Atkins on 71

8/3/2019 Atkins on 71

http://slidepdf.com/reader/full/atkins-on-71 1/43

THE CONTROL PROCESSES OF SHORT-TERM MEMORY

by

R. C. Atkinson and R. M. S h i f f r i n

TECHNICAL REPORT 173

A p r i l 19, 1971

PSYCHOLOGY SERIES

Reproduction in Whole o r i n P a rt i s Permitted f o r

any Purpose of the United S ta te s Government

This research has been supported by th e N atio na l Science

Foundation under grant NSFGJ-443X.

INSTITUTE FOR MATHEMATICAL STUDIES IN THE SOCIAL SCIENCES

STANFORD UNIVERSITY

STANFORD, CALIFORNIA

8/3/2019 Atkins on 71


'rHE CON'rROL PROCESSES OF SHORT-TERM MEMORY

R. C. Atkinson and R. M. Shiffrin

Stahford University

Stanford, California 94305

Human memory is divided into a short-term working

memory and a long-term permanent memory. Control pro

cesses act within th e short-term working memory to makedecisions and regulate information flow, thereby con-t roll ing learning and forgetting.

The system by which information is stored in and retrieved from

memory has always been a topic of great interest to psychologists. The

English associationists and early experimental psychologists like Wilhelm

Wundt, William James, and Ernst Meumann relied upon introspective tech-

niques to generate their theories. Their introspections led them before

the turn of th e centu ry to divide memory into short-term and long-term

components. They discerned a clear difference between thoughts currently

present in consciousness and those that could be brought to consciousness

after a search of memory that often required considerable effor t . For

example, this sentence is in your current awareness, but the winner of

th e 1968 World Series, while probably in memory, reqUires some ef for t

to re tr ie ve and in fact may not be found at al l .

Despite i ts intuitive attractiveness, the short- versus long-term

view of memory was largely discarded when psychology turned to behaviorism

which emphasized animal as opposed to human research. The short- versus

long-term dist inct ion received l i t t l e further consideration unt i l th e

1950's when a number of psychologists, p a r t i c u ~ a r l y Donald Broadbent in

England, Donald Hebb in Canada, and George Miller in th e United States,

1

8/3/2019 Atkins on 71


reintroduced i t (see George A. Miller , Information and Memory, Scientific

American, 1956, 195 (2), 42-47). The growth of two-process systems was

accelerated by the concu rren t development of computer models of behavior

and mathematical psychology. The two-process viewpoint is now undergoing

considerable theoret ical development and is th e subject of a large re-

search effor t . In par t icu lar , the short-term memory system, which we

will cal l short-term store (or STS) , has achieved a position of pivotal---------importance. I ts importance stems from processes carr ied out in STS that

are under the immediate con tr ol o f the subject . These control processes

govern th e flow of information in th e memory system; they can be called

into play at the sUbject's discret ion, with enormous consequences on

performance.

Some control processes a re used in many situations by everyone, and

others are used only in special circumstances. Rehearsal, an overt or

covert repeti t ion of information, is employed in numerous s i tuat ions:

when remembering a phone number unt i l i t can be writ ten down, when re-

member;i.ng the names of a group of people to whom you have just been

introduced, and when copying a passage from a book, to name a few examples.

Coding refers to a class of control processes in which long-term retr ieval

is enhanced by placing the to-be-remembered information in a contex t of

addit ional and easi ly retrievable information. For example, students

sometimes learn the twelve cranial nerves w ith the use of the mnemonic

"S:n S:ld S:lympus I -,!,iny -,!,op !2 ! inn J2nd Qerman ~ i e w e d !:!ome Bops," where the

f ir st le tte r of each word corresponds to the f i r s t l e t t e r of each nerve.

Imaging refers to a control process in which verbal information is re-

membered through the use of visual images. The ancient Greeks made

2

8/3/2019 Atkins on 71


extensive ~ s of this process. Cicero suggested learning long l i s t s (or

speeches) by placing each member of the l i s t in a visual representation

of successive rooms of a well-Known mansion. A number of other control

processes, i n c ~ u d i n g decision rules, organizational schemes, retr ieval

strategies and problem solving techniques, will be encountered in this

art ic le . The point to Keep in mind is th e optional nature of control

processes. I n con tr as t to permanent structural components of the memory

system, th e control proceSses are selected at th e subjec t 's d i scret ion;

they may vary, not only with different tasks bu t even from one encounter

with the same task to the next.

An Outline of th e Memory System

We believe that the o v e r a l ~ memory system is best described in terms

of the flow of information into and o ~ of STS and th e subject 's control

of this flow. Before describing the system i t is helpful to introduce

terminology with which to discuss information flow. All phases of memory

are assumed to con sis t o f small units of information which are associ-

atively related. Any set of information units that are closely inter-

related will be termed an "image" or "trace" (thus "image" does not

necessarily imply a visua l represen ta t ion). I f the pair "TKM - 4" is

presented for memory, the image stored might include the size of th e

card on which th e pair is printed, the type of print , the sound of th e

various symbols, the semantic codes , and numerous other units of infor-

mation.

The basic phases of th e memory system and th e types of information

flow are i l lUstrated in Figure 1 . Information from the environment is

accepted and processed by the sensory registeps in th e various sensory

3

8/3/2019 Atkins on 71


Figure Caption

Figure 1. Information flow in the memory system . Env ironmen tal in-

formation is processed by sensory r eg is te rs in th e various physical

modalities and entered into shor t- te rm store (STS). The information

remains temporarily in STS, th e length of stay depending on control

processes. While information remains in STS i t may be copied into

long-term store (LTS). While information remains in STS, information

in LTS associated with i t may also be activated and entered in STS.

Thus, i f a picture of a t r iangle is presented, this visual information

is processed and entered into STS. Then, since the verbal name

"triangle" is associated with this visual information in LTS, this

verbal label is also entered into STS.

3a

8/3/2019 Atkins on 71


ENVIRONMENTAL INPUT·

IVISUAL

IIAUDITORY 1··· ...... ···1 HAPTIC

I..

