few cognitive handi- caps. Pursuing this issue was difficult because patients with a pure shoreterm memory deficit are rare. We therefore attempted 10 simulate this condition in unimpaired subjects by using a dual-task technique (5). We argued as follows: if the digit-span procedure depends on the short-term store, with the number of digits retained determined by the capacity of the stor, then it should be possible 0 inerfere systematically with the operation of the working memory system by requiring the subject to remember digits while perform- ing other cognitive tasks. As the concurrent digit ioad is increased, the remaining shor-term capacity would decrease and the interfer- ‘ence would increase, with performance presumably breaking down as the digit load reached the capacity of the system. Reasoning, comprehension, and learning casks all showed a similar pattern. As concurren’ digit load increased, performance declined, but che degree of disruption fell far shore ofthat predicted. Subjects whose digit memory was at full capacity could reason and lean quite effectively. ‘These results, rogether with others, encouraged the abandonment of the idea of single unitary short-term store that also functions as 1 working memory. Instead, we proposed the tripartite system shown in Fig. 1, which comprises an attentional controller and the central executive, supplemented by two subsidiary slave systems. ‘The articulatory or phonological loop was assumed to be responsi- ble for maintaining speech-based information, including digits in ‘he digit span test, whereas the visuospatial sketch pad was assumed 0 perform a similar function in setting up and manipulating visuospatial imagery. ‘The concept of working memory has increasingly replaced the older concept of short-term memory (6). Research has subsequently tended to concentrate on one of two complementary but somewhat dierent approaches. One of these defines working memory as the system tha is necessary forthe concurrent storage and manipulation of information; asks are devised that combine processing and. storage, and the capacity of such tasks to predict a range af other cognitive skile, sch as reading, comprehension, and reasoning, is rested. This psychometric approach, which has flourished most srongly in Norh America, frequently focuses on the extent to. which performance on working memory tasks can predict individual diferences in the relevant cognitive skills. An alternative approach, which has been more favored in Europe, uses both dual-task methodology and the study of neuropsy: ‘chological cases in an artempe to analyze the structure of the working memory system, Most effors has been devored t0 the nwo: slave systems, on the grounds that these offer more tractable problems than the more complex cenrral-executive system. ‘The two appreaches are complementary, and both have strengths and weaknesses; the psychometric correlational approach has the advantage that it can tackle what is probably the most crucial ‘component of the system, the central executive, and can furthermore work directly on problems of practical significance, such as readingFig. 1. A simplited represen- caton of the Baddeley and Fitch soeking memory mod dis) Visuospatial coer ‘Phonotogical ssexgag__sePom [I ¢ | IK )| | . ZI ‘comprehension or the reasoning tasks used in test of intelligence. ‘The weakness of this approach lies in the reliance on complex working mc:nory tasks that have a somewhat arbitrary construction and that do not readily lend themselves to a more detailed analysis of the component processes. The dual-task and neuropsychological approach can be utlized 10 successfully analyze the constituent processes of the slave systems but has made less headway in teasing apart the complexities of the executive controler. Individual Differences in Working Memory ‘The essence of the psychometric approach isto develop casks that ‘require the combined storage and manipulation of information and to correlate performance on these rasks with the performance of practically and theoreticaly important cognitive skills. One influen- ‘ial smady in this area was carried out by Daneman and Carpenter (7), who examined the processes involved in reading comprehen: sion. They devised a series of working memory tasks, one of which required subjects to read aloud or listen to aseres of shore sentences while retaining the lst word from each sentence for subsequent immediate recall. Hence, subjects might read or heat: “The ssilor sold the partor. The vicar opened the book” They should thea respond “parrot, book.” The test rypically stars with ro sentences and increases to a point at which subjects are no longer able ro recall all the erminal words. Ths point is designated the subjects working memory span. Daneman and Carpenter, and others asing similar techniques, ‘ypically found a correlation coeficens of about 0.5 ot 0.6 benween working