EIREE|SG0"Votume 208, pp. 111-1182 SCIENCE Acquisition of a Memory Skill K. Anders Ericsson, William G. Chase, and Steve Faloon Copyright © by the American Association for the Advancement of ScienceAcquisition of a Memory Skill Abstract. After more than 230 hours of practice in the laboratory, a subject was ‘able to increase his memory span from 7 to 79 digits. His performance on other ‘memory tests with digits equaled that of memory experts with lifelong training, With ‘an appropriate mnemonic system, there is seemingly na limit to memory perforni- ‘ance with practice. One of the most fundamental and stable properties of the human memory system is the limited capacity of short: term memory. This limit places severe ‘constraints on the human ability to pro- cess information and solve problems (I) a the other hand, this limit (about sev- cn unrelated items) stands in apparent ‘contrast to documented feats of memory experts (2). Whether these memory skills. are the result of extensive practice or of exceptional ability has often been dis- pputed. The goal of this research is to ana- lyze how a memory skill is acquired. ‘An undergraduate (S.F.) with average memory abilities and average intelli ‘gence for a college student engaged in the memory span task for about | hour a day, 3 to 5 days a week, for more than V/s years. S.F. was read random digits at the rate of one digit per second; he then re- called the sequence. Ifthe sequence was reported correctly, the next sequence was increased by one digit; otherwise it was decreased by one digit. Immediately after half the trials (randomly selected), S.F. provided verbal reports of his thoughts during the trial. At the end of each session, he also recalled as much of the material from the session as he could. On some days, experiments were substituted for the regular sessions, During the course of 20 months of practice (more than 230 hours of labora- tory testing), S.F's digit span steadily improved from 7 to almost 80 digits (Fig 1). Furthermore, his ability to remember digits after the session also improved. In the beginning, he could recall virtually nothing after an hour's session; after 20 months of practice, he could recall more SCIENCE, VOL. 206, 6 UNE 1980 than 80 percent ofthe digits presented to him. On one occasion (after 4 months of practice), we tested S.F.'s memory after the session with a recognition test (be- cause recognition is a much more sensi- tive measure of retention than recall is); he not only recognized perfectly 3- and ‘digit sequences from the same day, but it spon Proce (Sey backs) Fig. 1. Average digit span for Fas a fune- tion of practice. Digit spun is defined as the Ieogth ofthe sequence that fs correct $0 per. cent of the time: under the procedure fo. lowed, tis equivalent to average sequence length’ Bach day represents about | hour's prctice and ranges from $5 tals per day in the beginning to 3 uals per day forthe longest sequences. The 38 blocks of practice shown, hete represent about 190 hours of practice in- terspersed among these practice Sesions are approximately 40 hours of experimental ses- sions (nat shown). ‘reto7swaD6- 1181500 300 Copyigt 1980 AAAS also recognized sequences from in the week, With only a few hundred hours of practice, S.F. would be classified as beginner at most skills. However, in his field of expertise, memory for random digits, he compares favorably best-known mnemoni Sand Hunt and Love's VP. @). For ex- ample, after about 6 months of practice, wwe set S.F. the task of recalling a matrix of 50 digits because data on this task are available for both S. and VP. S.E’s study times and recall times were atleast as good as those of the lifetime mem- ory experts. ‘The Key to understanding this skill comes from analyses of S.F.'s verbal re- ports and his performance on various ex- perimental tests. We will first describe two essential components of this skill: his mnemonic associations and (i) his retrieval structures. Then we will ad- dress the question of whether or not S.F. as able to increase his short-term mem- ory capacity ‘The most essential part of S.F.'s skill is his mnemonic associations, which he described in great detail in his verbal re- ports. The principle of a mnemonic is to associate unknown material with some- ‘hing familiar; the advantage is that it re- lieves the burden on short-term memory because recall can be achieved through & single association with an already-exist- ing code in long-term memory. What S.F. did was to categorize 3- and 4digit groups as running times for various races @). For example, 3492 was recoded as. “3 minutes and 49 point 2 seconds, near world-record mile time" ¢). During the first 4 months, S.F, gradually con- structed an elaborate set of mnemonic associations based initially on running times and then supplemented with ages (€93 was “89 point 3, very old man") and dates (1944 was “'aear the end of World War II") for those sequences that could not be categorized as times. Running times (62 percent) and ages (25 percent) account for almost $0 percent of S.F.'s ‘mnemonic associations. ‘There are several lines of evidence conceming the