SENSORY REGISTERS

STS

TEMPORARY

WORKING

MEMORY

1---------,I CONTROL PROCESSES I

I II REHEARSAL, II CODING, II DECISIONS, II RETRIEVAL II STRATEGIES IL .J

RESPONSE

OUTPUT

LTS

PERMANENT MEMORY STORE·

3b

8/3/2019 Atkins on 71


modalit ies, and entered into STS. Information resides in STS for a period

of time that is usually under the control of the subject. By rehearsing

one or more items the subject can keep them in STS, but th e number that

can be maintained in this way is s tr ic tly l imited. For example, most

people can maintain 7 to 9 digits . Once an image is los t from STS, i t

cannot thereafter be recovered from STS. During th e period in which in

formation resides in STS i t may be copied into long-term (or LTS);

we shal l see that th e t ransfer of information from STS to LTS is highly

dependent upon rehear sa ls o f tha t information in STS. LTS is assumed to

be a relat ively permanent memory store, from which information is not

los t . Information is copied from LTS to STS as well as in th e reverse

d ire ctio n; in fact , we assume tha t during the period an image resides in

STS, some information in LTS closely associated with that image wil l be

activated and also entered into STS. Thus information entering STS from

th e sensory regis ters will in i t ia l ly be specific to the modality of input,

but almost a t once close associations from LTS in a l l modalities wil l be

activated and placed in STS. For example, a word may be present ed v isu

al ly, but immediately af ter input th e articulatory-verbal "name" and

associated meanings wil l be activated from LTS and placed in STS.

Our account of STS and LTS does not require that th e two stores

necessarily be in d iffe re nt p arts of the brain, or involve different

physiological st ructures . I t is pos si ble, f or example, to view STS

simply as a temporary activation of some portion of LTS. The same phy

siological structures might be involved in both instances, th e only

dist inct ion being whether or not a given structure is current ly ac tivated .

AlSO, in our thinking we tend to equate STS with "consciousness"; the

4

8/3/2019 Atkins on 71


thoughts and information of which we are currently aware can be considered

to be part of the current contents of STS. Such a statement l ies in th e

realm of phenomenology, and as stated cannot be scientif ically verified.

Nevertheless, thinking of STS in this way may help the reader conceptual

ize the short-term system. Because consciousness is equated with STS,

and becaUse control processes are centered in and act through STS, this

store is cQnsidered to be a "working memory": a store in which decisions

are made, problems are solved , and information flQW is directed.

Retrieval of information frQm STS is quite fas t and accurate. Ex-

periments by Sau l S te rnberg a t Bell Telephone Laboratories and others

have shown that the retr ieval time for information in STS such as let ters

and numbers ranges from lO to 30 milliseconds per cha ract er . The re

tr ieval of information from LTS is considerably more complicated. So

much information is contained in LTS that the major problem is finding

access to some sma ll subse t of this infQrmation which con ta in s the

desired image. This problem might be likened to the task of locating a

part icular book in a l ibrary. Once the book is lQcated, i t may then be

scanned in an attempt tQ reCQver the desired information. We prQpose

that the subject activates a l ikely subset of infQrmatiQn, places i t in

STS, and then scans STS fQr the desired image (which may no t be present

in the current SUbset). The retr ieval process therefore becomes a search

in which various SUbsets are activated and scanned. The conception is

depicted in Figure 2. On th e basis of the tes t query th e subject selects

a small se t of features t ermed "probe information" and places the probe

in STS. The subset of information in LTS c losely associated with the

probe wil l then be activated and entered into STS; this SUbset is termed

5

8/3/2019 Atkins on 71


Figure Caption

Figure 2. Information flow and decisions during the search of long

term memory. The probe is placed in STS, then information in LTS

c lose ly associated with the probe is activated and placed in STS. This

set of informationis

called the "searchcset." Eefore the search-set

is los t from STS, th e subject draws images from the search-set for ex

amination. I f the desired information is not found, th e search is

ei ther stopped, or recycled to another selection of probe information.

5a

8/3/2019 Atkins on 71


PRESENTATION OF

TEST INFORMATION

CHOICE OF RETRIEVAL

STRATEGY

SELECTION OF

PROBE INFORMATION

ACTIVATION OF RELATED SEARCH SET

IN LT S AND ITS TRANSFER TO STS

DECISION TO CONTINUE

OR TERMINATE SEARCH

RESPONSE CHOICE

AND ITS OUTPUT

5b

8/3/2019 Atkins on 71


the "searqh"set." The subjeqt seleqts from the searqh-set some image,

whiqh is then examined. The informat ion extraqted from the seleqted

image is uti l ized for a deqision: has the desired information been found?

I f so, the searqh is terminated. Even i f the information has not been

found, termination may occur i f the subject decides continuation is un

l ikely to be productive. I f the search does continue, the subject begins

the next cycle of the search by selecting a probe once again. This may

or may not be th e same probe used on the preceding cycle, depending upon

the subject 's strategy. For example, the subject may be asked to search

for s.tates of the United States start ing with the l e t t er M. He may do

so by generating s ta tes at random and checking their f i r s t l e t te r ( in

which case the same probe information may be used on each search cycle)

or he may generate success ive s tates in a regular geographic order ( in

which case the probe information is systematically changed from one qyqle

to the next). I t qan be shown tha t s tr ate gie s in whiqh th e probe in

formation is systematiqally qhanged will more often result in sucqessful

retr ieval , but wil l take longer to do so than alternative "random"

st rategies . Note that t he F reud ian COnqept o f repre ss ed memories would

be handled in th is framework by an inabil i ty of the subjeqt to generate

an appropriate probe.

The E ffeqts of Rehearsal

The reader has undoubtedly notiqed th at th is theory portrays th e

memory system almost e nt ir el y i n terms of the operations of STS. This

is qUite intentional. In our view, information storage and retr ieval is

best desqribed in terms of the flow of information through STS, and in

terms of the subjeqt ' s qOntrol of th e flow. One of the most important

6

8/3/2019 Atkins on 71


of these control processes is rehearsal; rehearsal is an overt or covert

repetition of information that ei ther increases i t s momentary strength

in STS or otherwise delays i t s loss. Some examples have been mentioned

ear l ier ; other uses of rehearsal occur during the taking o f lecture notes

or .during the act of performing mental arithmetic. Rehearsal can be

shuwn not only to maintain information in STS bu t also to control t ransfer

frOm STS to LTS. We.will pre sent several experiments concerned with an

analysis of th e rehearsal process.

