The episodic buffer: a new component of working memory? Alan Baddeley In 1974, Baddeley and Hitch proposed a three-component model of working memory. Over the vyears, this has been successful in giving an integrated account not only of data from normal adults, but also neuropsychological, developmental and neuroimaging data. There are, however, a number of phenomena that are not readily captured by the original model. These are outlined here and a fourth component to the model, the episodic buffer, is proposed. It ‘comprises a limited capacity system that provides temporary storage of information held in a multimodal code, whi capable of bi formation from the subsidiary systems, and from long-term memory, into a unitary episodic representation. Conscious awareness is assumed to be the principal mode of retries from the buffer. The revised model differs from ‘the old principally in focussing attention on the processes of integrating information, rather than on the isolation of the subsystems. in doing so, it provides a better basis for tackling the more complex aspects of executive control in working memory. "Tire src ihn opine sence comin i ferent forms, ranging from detailed mathematical or compute tional models of narrow and precisely defined phenomena, co ‘broad theoretical frameworks that atempe to make sense of wide range of phen more detailed speciation. The purpose of such framework is wo represent prompting further questions that ae tactble, This is likely mena and that kave open much of the shai currently known while atthe same time cither to excend the range of applicability of the model, oft increase ins theoreical dept, subsequendy lading to mote precisely specified sub-model. The concept of working mem ‘ory proposed by Baddeley and Hitch provided such 2 fame. ‘work for conceptualizing the role of temporary information storage in the performance of wide range of complex cogni tive tn (ce Box 1. Te represented a development of eaicr model of shore-serm memory, sich as thote of Broadbent snd Adkinson and Shili?, bus dileed in ewo ways, First i abandoned the concept of unitary or ‘component system, and second it emphasized the Function of sofa mul in fa such asytem in complex cognition, rather than memory perse (Over the 25 yeas since the publication of our initial pape, the concept of wotking memory (WM) has proved vo le, In one form or another, i continuer to be actively ued within many areas of cogn including mainstecam cognitive prychology*, newopsychol ‘oy, neuroimaging’, developmental prychology” and com: putational modeling”. However, there have always been Phenomena that did not fit comfortably within the Baddeley and Hitch model, particularly in its more recent farm. An artempt co come to serms with thee hat led to a lation of che sheoretcal framework, which will be described below. The reformulation leads o the proposal of new component of working memory, the ‘pizodic buffer Problems for the current model The phonological loop: imi an lirsitations Te phonological oop gives a reasonably good account of a wide range of data (ee Box 2) «na that donot seem to ft nes into th rious futher modification, Consider, ist the effet of aticw- There ate, however, phenom ictute withous Iacory suppression, whereby the subject continues to utter an ierelevane word such a ‘the’, while attempting to remember and repeat back a visually presented sequence of numbers. Acconing tothe model, suppresion should prevent the seg isration of visual material in the phonological loop, produc ing a devastating impact on subsequent recall. Suppression does have significant fet, bu by na means devastating: in a sypical ud -mony span might drop from 7 0 5 digi, Furthermore, patient with grosily impaired sho ‘erm phonological memory, resulting in an auditory memory span of only one digit, can ypicaly recall about four digie with visual presentation", How ae such digits stored? An obvious possibilty isin tems of the vswonptia ste pad. However, the evidence indicate hat hi sete ie good at snoring a single complex pattem, but not suited to sedate” more i vil coding were involved then one might x pecteuppresion to make eal performance very sensitiv toc fects of vss smarty A recently by Logie eal dos i auditory Eusth deed show visual similarity efects®. They are, however, small and not limited ro conditions of aiculatrysuppresion pr sniseserspnense2 Pparmen of eae Pathog, Urey fis 2 Wanda Rad Beal OK B54 TW ade baer 9n8Baddeley ~ The episodic buffer In working memory Box 1. The concept of working memory The tern working memory is wid in at lat chic different way in differen areas of copaitive since. Ti wad bee, and ts engnitive pychology gener to ree ated expcy system allowing the temporary storage and manipulation of n- Formation neetery fr sich complex sks ar comprehension, leaning and reaeoning (Refs 3), In the animal leaning lb oratory the term refer co the worge of information