muemonic associations (On the basis of S.F.'s verbal reports, we were able to simulate his mnemonic as- sociations, that is, to abstract a set of rules that categorizes a sequence of dis- its a5 3- and 4-digit running times. When we compared the simulation to the ver~ bal reports, between 85 and 95 percent of the time the computer categorized the digit sequences as $.F. did. By means of the simulation, we were also able to de- termine which sequences of digits would be categorized as running times and st ts, such as Lutia’sosipeterrans peepee plies Reece ea ere 7, digit groups followed by two 3-digit bering the order of more than four ua, groups within a supergroup. Introducing another level of organization by sub- dividing these supergroups allowed SE's performance to improve rapidly so that he now averages almost 80 digits. His current retrieval organization can be described as a hierarchy with three lev- els, and his retrieval structure for 80 dig- its can be illustrated ia the following ‘way, with spaces corresponding to levels in the hierarchy: 444444 333533 444 333 448 5 Besides the verbal descriptions, there js a greal deal of additional evidence that S.F. uses hierarchical retrieval struc- tures, Probably the most straightforward evidence comes from his speech patterns during recall, which almost invariably follow the same pattern, Digit groups are reealled rapidly at a normal rate of speech (about 3 digits per second) with pauses between groups (about 2 seconds between groups, on average, with longer pauses when he has difficulty remember- ing), At the end of a supergroup, how- ever, there is a falling intonation, gener- ally followed by a longer pavse (5) In several experiments, we verified that groups are retrieved through the crarchical structure rather than through dliteet associations between groups. In cone experiment, instead of asking for re- call after presenting the digits, we pre- sented SF. with a 3- or 4digit group and asked him to name the group that pre- ceded it or followed it in the sequence. He required more than twice as long, on the average, if the preceding or following group crossed a supergroup boundary (10.0 seconds) than if it did not (44 see- fonds). In another experiment, after an hour's session, we presented S.F. with 3- and 4-digit groups from that session and asked him to recall as much as he could about each group. He invariably recalled the mnemonic associations he had generated, and he often recalled a great deal about the location of the group within the hierarchy, but he was virtually never able to recall the preceding or fol- lowing group. After all this practice, can we con- clude that $.F. increased his short-term memory capacity? There are several tea sons to think not. () The size of S.F.'s groups were almost always 3 and 4 d its, and he never generated a mnemonic astociation for more than 5 digits (6). (i) He almost never allowed his rehearsal ‘group to exceed 6 digits. (ii) He general- ly used three groups in his supergroups ‘and, after some intial diticulty with five ‘groups, never allowed more than four groups in a supergroup. (iv) In one ex- perimental session, SF. was switched from digits to leters ofthe alphabet after 3 months of practice and exhibited no transfer: His memory span dropped back to about six consonants These data suggest that the reliable working capacity of short-term memory is about three or four units, as Broadbent has recently argued (7), and that it is not possible to increase ‘the capacity of short-term memory with extended prac- tice. Rather, increases in memory span are due to the use of mnemonic associa- tions in long-term memory. With an ap- propriate mnemonic system and retrieval stoucture, there is seemingly no limit to improvement in memory skill with prac- tice K. Anpens Ericsson WittiaM G. CHase ‘Steve Fatoon Department of Psychology, Carmegie-Melion University, Pittsburgh, Pennsylvania 15213 1G. Aas, Rocha Rew, 1 936% A Stalin 2°Stsin mer mabe Ing" cence al"Englewood. Cis” N 2 KR vais as documented the cae tory of dee exceptional peso, wip temed ee ibe bigs aun vival ran for eerammeatonctimeny (ae indel nai (avon; New Yt) 63) and Hon and T. Love fae doserbed ann eX lng areocttons i sever inpunges it ‘SW: Bele and E” Marin Eas, Couns Fro amit Humare Memory (Wino, Wain fon BC 918, 9 0 2 SE InS ood leg ‘ists He clases runing tees ito ot feast iUSlor ees fo es aon, 4 Thecategory bel by hast was mat alicieat to snore i ena peste. complete Sci fees seam eae Won ane wh ean ‘astern ated 1 Bs. ionaton, apd sess stems ae wel sce tins ols Sines he Brgitn Glowon ‘the faucet The drionaton of American Engl (Uv. ot Hea he hes ey ion RIDA te Compre tn oi ere inlet) te poste aes re EUR herp les reseed SF SEL Ned Betcha apetnet nes fealy peter 6 The rion azpciations of tihng cae [lsprapper to be ented Sort ait [@.E Males Rocha. Ereirungsbond' (910) 1 BER Beadet nSate ong Teo Meme aera Kenge aed A Wis, Es (Wo, Wei Yon 13,93 1 Riper yet omar conrs toe (Pant HHTIREY ote Racal Isle ot int Rea, Wennoat’ R Andewon MT HL Chi, We Jones, MW Sehanaskd HEA, Sinon for tervals comments 27 February 1980 SCIENCE, VOL. 208