The research in ques tion involves a memory paradigm known as "free

recal l ." Because the exper iments to be considered here are based on one

or another variation of this paradigm, i t will be described in some de-

t a i l . The si tuat ion is analogous to one in which you are asked to name

th e people present at the la s t large party you attended. The experimental

procedure is extremely simple. A l i s t of random items (usually common

English words) is presented to the SUbject one at a time. Following

presentation the subject attempts to recall as many words as possible in

any order. Many psychologists have worked with this paradigm and major

research efforts have been carried out by Bennet Murdock at the University

of Toron to , Endel Tulving a t Yale University, and Murray Glanzer at New

York University. The result of principal interest is the probabili ty of

recall ing each item in a l i s t as a function of i t s ser ia l presentation

po sitio n. P lo ttin g this function yields a U-shaped curve of the form

presented in Figure 3a. The increased probabili ty of recal l for th e

f i r s t few words in th e l i s t is called the primacy effect ; th e large in-

crease for the las t 8 to 12 words is called the recency effect. There

is considerable evidence that the recency effect is due to re t r ieval

7

8/3/2019 Atkins on 71


Figure Caption

Figure 3. The probability of recall as a function of ser ia l presenta-

t ion position for various free recall experiments. In free recal l a

l i s t of words is sequentially presented to the subject and then he is

asked to recall them in any order. (a ) The basic ser ia l position curve:

th e rise on the r ight is called the recency effect because these words

are the most recently presented; the rise on the l e f t is called the

primacy effect because these are the f i r s t words presented. (b ) I f an

a ri thmet ic t ask is interpolated between presentation and recal l the

recency effect disappears bu t earl ier portions of the curve are unaf-

fected. The recency effect t hu s appears to be due to retr ieval from

STS, and is eliminated when arithmetic causes the words to be lost from

STS. (c) The l i s t length effect demonstrates LTS retr ieval failure in

free recal l . Words in long l i s t s are recalled less well than words in

short l i s t s . (d ) The time available to rehearse each word affects LTS

retr ieval: slower presentation resu lts in better recal l . The above

graphs are idealized, and are based on experiments reported by James

Deese , Benne t Murdock, Leo Postman and Murray Glanzer.

8/3/2019 Atkins on 71


o

1.0 . . - - . - - - - - - - - ~ . . . . _ - - ' - - ' - ~ - - - - ........

': r - - - i " ' ~ " ' ~ " ' ~ " ' 1 - 0 - a - - - - - - - ; - - - - =r ; ; i K ~~

40 WORD LIST

J ~ - - 20 WORD LIST

WITH INTERVENI NG ARITHMETIC

WITHOUT INTERVENING ARITHMETIC -_

::l .5

<luUJ

n:

II..o 0 L - . . . I - _ ~ -- ,-_.. . .L LJ

>-1-1 .0 . . -. . . .---------.-, . .------------, . . . . . . ,

-J

IX!

<lIX!

o

lt .5

01..-...1---------.1...---------.1-.1

1.0 . . - . . , . . ---------.....------------,,.......,

.5

___-- 2 SEC. PRESENTATION RATE

I SEC. PRESENTATION RATE

01..-...1--,--------.1...---------.1-.120 40

SERIAL PRESENTATION POSITION

7b

8/3/2019 Atkins on 71


from STS, and that the earl ier portions of the ser ia l position curve re

flectLTS retr ieval only. In one paradigm, for example, the subject is

required to carry out a di f f icul t arithmetic task for 30 seconds immed-

iately following l i s t presentation and then asked to recal l . One can

assume that the arithmetic task causes th e loss of a l l words in STS so

th at r ec al l re flec ts LTS retr ieval only. Indeed, the recency effect is

eliminated when this experiment is performed; fur thermore , the ear l ier

portio ns o f the ser ia l position curve are unaffected (see Figure 3b).

Variables that influenceLTS

but notSTS

also can be manipulated.

In these cases, the recency portion of th e ser ia l position curve should

be relatively unaffected, while the ear l ier portio ns o f the curve should

show changes. One variable that affects LTS but not STS is the number

of words in th e presented l i s t . As seen in Figure 3c, a word in a longer

l i s t is less l ikely to be recalled, but the recency effect is quite un

affected by l i s t length. Similarly, increases in presentation rate

decrease the l ikelihood of recall ing words prior to the recency region,

but leave the recency effect largely unaffected ( se e Figu re 3d).

In free recal l experiments many l i s t s are usually presented in a

session. I f the subject is asked a t the end of the ses sio n to recal l

a l l the words presented during the session, we would expect his recal l

to reflect LTS re t r ieval only. The probability of recalling words as a

func tion o f thei r ser ia l position within each l i s t can be plotted for

end-of-session recal l and compared with the ser ia l position curve for

recal l immediately following presentation. The results of such an ex_

periment are shown in Figure 4. For th e delayed recal l curve the primacy

effect remains, but as predicted the recency effect is eliminated. In

8

8/3/2019 Atkins on 71


Figure Caption

Figure 4. In addition to a r e c a l l tes t immediately f ol lo wi ng e ac h l i s t

p re se n ta tio n , the s u b j e c t may be asked a t th e end of an experimental

session to r e c a l l a l l th e words from t h a t session. The delayed tes t

should r e f l e c t LTS r e t r i e v a l only. This p re d ic tio nis

verified by the

ser ia l p o s i t i o n curve f o r the delayed tes t : th e increasing recency

e f f e c t is missing. The d a ta are from Fergus Craik.

8a

8/3/2019 Atkins on 71


1.0 r - - _ _ . _ ~ - - : - - . . , _ ~ ~ - ~ ~ ~ _ _ . _ - - - - - ~ _ , _ _ _ _ _ ,

DELAYED RECALL

r IMMEDIATE RECALL

...J

...J

«<.>

w

cr:

IL.

o

>- .5

...J

m«mocr:

a.

O ' - - - ' - ~ ~ ~ - - " " " " ' - - - - ~ ~ - ' - ~ - ~ - ~ ~ - - ' - - - - '5 10 15


8b

8/3/2019 Atkins on 71


summary, the J;'ecency regi.on appears to re fle ct retr ieval from both STS

and LTS, whereas the ser i a l position curve prior to th e recency region

ref lects retr ieval from LTS only.

In 1965 at a conference sponsored by the New York Academy of Sci

ences we put forth a mathematical model explaining these and other effects

in teJ;'ffis of a rehearsal process. I t was postulated that th e subject was

rehearsing a small number of words in STS at a l l times during presenta

t ion, including the item most J;'ecently pJ;'esented. The words s t i l l being

rehearsed in short-teJ;'ffi store when the l as t l i s t item has been presented

were assumed to be output at once (giving rise to the recency effect) .