memory span and reading comprehension, as measured by standardized tests (8). The span task does not have to involve language processing because similar coreelations ate found when simple arithmese, combined with word cecal, is substituted for sentence processing (9). ‘Subsequent studies have indicated that students with high work: ing memory span were better at coping with “garden path sentenc ¢,” which contain misleading context, and that dhey are better at rawing inferences from text, suggesting that they have a better _prsp of its meaning (10). ‘A second area in which the individual differences approach has been applied to the analysis of working memory is concerned with the study of reasoning and concentrates particularly on tasks that have radivonally been used ro measure intelligence. One example of this is the working memory analysis by Carpenter, Just, and Shell (11) of performance on the Raven's maccces ask a fest in which one Sector is missing from a complex pattern and the subjec is required to choose which of six possible options offers the best Completion. Christal (12) has also shown chat working memory tests provide improved prediction of technical learning capacity in U.S. Air Force receuits, when compared with more scholastic measures. Kyllonen and Christal (13) have carried out a series of studies, ‘each involving several hundred subjects who were requited t© perform a numberof standardized test of reasoning ofthe type used. to assess intelligence as well as a range of tasks that had been devised to estimate working memory capacity. For each study, their results “suggested 2 very high correlation beaween working memory capacicy 3) TANUARY 1992 and reasoning skill They concluded, however, that the two con- cepts, although closely related, were not synonymous; reasoning performance was more dependent on previous knowledge than was ‘working memory, which in contrast appeared to be more dependent ‘on sheer speed of processing. Components of Working Memory Although concurrent storage and processing may be one aspect of working memory iis almost certainly nor the only feature; indeed, Baddeley, Barnard, and Schneider and Derweiler (14) all suggest that che coordination of resources is the prime function of working ‘memory, with memory storage being only one of many potential demands that are likely to be made on the system. ‘One proposed role for the central executive i tha of coordinating information trom two or more slave systems, This feature of the ‘central executive was used in an attempt to test che proposal that Alzheimer’s disease i associated with a particularly matked deficit in ‘ental executive functioning (15). Patients with Alzheimer’s disease and both young and elderly normal subjects, were required t0 perform two tasks concurrently, one visual and one verbal. The difficulty of each rusk was adjusted so thar the Alzheimer patients were making the same proportion of errors as the control subject, and subjecss were then required to perform both tasks at the same time, Normal elderly subjeces were no mote impaired than young consrols by this requirement to coordinate, whereas the Alzheimer patients showed a marked impairment in performance on both the ‘memory and tracking tasks when required to combine them (16). AS the disease progressed, performance on the individual tracking and ‘memory span tasks held up very well (Fig. 2), whereas performance fon the combined tasks deteriorated markedly, as would be predicted by the hypothesis of a central executive deficit in Alzheimer's disease «a. The Slave Systems of Working Memory Although an analvtic approach to the central executive is begin ring to bear fruit, there is no doube that considerably more progress has been made with the simpler task of understanding the peripheral slave systems of working memory. The dual-ask paradigm has been _used t0 demonstrate the separability ofthe memory systems respon sible for lexrning by means of visuospatial imagery and of learning bby rote repetition. Imagery is disrupted by the requirement of performing a visuospatial fask, such as cracking a spor of light ‘moving on a screen, by certain types of eye movement, or by the presentation of irelevane visual material during learning (18) ‘There are separable spatial and visual components, with different Memorrspan arora) MS « Fig. 2. Dustask performance of mtients with Alzheimer’s disease in sere of thee sequential test (1, 2. and 8) 6 months apart. T. tracking task, MS, ‘memory span ask. Normal subject dd noe show a dference between single fd dul tsk condivions. Data tom Baddeley eal. (17)casks differentially recruicing the ‘wo. Farah (19) distinguishes one imagery component that is principally concerned with the represen ration of pattern information and that involves the occipital lobes from a second more spatial component that seems co be dependent ‘on parietal lobe functoning, Neuropsychological