scot cra al performed within te same dy, a demanded by ike such asthe radial arm maze (Re © In acl ineligene, production system achitecues apply the ce tothe com ponent, ofen ualinted in capaci, tha i assumed te be responsible for holding the productions (Ref. ‘These thiee meanings re thus not serchangeable Performance of au on s the radial ams maze fr exmple probably eles upon longtecm memory (LTM), while che elite capaci of the woking memory companent yp cally asnmed by peodicton rtm architctres dif smatkedly fom the expaiy Limitation axed by mas ofthe model proposed within cognitive peychology. ‘Themls component model of working memory (WM) shat form the bat ofthis revew developed from an eter concept of shoreterm memory (STM), tac was assumed co eomprise & ‘unitary temporary soage sytem, This appreach was cpt by «he model of Acknson and Shin (Ref), Howevet thet (@) — Vissospatal |“ 7Cental) ~ | Phonotogia stetorped | _ (execute oop Sa Visvospatial Phonological sketenpad oop T F vvitual Episodic i ‘semantics mM ee memory propose by Baddeley and Hitch (Refsb). The ieee Component model assures en attentional contol, he cena ‘hl ketehpag, which perfome 3 sma funeson fer vuln ‘avon. The to subs stems hemseles form ctv ores ‘hat ate capable of combining Information from Sensory Np, sd fom the cena execute. Hence a memory tae In the Phonological stare might stem ether rom re aur inpu, (from he uve relation of aly presented tam ch ‘i leer (b) A further development of the WM madel ‘became clear that he phonological lop lay an important ole ‘nfng-termphonolelcaiearning nadtionta shorter tor sige. As such ts sociated with the development of vrabulry ‘neuen andthe speedo acquisition of ferign language ‘ocabulayin au The heded areas repent ‘rytalied co titive tins eapabe of scumiting longer kel {eg language and eran nowlege). Urohsaed tere are assumed tobe Au capaci, sah a5 terion and temporary Storage, and are themselves unchanged by learning the than ‘ety wa the cysaliaed systems et ‘This lfer of visual similarity on span for verbal materials presents something of a problem: given that it occus under ‘model encountered pioblem (I) in aecounting fr the elation ship berween pe of encoding and TM (Ref, (2) in expla ing why patients with groey defective STM had apparently normal UTM. and 3) in accounting or the ees of ange fon: current eon lamingcomprchending and reasoning eb) Baddeley and Hitch proposed che three component WM ‘model (own in Fig. 1a) ro accoune fo thi pate of dts The model compried an arentonal contol ye, the “ene executive aided by two subidiay save systems, the ‘phone- log loop and the visuospacal seca’ (Ref). The lop ic assumed o hold vtbl and scout information using en poray sot and an aiulatry teat system, which lsc lesion studies, and subsequenly aeworadilogcl studi, ug ested ate principally ected with Brodsann aes, 40 and 14 respectively. The teichpa eave to held visscopatil information, o be Sactinable aca separate vu, patil nd posibly Hnacahrtc component and tobe principally rp seated within the ight hemiiphere area 6 19, 4Dand 7). The central exeetive ir alo assumed tobe Factonable. Although it ie ear wel unentod, font abe arse appear robe strongly implcsted. An excelent recent overview of shoreserm and working memory i given by Guthereae (Ref) Working memory and long-term memory wete inially tweated a quite separate because paints wih car shorter phonologil deicis appear to have inact LIM (Ref. b) Subiequent seach hs shown tha ich patients do have spe- sili dfcis in Long-term phonolgieal leaning, for example leaning the vocabulary ofa ne language (Ref. Further ev dence bated on the lik bere phooolgial loop pefor smance and vorabulay level incon, gets thatthe lop right have evolved to enhance language sequinion (Ref. ‘As prediced by this supposition, patients with phonological loop deficit have gzest dificlyn aequrng novel voabelry. Te seom klyehe similar ancion served by she wep ‘al ecchpad although thee ie ae yet ile nvesigation of hie topic H one aceps the fypothess ofan equialenefnction forthe serchpad,posbly in cquting visuospatial seman, ‘hen the famework it modified to that shown in Fig. 1 ler, er (1960) Ran and he Sucre of eh Hale Desai, 20. ad te, (978) Worng memory. The Paolo of leaming tnd Motivation (homer, GA. ed) {Dt D5. a (58) Hippecamesfuncton: woking memary srcognive napping? Py eal 8239-248 nena, A aps Simon, HA. (972) Haman Problem Sabin seinen, RE. And Siti, RML (SER) Human memory & ‘ropoted sem an ie convl proces. The Phaleay of ‘Spence, ep. {Foe FIM an Ldn, FS. (972) Leva procesing: Frarenoth for memary rset J Vara enn Verbal Behav 9 cathe, 5. (1998) Cognit approaches othe deeloment Iueey, A. ea 1998 Wien gtr eming depend on 1 cate Ra. 1989) Meas a id and yale ntligenc: 2 “el experiment Ebue Pho S22 standard non-suppresed conditions, it indies that visual and phonological information ate combined in some way. The