The t ransfer of information to LTS was assumed to be a func tion o f the

amount of rehearsal g iven e ach item during l i s t presentation. Since th e

words presented f i r s t in the l i s t do not have to share rehearsal with

many o t ~ e r items, they were assumed to J;'eceive addit iona l rehearsa l.

This ex tr a r ehear sa l was supposed to cause more t ransfer of information

to LTS for th e f i r s t items ( th us g iv in g r ise to the primacy effect) .

This rehearsal model was given a fonual mathematical statement and

f i t to a wide array of experiments. The model provided an excellent

quantitative account of a great many results in free recal l , including

those d iscussed in this paper. A more direct confinuat ion of th e model

has J;'ecently been provided by Dewey Rundus a t Stanford University. He

carried out free recal l experiments in which subjects rehearsed aloud

during l i s t presentation. This oveJ;'t rehearsal was tape-recorded, and

compared with the recal l results . The discussion is simplified i.f the

term "rehearsal set" is used to refer to those items overtly rehearsed

between successive presentations of wOJ;'ds. The number of different words

9

8/3/2019 Atkins on 71


contained in th e rehearsal set was found to s tar t at 1 following the f i rs t

word presented and then to rise unt i l th e fourth word; from the fourth

word on the number of different words in th e rehearsal set remained fair ly

constant (a t about 3.3) unt i l the end of th e l i s t (see Figure 5a). The

sUbjects almost always reported the members of th e most recent rehearsal

set when the l i s t ended and recal l began. Some particularly interesting

data from these experiments ar e seen in Figure 5b. The f igure super -

imposes on the ser ia l position curve a curve giving th e mean number of

to ta l rehearsals for items presented in various ser ia l positions. Aclose correspondence is evident between number of rehea rsal s and recal l

probab il ity fo r words prior to th e recency effect ; in the recency region,

however, a sharP disparity occurs.

These findings provide considerable support for the assumption that

LTS storage is a function of the number of rehearsals, and that the

recency effect arises from STS retr ieval rather than LTS. The hypothesis

that storage is a function of the number of rehea rsal s can be checked in

other ways. For example, th e recall probab il ity for a word prior to th e

recency region was plotted as a func tion o f th e number of rehearsals re-

ceived by that word. The result was an almost l inear, sharPly increasing

function. Furthermore, consider words presented in th e middle of the

l i s t that happened to be given the same number of rehearsals as th e f i r s t

item presented. The recal l p robab il it y f or such items was ident ical to

that for the in i t ia l ly presented item.

Having established the efficacy of rehearsal both in storing infor-

mation in LTS and maintaining information in STS, an experiment was

carried out .in which th e subjects ' rehearsal was manipulated direct ly.

10

8/3/2019 Atkins on 71


Figure Caption

Figure 5. (a ) Example of a subject 's rehearsal protocol. A par t ia l

l i s t ing of a SUb ject 's ove rt rehearsa19 ir) an experiment by Dewey Rundus.

Note that the f i r s t word presented ~ given more to ta l rehearsals than

la ter words. (b) Probabili ty of recal l compared with to ta l number of

rehearsals, at eaoh presentat ion pos it ion . P rio r to the recency region,

rehearsals and recal l are c lose ly rel at ed . ThUS, LTS storage depends

on the number of rehearsals given an item, and th e storage differences

appear in LTS ret r ieval .

lOa

8/3/2019 Atkins on 71


I i I T'I r I I '-::-'t I I J I I I I I I i I I 15

®

.......--TOTAL REHEARSALS PER ITEM

en9 ...J

«en

a:7;;';

:I :Wa:

5 ...J

3

13 :aw!::

II a:wa.

19

6-6--.0.-""-,,,--<>-..0......

'"5 7 9 II 13 15 17


. PROBABILITY OF RECALL

3

@

Item 1 . 1.0.Presented Item's Rehearsed (Rehearsal Sell

I. Reaction . Reaction, Reaction, Reaction, Reaction .9

...J

2. Hoof Hoof, Reaction, Hoof, Reaction ...J

«

3. Blessing Blessing, Hoof, Reactionf;:l .8

f- '

Ia:

0 l.L.d 4. Research Research, Reaction, Hoof, Research 0.7

5. Candy Candy, Hoof, Research, Reaction>-

::::i .66; Hardship Hardship, Hoof, Hardship, Hoof lIJ

«

7. Kindness Kindness, candy, Hardship, Hoof I.5

a:

8. Nonsense Nonsense, Kindness, Candy, Hardship Ia.

.4

.3

20. Cellar Cellar, Alcohol, Misery, Cellar

8/3/2019 Atkins on 71


Subjects were t rained to engage in one of two types of rehearsal (see

Figure 6a). In the f i r s t (One-item rehearsal set ) the most recently

presented i tell! was rehearsed exa ct ly t hr ee times before prese.ntation of

the nex t i tem- -no other items were rehearsed. In the second (Three-item

rehearsal set) the subject rehearsed the three most recently presented

items) once each before presentat ion of the next item; th e f i r s t rehearsal

se t contained three rehearsals of the f i r s t word, th e second rehearsal se t

contained two rehearsals of the second word and one rehearsal of th e f i r s t

word, and a l l sUbsequent rehearsal sets contained one rehearsal of each

of the most recent three i tems. The resul ts are shawn in Figure 6b .

When only one item i s rehearsed a t a t ime, each item receives an ident i

ca l number of rehearsa ls . In this case th e primacy ef fec t disappears,

as predicted. Note tha t the recency effect appears for items pr ior to

the l a s t i tem, even though the l a s t Hem is the only one in the l a s t re-

hearsa l se t . This fact indicates that items, even when dropped from

rehearsal, require an addi t ional period of time before they are completely

los t from STS. The curve for the three item rehearsal condition shows

th is e ffe ct also: the l a s t rehearsal se t contains the l a s t th re e p re

sented items and these are r eca ll ed pe rf ect ly . Nevertheless, a recency

ef fec t i s s t i l l seen fo r items pr ior to these three. I t also should be

noted tha t a primacy effect occurs in this condit ion. This ·was predicted

because the f i r s t item in th is condition received a to ta l of 5 rather

than 3 rehearsa ls .