evidence supporss this dichoromy, with some patients having great dffeity in umag- ing and recalling such visual fearures as the shape of the cars of a spaniel dog or the color of a pumpkin but having no difficulty in spatial taser such as describing, routes or locating towns on maps; ‘other patients show exactly the reverse partern of deficits (20), Having found ways of separately disrupsing spatial and verbal processing, one can explore the relative conmibution of diferent subsystems co complex tatks, One example of this application ‘concerns the nature of the cognitive processes involved in playing chess, The literature reviewed by Holding [in (2/)} indicates that both visual and verbal coding have been claimed t0 be crucial by diferent seudies that principally rly on subjective report. We have ‘sought more objective evidence through a series of experiments that utilize the secondary-task technique co disrupt either the phonolog Jes! loop, the sketch pad system, or the central executive, Our first study involved memory tor complex chess positions and tested subjects ranging from the modest club player to the internaional grand master. As expected, expernse correlated highly with memory performance, but all subjects showed the same basic partern: n0 disruption from the concurrent verbal task but clear impairment from the tasks occupying che visuospatial sketch pad or the central executive. A second study required subjects to choose the optimum ‘next move from a complex midle-game position and found exactly the same pattem. Disruption of verbal acivity had no effec, whereas visuospatial disruption was clear, and this problem-solving task was ‘even more susceptible to central xecutive disruption than the rask in the first study (22), Analyzing the Phonological Loop ‘The phonological loop is probably the simplese and most exter sively investigated component of working memory. It les closest 0 the earlier concept of short-term memory and has been investigated ‘most extensively with the memory-span procedure. I is assumed to comprise two components, 2 phonological store that can hoki acoustic or speech-based information for 1 to 2 seconds, coupled with an articulatory control process, somewhat analogous 10 inner speech. This system serves ewo functions; it can maintain material within the phonological store by subvocal repetition, and it can take visually presented material such as words or oameable pictures and register them in the phonological store by subvocalizanon. “This simple model is able fo give a good account ofa rich range of laboratory-based findings. These include the following: 1) The acousuic similarity effect, This is the observation that the immediate ordered recall of items is poorer when they ate similar rather than dissimilar in sound (23). Hence, hearing and repeating alssimilar words such as “pit, day, cow, pen, rig,” is easier than a phonologicall similar sequence such 25 man, cap, ean, map, mad.” ‘Thus phenomenon is assumed tO occur because the basic code Involved in the store is phonological, similar items have fewer cistinguishing cues than dissimilar items and are therefore mote susceptible ro being forgorten. Similarity of meaning does not have this effect, suggessing, that this subsystem docs not relect semantic coding 2) The irelevant speck effet. This reers ro 3 reduction in recall of lnts of visually presented items brought about by the presence of irrelevant spoken material (24). Once again, the semantic character- istics of the material are fot important, with a language that is ss unfamiliar to the subject being just as disruptive as words in his or her native congue and nonseme spilables being as disruptive as meaningful words, The etfect is not due to simple distraction, because Joud bursts of noise have lite or no effect (25), These results ae interpreted under the assumption that disruptive spoken ‘material gains obligatory access to the phonological memory store. 3) The word.lengt ef. This provides evidence on the nacure of the subvocal rehearsal process. Memory span for words is inversely related to spoken duration of the words. Subjects can generally remember about as many words a5 they can say in 2 seconds (26). } This phenomenon accounts for diffrences in digit span when subjects are tested in diferent languages; languages in which digits tend 10 have long vowel sounds or more than one syllable cake longer to rehearse and lead 10 shorter memory spans (27). The model ean also explain the marked tendency for digit span in children to increase with age; as children get older, they are able to rehearse faster (28), 44) Antculatory suppression. it is possible to disrupt the we of subvocal rehearsal by requiring subjecss to urer some repeated irrelevant sound, such as the word "the." This process, known as ariculatory suppression prevents the subjects from rehearsing the ‘material dey are reying