In addition to these resul ts , a delayed reca l l t e s t for a l l words

was given a t the end of the experimental session (s imilar to the proce

dure the resul ts of which are given in Figure 4). The data from this

11

8/3/2019 Atkins on 71


Figure Caption

Figure 6. (a) Schematic outline of rehearsal sequences for two different

rehearsal st rategies given to the subject by the exper imenter. In the

f i r s t only the item currently being presented is rehearsed, and a l l items

receive an equal number of rehearsals. In th e second, the l as t three

items present ed a re r ehea rsed ; in this case the f i r s t two items receive

more rehearsal than the remaining items. Letters in the figure repre-

sent words in the study l i s t . (b ) Results for the two rehearsal schemes.

Note that for immediate recal l primacy disappears when a l l items are

g iven equal rehearsal. For both types of rehearsal, a not iceabl e r e-

cency effect exists for items prior to those in the las t set being

rehearsed. This indicates that items when no longer rehearsed take

SOme additional period before they are lost from STS. The lower curves

give the results from a recal l t es t given at the end of the experimental

session. These curves, which ref lect LTS retr ieval only, closely mirror

the number of rehearsals acco rded items during presentation.

l la

8/3/2019 Atkins on 71


0 ®One-Item Researsal Scheme 1.0

Serial Item Items Total• I THREE-ITEM REH EARSAL

Position Presented Rehearsed Rehearsals Per I tem -+- SCHEME

I A AM 3

I .al-< l - IONE- ITEM

REHEARSAL2 B BBB 33 C CCC 3

--0-- SCHEME

4 D DDD 3

5 E EEE 3

6 F FFF 3 . I .c.J

.c.J

<tUw

14 N NNN 30:: .6

15 0 000 3 IJ..

016 P PPP 3

I \~ I M M E D I A T E-

f-'

IThree-Item Rehearsal Scheme I

I - RECALLf-'

.c.Ja '

.4

lD

Serial Item Items Total0

-"""'\::DELAYEDPosition Presented Rehearsed Rehearsals Per Item a.

" ! l - R ECAL LI A MA 5A \ ... . A , / ~ " X " ' ~ - - ....... ~ ~ ' ' O ' " ' ' ' . ( )B BBA 4

.2 / ... <f" / " "0. '"' . . 0 - -3 C CBA 30 ' ' 0 - - ""tf -"0''''''' 'cY'" '-e.

4 D DCB 3......... 1

5 E mc 3 I-- j

6 F FED 3

0 11 13 154 N NMl 3 I 3 5 7 915 0 ONM 2

SERIAL PRESENTATION POSITION16 P PON I

8/3/2019 Atkins on 71


delayed tes t , which reflect re t r ieval from LTS only, are also given in

Figure 6b (the lower two curves). Note that LTS retr ieval closely par

al lels the number of rehearsals given an item during presentation, for

both rehearsal schemes.

The Structure and Function of th e Short-term Store

These results strongly impl icate rehearsa l in the maintenance of

information in STS and the t ransfer of that information to LTS. The

question then arises: what are the forgetting and t ransfer character

i s t icsof

STS

in the absence o f r ehea rs al ? Attempts to cont ro l rehearsa l

have usually involved a di f f icu l t verbal task such as arithmetic. For

example, Lloyd and.Margaret Peterson at Indiana University (see Short

term memory, Scientific American, July, 1966, 215 (1), 90-95) presented

a three le t te r trigram to be remembered; th e subject next engaged in a

period of arithmetic and then was te ste d f or his trigram memory. Figure

7 i l lustrates the probability of recall as a function of th e duration of

arithmetic. The loss observed over time is similar to that seen in the

recency effect in free recal l . The re asons a re apparent: STS loss caused

by an arithmetic task is similar to STS loss caused by a series of inter

vening words to be remembered. The asymptote of the curve in Figure 7

reflects the retr ieval of the trigram from LTS alone, and the ear l ier

portions of the curve represent re t r ieval from both STS and LTS. The

loss of th e tr ig ram from STS is t hus represented by a decreasing function

prior to the asymptote.

I t has often been assumed tha t th e forgetting observed during arith

metic reflects an automatic STS decay which inevitably occurs in th e

absence of rehearsal. These views have ignored the effect of the

12

8/3/2019 Atkins on 71


Figure Caption

Figure 7. Probabili ty that a three le t te r trigram wil l be recalled

following a p eriod o f arithmetic. The arithmetic is presumed to prevent

rehearsal. Earl ier theories proposed that STS decay in the absence of

rehearsal was represented by the curve in this i l lustra t ion, but they

failed to take into account th e effect of the arithmetic task i t se l f .

The data is based on experiments by Lloyd and Margaret Peterson.

l2a

8/3/2019 Atkins on 71


arithmetic i t se l f . Other theories have t r ied to implicate the inter

vening act ivi ty as the cause of the loss. Indeed, evidence is available

showing that the amount of new material between presentation and t es t on

a given item of information is a much more important determinant of STS

loss than the @count of time between presentat ion and t es t . However,

there are at l eas t two exp lana tions o f this finding. The f i r s t holds

that th e act ivi ty between presentation and t es t is the direct cause of

an i tem's loss from STS. The second explanation proposes that STS in

formation decays a t a fixed rate in time in th e absence o f r ehea rs als ,

but that rehearsals wil l delay the loss. The l a t t e r explanation supposes

that the rate of intervening activi ty wil l affect the number of rehearsals

that can be given to the to_be-remembered item, and thus indirectly de

termine the rate of loss.

I t has recently become possible to choose between these two explan

ations of STS loss. The impetus arose in a thesis by Judith Reitman at

the University of Michigan. She substituted a signal d ete ctio n ta sk fo r

the a ri thmeti c t ask in the Peterson and Peterson paradigm. The signal

detec tion t ask consisted of responding whenever a weak tone was heard in

a continuous background of white noise. Surprisingly, no STS loss was

observed af te r 15 seconds of this task, even though subjects reported no

rehearsal during the signal detection. I f we acc ept th e view that re

hearsal is not occurring, then we could conclude that STS loss is due to

the type of interference during t he i nt ervening interval . Another

important issue which could potent ial ly be r eso lved u si ng th e Reitman

paradigm concerns the t ransfer of information from STS to LTS: does

t ransfer occur only a t i n i t i a l presentation and a t SUbsequent ~ e h e a r s a l s ,

13

8/3/2019 Atkins on 71


or instead throughout th e p erio d that the informa tion resides in STS,

r eg ardle ss o f r eh ea rs als ?

To answer these questions, the following experiment was carried out.

The design is given in Figure 8. A consonant pentagram (SUCh as QJXFK)

was presented for 2.5 seconds for the subject to memorize. This was

followed by a signal detection task in which pure tones were presented

at random intervals in a continuous background of white noise. The sub

jects pressed a key whenever they thought they detected a tone ( this task

was diff icul t--only about three-fourths of the to nes presented were

correctly detected). The signal detection period lasted for e i ther 1,

8, or 40 seconds, with tones occurring on the average every 2.5 seconds.

In conditions 1, 2, and 3 the SUbjects were tested on th e consonant

pentagram immediately following the signal detection. In conditions 4,

5, and 6 the subjects were required to carry out 30 seconds of diff icul t

arithmetic following the s igna l de tec tion before they were tested. To

insure that rehearsal would not occur, subjects were paid for performing

well on s igna l de tec tion , and for accurately performing thei r arithmetic,

but were not paid for l e t t e r memory. I n add iti on , they were instructed

not to rehearse l e t ters during signal detection or arithmetic. When

queried, subjects reported they were not consc ious ly aware of rehea rs ing.

However, because the que stion o f rehearsal is quite important, an addi

t ional experiment was carr ied out. Upon completion of the f i r s t experiment,

SUbjects were run in the same conditions, but with new instruct ions to

rehearse the pentagram aloud following each tone detection.

The pattern of resul ts is depicted in Figure 9. The two lower

curves represent retr ieval from LTS only, since these curve s g ive

14

8/3/2019 Atkins on 71


Figure Caption

Figure 8. The design of a short-term let ter memory experiment in which

arithmetic and signal detection are both used to prevent rehearsal. A

random array of five consonants, such as XKVJZ, is presented for 2.5

seconds. The SUbject then carries out 1, 8, or 40 seconds of signal

detec tion (detect ing qrief tones in a continuous background of white

noise), followed by 0 or 30 seconds of arithmetic, followed by a tes t

for l e t t er memory. In one variation of th e experiment subjects rehearsed

during th e signal detection task, and in th e other they did not.

14a

8/3/2019 Atkins on 71


f-'

0 '

CONDITION

2

3

4

5

6

PRESENTATION OF CONSONANTS

SIGNAL DETECTION TASK

• ARITHMETIC TASK

II] TEST

~ ~ [ T I~ I ? ? ? @ m~ _ I T J~ [!J

~ I & @ l IT]~ IT]II I I ! I II I

Seconds 0 9 18 27 36 45 54 !>3 70

8/3/2019 Atkins on 71


Figure Caption

F:j.gure 9. The r e s u l t s o f th e ex perim en ta l o.esign presenteo. in Figure 8 .

The probability t h a t a l e t t e r w i l l be recalleo. in i ts c or re ct p o si ti on .

The upper curves show t h a t s i g n a l o .e te ct io n , w it h o r without re h e a rsa l,

leaves STS r e l a t i v e l y unaffecteo.. The lower curves give LTS r e t r i e v a l ,

since the long perioo. o f a ri th m et ic causes th e co nte nts o f LTS t o be

l o s t . The lower o.asheo. curve r i s e s throughout the perioo. of s i g n a l o.e-

t e c t i o n because the re h e a rsa l r e s u l t s i n ao.o.itional t r a n s f e r from STS

to LTS. However, th e lower s o l i d curve is h o riz o n ta l over th e las t 32

Secono.s o f s i g n a l o.etection. Thus a tra c e can remain in STS fo r con-

sio.erable perioo.s without s i g n i f i c a n t t r a n s f e r from STS to LTS, as long

as re h e a rsa l is not useo..

l4 c

8/3/2019 Atkins on 71


I \

I \I \I \

('\I \ ...J

I <tUl

I \...JO:<t<t

I \UlW ZO:J: 0

I \<tWWo: ....J: .u

I \WO W

o :Z ....\ --

w0

I \u

...J i= ...J

I<t

\W <l:

...JUl ::::;: zI

<to:

\J: S:2l< t I - C/)

Io:W

\tJ : 0:wW <t IL.

I J :O: \ Ul 0

Iwo

\0

o:z z C/)

I0 0-- \u zW

I

uUl 0

i=

\(, )

W 0 wI ::::;:

\r<'l C/)

I \:

I <t\

Ul

z

I0

\W

I Ul \I

0\

I \

0 Ol (X lI': (l ) 10

- 1 1 " ~ 3 ~ .:W A l . I 1 1 8 " 8 0 ~ d

14d

8/3/2019 Atkins on 71


performance folloWing 30 seconds of arithmetic; we assume that arithmetic

causes the pentagram information to be lost from STS. The top curves

give performance when arithmetic is no t used; they show that the signal

det ec tion t ask alone causes no loss whatever. Since performance is es-

sential ly perfect immediately following signal detection, regardless of

the duration o f s ig nal d etec ti on , i t may be presumed th at th is task does

not produce forgetting from STS for verbal material.

What then does produce forgetting from STS? Information input and

i t s analysis is not by i t se l f enough: the subject is performing a diff icul t

information process ing task dur ing th e signal detec tion period , but no

pentagram forgetting occurs. The same data show that time alone causes

no noticeable forgetting. Yet verbal information (arithmetic) does cause

a large loss. Thus Reitman's conclusion appears to be correct: for-

getting is caused by the entry into STS of other similar information.

Note that th e above arguments depend upon the assumption that re-

hearsal is not occurring during the signal detection period. Evidence

that this is the case is seen i f one compares the two lower curves in

Figure 9. The dotted curve shows that performance improves i f subjects

rehearse overtly during the signal detection period. Presumably the

rehearsal transfers information about the pentagram to LTS. This addi-

t iona l t ran sf er to LTS is reflected in the retr ieval scores, and the

dotted curve r ises. But the lowest curve is horizontal over the las t

32 seconds of signal de tect ion , ind ica ting that no rehearsal was occurc

ring during this period.

The fact that the lowest curve is f la t over the l as t 32 seconds

has important implications for STS to LTS t ransfer . This curve indicates

15

8/3/2019 Atkins on 71


that essential ly no STS to LTS t ransfer occurred dur ing this period. Yet

the top curve shows that the trace remains in STS throughout this period.