to remember and thus removes the effec of word length. Suppression also prevenes subjects from registering visually presented material in the phonological store. Recall of such visual material is reduced, and the acoustic similarity effect is abolished (29) The performance of neuropsychological patients with impaired short-term memory can also be explained as a deficit in the phono- logical store. They typically show’ no evidence of phonological coding in memory tks when presentation is visual, no word length, effect, and no influence of articulatory suppression, suggesting that these patients make lire or no use of their defective phonological short-term store (30) The Function of the Phonological Toop Patients with 3 specific phonological loop dencit seem to have remarkably few signs of general cognitive impairment. Although they pally have difficulty in comprehending certain types of complex sentences, interpretation of results in this area remains controversial (31). The most commonly held view is that the phonological store serves as 2 backup system for comprehension of speech under taxing conditions bur may be less important with simple, clearly presented material In recent years we have been exploring another possible function of this system, namely, its role in long-erm phonological learning, such as acquiring the voczbulary of one’s nanve, or even a foreign, language. in one study, we asked a patient with 2 very specific shorererm phonological memory deficit 10 lea eight items of Russian vocabulary, language with which the patient was unfamiliar; Wwe compared the ‘results with che patient's capacity to lear 10 associate arbitrary pais of words in the patent's nave language (32), People tend to lear pate of familiar words in terms oftheir meaning, and, as expected, the patient’s performance on thie cask was entirely ‘portal. In contrast, the patent filed eo iearn the Russian words with auditory presentation and was severely impaired relative to control subjects even wen presentation was visual, This result suggests chat shorc-teem phonolortical storage is important for new longterm phonological learning, Subsequent studies with normal adults have shown that factors that influence the phonological loop, such 35 articulatory suppression, word lengh, and phonological similarity, strongly influence foreign vocabulary acquisition yer show no eect on learning to associate pars of familiar words (33) SCIENCE, VOL. 255Evidence tor the importance of the phonological loop in native- language learning comes from a number of sources. Gathercole and Baddeley (4) studied a group of children with a spectc language disorder and found that their most striking cognitive deficits oc: ‘curred in a task involving hearing and repeating back unfamiliar fnonwords; on this nonword repetition tsk, S-year-old children with che language development of 6-year-olds functioned ike ‘A-year-olds, Further invemigation suggested thar dis was due neither £0 perceprval difficulties nor to difculties in speech procuc- tion but probably resided in the operation of the phonological shor-term store ‘A subsequent study assessed the role of che phonological shorter store sin the development of vocabulary across the normal range (35). A sample of 118 children was tested after starting school between the ages of 4 and S years. Their capacity for nonword repetition was measured, as was their nonverbal intelligence and their vocabulary, which was tested by speaking a series of words to the children and requiring them to point to appropriate picrures. Nonword repetition proved to be highly correlated with vocabulary and to be a powerful predictor af vocabulary I year lace. In an experimencal sioulation of new word learning (36), we ‘aught children new names for toy monsters. Two groups were rested that were matched for nonverbal intelligence bur that differed in nonword repetition capacity. Those with low capacity showed poor learning, particularly in the case of unfamulia invented names. Service (7) has studied the acquisition of English as a second fanguage by young Finnish children. Service took a number of measures of cognitive skill before the course began, including ‘measures of nonverbal ineelligence and of nonword repericion capacity, Two years later the chikiren’ performances on a range of fests of English language were correlated with these earlier mea sures. Once again, nonword repestion capacity, which is assumed to depend on shore-term phonological storage, was clearly the best predictor of subsequent success. Thus, the evidence supports the View that shor-erm phonological memory is crucial in dhe aequisi- ‘ign of vocabulary, Conclusion The concept of a working memory system that temporarily stores information as part ofthe performance of compiex cognitive task is proving ro be productive. Studies that have utilized the individu Ulference approach have linked working memory to performance on «range of imporrant tasks, including language comprehension and ‘reasoning. The more analvtic approach has shown that che concept forms a useful conceprual too! in understanding a range of neuto- psychological deficits, which in rurn have throsva light on normal ‘cognitive functioning. Working memory stands ae the crossroads between memory, attention, and perception. In the case of the slave systems, the phonological loop, for example, probably represents an evolution of the basic speech perception and production systems t0 the point at which they cin be used for active memory. Any adequate model of the phonological loop is thus likely © overlap substantially with an adequate model of speech perception and speech production. The visuospatial sketch pad is probably innately related £0 the pro- cesses of visual perception and action. The central executive clearly reflec a system concerned with the attentional control of behavior, with subsequent developments almost cerainly depending on par. let developments in the studv of attention and of the control of action, If these links can be sustained and developed, the concept of Working memory i likely eo continue t0 be 3 frutfial one. SUIANUARY 1992 REFERENCES AND NOTES See A, D, Baie (Huma Memory: Theory and Prate (Al am Bacon, Needtam Heh. MA. 1990), pp 3.66) for a teew. Nn ESK Yarrgon, J. Yok, Lam Veh Behar. 9,176 (1970), 8 “ier, cma, CW. Wh and © Lang Es. (terme Lonson, 1646) pp (O93, T Sullezad EK. Warnagton,Q. J. Exp. Pathol: 22,261 (1970; A. Ras. FL Spice, G. Valen E2anabo, Nevmpryons 2, 263 (1982). 6, Vall nd Slice Narnia pao of Shomer Memory (Cambodg ‘iw ees, cimbeae,1990) FRC, Ackinan and RM, Shin, The Pca of Leaingond Menno: ane or Resa onl Toy, KW. Spence BAe ex, New York, 1968) va 2 pp 3198, AD. Beer and Hitch, 0 The Poly of Lamina Motion, C. 1X Boner, Ea Academe Pes New Von 1979), vo By pp. 47-89. RG. Crowder. ts Phat 50, 291 (1982), M Daneman a PA. Carpenter J. Vor. Lea Vorb. Bas. 19.450 (1980) AD. Badaee,R Loe, Nuno South N. Breet tem. Can. 28, 119 (985), MB} Mactan & Miler J. Ede, Pp, 1314 (19%, FV Qabh. Yu AT Pastner Read 9,80 198: 84 T.Tumer and RW. Engle J. Mom: Lang 28,127 (1989 M, Daneman tnd A. Carpenter, J. Exp. Pye: Lea, Mem. Coe. 9, S61 UBBEN. IW, Oat Be J Baw Pehl A, 1 11980), B.A Carpenter MA. fut, Shell Pht R97, $4 (29% RE, Chal ArauoneLeery Han Rees Donte Fal Rep ScsiP- atch (Brocka Ais Fonte Bam T1991). 2-6. Kylonen and RE Ct linen 14,389 (2500), A.D. Bude, avkng Momany (Oxford Unie Pres, Ox, 1986); PBurar {2 Poa nh Pho of Langue, A Elis, E (Esau, Lon, 1985) ‘LZ bp 197-258: W. Shvesderns M. Dew in Th che of earn setlists GH Bows 8 nse fem New Yo ORT 1-7. Becker Alsemers Das: Advone i Bae Reeh and Therap, B. J Vruraman, $H: Conny ] H Cromdons Eas (Cer far Brain Sots a ‘Metin Chante Thue, Came, 1987), pp 343-348; Ht. Spina, 5 Dali Sa & Banden, A'D. Bader Com, Nosepyeo 198 (1988). ‘AD Baddeley, K- Lope, Brea Dei us, 66 Spee, QJ. Esp. Popa SA 0 (985) AD, Badly, 5. Bes 5. Dela Su R Lope. H. Spier, Brinn pee TLR Bracks... Esp. Pact. 19,389 1967), A.D. Bade. 8, Grok E Wighe, N"Thorvon i laeton tnd Peorancey MA Rabbit ad ‘omic ds easemic Pres London, 1973), 8. 9p 208-217 fee RF ope and A.D. Bade in Imagen” Cat Diop]. Ratan, D. Moths, P. Hampson Ea (Rouuedge and Kegan Mh London, 1990). pp. 103-138) for srenewfs A.D. Badly snd Lieberman, io denon Sod Pefamanc, RS. Nickerson, Bd (eos ils NJ 1980), 90: VIl pp IMJ. Fara, Pha. Re. 98,307 (1988) 'M Hammond, DN. Lenne, R, Cabana Can, yl. 20,439 Tam DFE Holling, The Pash of Chex Ski (Esha, Hilal, NJ, 1985): A D. Baddeley Anton Seaton Aue nd Cont A.D, ey a ‘Westra (Oxford Un raw, Oven pres TW Robins an preparaon, Goan, dr) Peto 88,75 1964); AD Badd, ut 18, Son veg : Oo Bot HLA Cole and A. Webb, J. Verh. Loam. Vo, Bean 15,17 (1976) P Slane Sai 8D. Baal. ha, 180 (198) HA Cite a 19,72 (HD, Sina A, D, Bley, geo 30 AD. Bude eral. J. Verh, Lear Ver. Behn. 14, 578 (1975), NG Flan A Henney, Beye Pt (1980) M Nave Benin ad Fees. QJ. Esp. Peal 3.739 (1986). Nicolson niger an Leong MP. Freda |, ®- Dae O'Conee, Fir (Penun, London, 1981), pp. 9-186, G Hlich and M.S. Hala. Phi Saami 92, 325 1983). Hale, N Thomson, © Mase A Lovrence Ep Child Phe. 38,24 (198). ‘AD. Bade, VT Lema G, Vala Qf By. Pcl. $6,233 (1984); D urate ap edt 78-679 1968) Gale anh Daly J. Vem Lom. Vor, Bee, 23.151 (198). SG. Valle tnd T. hale. Fis, Newepayteleual Inpaimens of Shore ‘Monory (Cambrige in: Pet: Cane 1950) I'D Basel CP3pegna, G,Valun, J- Mem Lng, 27,586 (988), Capagno. Vantin A. Body s.r. Papago and G Vale 27 an Pet bc 7 (1990) pres Nps S$ Gathercle and AD. Bade, J. Mom Lan. 28,336 (19%) i 24,200 (1989), Be) Pehl Bl, 439 (1990, fran cnt Fin. 11 po a.” Memes Tur ace was ween whe? wating the Unveiy of Teas a Asn, strc T wt supported bythe Wee Chae of Human Ptormaace at pach os Oe suns Gates Lope Peron snd Spee ‘fe commune vote an eed sper sareces 559