Hence the pre sence o f a trace in STS is not alone enough to result in

t ransfer to LTS. Apparently tr an sf er to LTS occurs primarily during or

,

shortly af ter rehearsals. The r ise in th e lowest curve over the f i rs t

8 seconds may indicate that the t ransfer effects of a presen ta ti on o r

rehearsal take a t least a few seconds to reach complet ion. Work along

these l ines is being pursued, and many guestions remain. St i l l this

discussion should indicate how an understanding of control processes,

and experimental manipulation of them, can provide answers to basic

guestions about the structure and function of the memory system.

The emphasis we have given in th e above discussion to rote rehearsal

does not imply that other control processes are of lesser importance.

Although much evidence indicates STS to LTS t ransfer is strongly depen-

dent upon rehearsals, effective LTS retr ieval can be shown to be strongly

dependent upon th e type of information that is rehearsed. The choice of

part icular information to be rehearsed in STS i s a control process called

coding. There are many coding s tr at eg ie s t ha t have dramatic effects on

l o n g - t ~ r m retr ieval . In general, these st rategies consist of adding

appropriately chosen information from LTS to a trace to be remembered,

and then rehearsing th e entire complex in STS. Consider, for example,

th e following verbal mnemonic. Suppose you are given (as is ty pic al in

memory experiments) the st imulus-response pair "HRM - 4"; la ter "HRM"

wil l be presented alone and you wil l be expected to respond "4." I f

you simply rehearse HRM - 4 several times, your abil i ty to respond cor-

rectly l a te r wil l probably not be high. Suppose, however, you notice

16

8/3/2019 Atkins on 71


that HRM reminds you of "homeroom" and you think of various aspects of

your fourth grade classroom. In this case your abil i ty to recal l the

correct response wil l increase markedly.

Numerous experiments have demonstrated th e efficacy of such coding

techniques. Why does such a control process enhance retrieval? Firs t

of a l l , the amount and range of information stored during coding appears

to be greater than that stored when rote rehearsal is used. Secondly,

the coding operation provides a straightforward means by which th e sub

ject can gain access to an appropriate and small region of memory during

retrieval. In the above example, when HRM i s presented at the moment of

tE,st, th e SUbject who has been coding would be lik ely to notice, just as

he did during the in i t i a l presentation, that HRM was simil iar to "home-

room." He could then use "homeroom" (and the current temporal context)

as a further probe and would almost certainly access "fourth grade" and

thence generate the correct response.

There are numerous coding techniques, and many popular books have

been written by professional mnemonists who put forth techniques of th is

sort as "tricks" to aid one's memory. Some of the most effective tech

niques use imaging, a type of coding in which the subject forms visual

images as an aid t o r et ri ev al . Many special ized coding techniques have

been shown to be especially h elp fu l f or part icular tasks. Techniques

have been developed to help l earn pa ired-assoc iates, dates of special

events, commonly used numbers, long l i s t s , speeches, and outlin es fo r

lectures. The interested reader can turn to specialty books for examples

(an historical treatment can be found in The Art of Memory by F. A. Yates,

Chicago: University of Chicago Press, 1966).

17

8/3/2019 Atkins on 71


In addition to coding, imaging and rehearsal, "decision making"

characterizes a class of control processes of great importance. Decision

making usually involves a rule or strategy that enables th e sUbject to

choose between several possible courses o f ac tio n, or several possible

responses. For example , suppose subjects are asked to recal l as many

state names as possible. One sUbject might decide to adopt the strategy

of naming states alphabetically, another might name them in geographic

order, and a third might decide to name them in the order that they f i r s t

come to mind. Considerable differences in performance can result from

such decisions.

Long-term Retrieval and Forgetting

The discussion of coding has indicated how retr ieval may be enhanced

by storing information in such a way that cues will la ter be a va ila ble to

probe LTS accurately. That i s , th e key to retr ieval is the selection of

probe information that will activate an appropriate search-set from LTS.

Since in ou r view LTS is a relat ively permanent store, forgetting is as

sumed to result from an inadequate selection of probe information, and

a r esu lt ing fa ilure of the re t r ieval process. There are two basic ways

in which th e probe selection may prove inadequate. The f i r s t refers to

cases ",here the wrong probe is selected. For example, you might be asked

to name the s ta r of a part icular movie. The name actually begins with T

but you decide that i t begins with A and include "A" in th e probe infor

mation used to access LTS. As a result the correct name may not be

included in th e search-set which is drawn into STS, and retr ieval wil l

not succeed. The second way probe selection can p rove inadequate depends

on th e size of the search-set accessed by th e probe. I f th e probe is

18

8/3/2019 Atkins on 71


such that an extremely large region of memory is accessed, then retr ieval

may fa i l even though th e desired trace is included in the search-set.

Currently research is being carried out to identify the precise

mechanisms that are responsible, but th e effect i t se l f is quite clear.

For example, i f asked to name a f ru i t w hich s ou nd s l ike a word meaning

"to look at" you might say "pear." But i f asked to name a l iving thing

w hich s ou nd s l ike a word meaning "to look at ," the probability of re-

sponding "pear" will be greatly reduced. As another example, suppose

you attend a party with five other people, one named "John Smith." I f

asked la ter to name the people a t the party, you might be lik ely to

remember "John Smith." I f there had been 20 people at the party, recal l

of this name would have been less l ikely. However, besides a failure of

the memory search, other explanations of th is e ff ec t can be proposed.

I t could be argued that more attention was given to "John Smith" at th e

smaller party. Or i f the permanence of LTS is not accepted, i t could be

argued that th e names of th e many other people met at the larger party

cou ld erode or des tr oy th e memory trace for "John Smith." Are these

objections reasonable? To answer this question we note that this example

is analogous to that seen in free recall when l i s t s conta in ing f ive or

20 words are presented for recall . As shown in Figure 3c, words in

longer l i s ts are less well recalled from LTS than words in short l i s ts .

We would l ike to show that the l i s t length effect in free recal l is

dependent upon th e choice of probe information used to access the to-be

recalled l i s t , rather than upon the number of words intervening between

presentation and recal l , or upon differential storage given words in

l i s t s of d if fe ren t s iz e. The second issue is disposed of r athe r easi ly :

19

8/3/2019 Atkins on 71


in many free recall experiments which vary l i s t length th e subjects do \

not know a t the beginning of the l i s t what the length of the l i s t will

be. Thus they cannot reasonably store different amounts of information

for the f i rs t several words in the l i s t s of differing length. N e v e r t h e ~

less the f i rs t several words are recalled a t different levels ( se e F igure

3c). To dispose of the interference explanation (Which, implicates the

number of words between presentation and recall) is more diff icul t . Until

fair ly recen tly in te r fe rence theories o f forge tt ing have been predominant

(see Benton J. Underwood, Forgetting, Scient i f ic American, 1964, 210 (3),

91-99, and John Ceraso, The Interference Theory of Forgetting, Scientif ic

American, 1967, 217 (4), 117-124). In these theories, forgetting has

often been seen as a matter of erosion of the memory trace--usually caused

by items presented following the item to be remembered, but also caused

by items pr ior to the item to be remembered. The l i s t length effect

might be explained in these terms since th e average item in a long l i s t

is preceded and followed by more items than the average item in a short

l i s t . On the other hand, the re t r ieval model presented in this paper

assumes LTS to be permanent and maintains that the strength of long-term

traces is independent of l i s t le ngth . F orgetting resul ts from the fact

that the temporal -contextual probe cues used to access any given l i s t

tend to e l ic i t a larger search-set for longer l i s t s , thereby produoing

less e f fi o ien t r et ri eva l.

In order t o d is ti ngui sh between the retr ieval and interferenoe ex

planations, t he following experiment was carried out. A series of l i s ts

of varying lengths were presented. Fol lowing each l i s t the subject

attempted to free recall no t the l i s t jus t studied (as in the typical

20

8/3/2019 Atkins on 71


free recall procedure) but instead the l i s t preceding the las t l i s t

studied. This procedure eliminates the confounding between the size of

th e l i s t being recalled and the number of words intervening between pre

sentation and recal l . A large or small l i s t to be recalled can be

followed by ei ther a large or small intervening l i s t . The retr ieval

model predicts that recall probabil i ty will be dependent upon th e size

of th e l i s t being recalled, assuming that th e subject has probe infor

mation to access th is l i s t . The interference model predicts th e major

component of performance to be determined by the number of words in the

intervening l i s t . In t he exper iments , l i s t lengths of five and 20 were

alternated randomly in a session.

The results a re dep ic te d in Figure 10. Four conditions are dis

t inguished: 5-5, 5-20 , 20-5, 20-20 where the f i r s t number gives the size

of the l i s t being recalled and the second number gives the size of the

intervening l i s t . The resul ts are plotted as ser ia l position curves

similar to those in Figure 3b . Note that there is no recency effect in

any of the curves; th is would be expected since there is another l i s t

and another recall intervening between presentation and recall . The

intervening act ivi ty causes th e words in the tested l i s t to be lost from

STS; consequently, th e curves represent retr ieval from LTS only. The

results presented in Figure 10. seem qUite clear: words in l i s ts of length

5 are recalled much better than words inl i s t s

of len gth 20 , but the

length of the intervening l i s t has l i t t l e , i f any effect . The retr ieval

model can predict these results only i f a probe is available to access

th e requested l i s t . I t seems l ikely in this experiment that th e subject

has available a t t es t appropriate cues (probably temporal in nature) to

21

8/3/2019 Atkins on 71


Figure Caption

Figure 10. Serial position curves for a free recal l task in which the

subject recal ls the l i s t prior to the one jus t studied. The top two

cu rves g ive the reca l l probabi l i t ies for words in l i s t s of length 5 with

ei ther a 5 or 20 word l i s t intervening between presentation and recal l .

The lower two curves give th e recal l probabili t ies for words in l i s t s of

length 20, with ei ther a 5 or 20 word l i s t intervening between presen-

tation and recal l . The curves give results averaged from those obtained

in a series of three experiments. The results show that a word's proba-

bi l i ty of reca l l is dependent upon the length of l i s t in which i t is

imbedded, and independent of the number of words intervening between

presentation and recal l . The results are predicted by a theory which

postulates LTS forgetting to be a failure of retr ieval .

21a

8/3/2019 Atkins on 71


.4 ...... - . - - - - - - . - - - - - ~ - - - _ _ . . - - - - _ . _ _ _ _ . ,

INTERVENING LIST OF 5 WORDS

INTERVENING LIST OF 20 WORDS

OF 20 WORD

OF 5 WORDS

' \ V - INTERVENING LIST

\

\\

\\

"-

\' ;

IN:RVENING LIST

.3

.1

...J

...J«(. )

wII::

U.

o

>- .2....

...J.

lD

«lD

oII::a.

5 10 15 20


21b

8/3/2019 Atkins on 71


e n a b ~ e him to select probe information pertaining to th e desired ~ i s t .

I f th e experimental procedure were changed so that th e sUbject was asked

to recall the lOth previous l i s t , then s e ~ e c t i o n of an adequate probe

would no longer be possible. The results in Figure 10 demonstrate th e

importance of probe selection, a control process of STS. There are other

control processes which can be used during retr ieval . These include the

strategy by which the search is terminated, the decision to accept a given

image as th e one sought, and th e method by which new probes are generated

when old ones prove unproductive.

Only a brief overview of th e memory system has been presented, but

hopefully i t has clar if ied oUr approach, which integrates the system

around the opera tions of sho rt -t erm store. There is no intent to imply

that the system as described is in any sense a f inal theory. As exper

imental t e c h n i q u ~ s and mathematical models in the memory area have become

increasingly sophisticated, theory has undergone progressive changes.

There is no doubt that th is trend will continue, but we think i t quite

l ikely that th e short-term store and i t s control processes wil l be central

to any future system.

22

8/3/2019 Atkins on 71


References

Atkinson, R. C., & Shiffr in, R. M. Human memory: A proposed system and

i t s control processes. In K. W. Spence and J. T. Spence (Eds.),

The psychology of learning and motivation: Advances in research

and theory, Vol. 2. New York: Academic Press, 1968, 89-195·

Atkinson, R. C., &Wickens, T. D. Human memory and the concept of

reinforcement. In R. Glaser (Ed.), The nature of reinforcement.

Columbus, Ohio: Charles E. Merri l l Books, Inc . , 1971, 66-120.

Rundus, D. An analysis of rehearsal processes in free recal l . Journal

of Experimental Psychology, 1971, in press.

Shiffrin, R. M. Forgetting: Trace erosion or re t r ieval failure? Science,

1970, 168, 1601-1603.

Shiffr in , R. M. Memory search. In Donald A. Norman (Ed.), Models of

human memory. New York: Academic Press, 1970, 375-447.

Documents

Atkins on 71