Professional Documents
Culture Documents
Many group learning curves from ing and retention. The terra "free-recall
rote-learning studies are approxi- verbal learning" describes a type of rote-
learning task in the same way that paired-
mately exponential in form. The associate learning or serial learning describe
present study started with an attempt other types of rote-learning tasks. In free-
to find a rote-learning task such that recall verbal learning retention is measured
the results of individual 5s learning after each presentation by the method of
unaided recall. It is suggested that learning
single lists would be a sufficiently be considered the slope of the curve showing
close approximation to an exponential R as a function of n. Thus, learning would
function so that this function could be the first derivative of R with respect to
be used to describe the data. De- n, dR/dn. Such a distinction between reten-
scribing the learning curves of indi- tion and learning is analogous to the distinc-
tion between position and velocity in me-
vidual 5s by a mathematical function chanics ; j ust as velocity is the rate of change
should not only make possible a more of position with respect to time so learning
exact description of the data but also is the rate of change of retention with respect
facilitate the discovery of inter- to the number of presentations. Such a
definition of learning is of course consistent
relationships among variables. with the idea that is accepted by almost all
It soon became apparent that in psychologists; namely, that learning is an
free-recall verbal learning the data improvement in performance. This defini-
of individual 5s learning single lists tion sidesteps all problems as to why the
could adequately be described by an improvement occurs; in so doing it makes
possible a clear separation between the
exponential function. In free-recall phenomenon itself and the factors on which
verbal learning a list of items is the phenomenon depends.
presented one at a time and after The fact that the curve R = f(«) is ex-
each presentation 5 recalls as many ponential in form means that the slope of the
line, or learning, is proportional to the num-
items as possible. The words may ber of items yet to be recalled; thus, dR/dn
be recalled in any order, and the — b(c — R), The learning becomes slower
order of presentation is randomly and slower with each successive presentation.
varied from trial to trial. The Since learning is considered to be the slope
exponential function that describes of the retention curve (i.e., retention as a
function of number of presentations) the
the data is R = c(l — e~'>n') where R obvious measure of learning is b, the slope
is the number of items recalled after of the straight line obtained by plotting
each presentation, c is an upper limit ln[(c — R)/c~\ as a function of n, where In
or asymptote, b is the rate constant or is log to the base e. Thus, the rate constant
b replaces such traditional measures of learn-
slope of the curve in a semilog plot, ing as number of trials to criterion or number
and n is the number of presentations. of correct responses. Finally, it is assumed
Thus, the exponential function shows that the exponential describes the results of
how R varies as a function of n. all 5s in free-recall verbal learning; individual
differences are assumed to be manifest in
To avoid any confusion it is necessary to differing numerical values of the parameters
discuss briefly the relationship between learn- b and/or c.
1
This article was facilitated by a Faculty The present study reports a number
Summer Research Fellowship from the Uni-
versity of Vermont. The author would like of separate experiments, all of which
to thank Clinton D. Cook for invaluable dealt with the immediate retention
advice at all stages of the project. of unrelated words. Some of the
222
IMMEDIATE RETENTION OF UNRELATED WORDS 223
based on rough averages obtained from the abbreviations already mentioned (R, c, b, n,
individual testing.2 and SC) the following abbreviations will also
To determine if the unequal recall times be used: L (length of list), PT (presentation
for Trials 1-4 would affect retention, a group time/item), t (total time required for presen-
of 22 5s was tested under SC. Then they were tation of the complete list), and RI (number
retested under identical conditions except of words recalled after one presentation).
that they were given the same recall time Further, in any numerical designation of a
(ISO sec.) for all four trials. For total num- specific list the first number indicates L and
ber of words recalled the mean difference the second number indicates PT; thus, 30-2
between the two conditions was less than one means a list of 30 words presented at a rate
word, and the difference did not approach of 2 sec./word.
statistical significance (t = 0.32). Thus, it Experiment I.—The purpose was to deter-
would appear that the unequal recall times mine if a learning-set developed. The 5s
used under SC had little effect on the number were tested under the SC on five successive
of words recalled. sessions each spaced one week apart. A
Since the words were randomly selected total of 29 5s participated in one or more
from the Thorndike-Lorge list it was assumed of the five sessions plus two make-up sessions,
that, except for sampling fluctuations, lists but only 19 5s participated in all five regular
of the same length were always of comparable sessions. In Exp. I and in all other experi-
difficulty. Therefore, within a given experi- ments except where specifically noted the 5s
ment lists were not counterbalanced among were students of both sexes in the intro-
conditions. However, it should be noted that ductory psychology course who served as
most of the conclusions are based on the experimental 5s to fulfill a course requirement.
results of two or more different experiments, No 5 served in more than one experiment.
and different lists were used in different Experiment II.—The purpose was to deter-
experiments. mine if a warm-up or fatigue effect occurred.
As has been mentioned there are no prac- All 5s (N = 14) learned six successive lists
tice effects, either warm-up or learning-how- in one 2-hr, session. The first list was learned
to-learn. Therefore in some (though not all) under the SC. Lists 1, 2, 4, and 5 were
of the experiments to be reported the order followed by a filler task consisting of 10 short-
in which 5s received the various treatments answer questions made up specifically for this
was not counterbalanced. In general the purpose (e.g., "Who is the Governor of
primary consideration was simplicity of Vermont?"). The 5s were allowed IS sec.
design; lists were never counterbalanced, and to answer each question. List 3 was followed
order was counterbalanced only when it by an 8-min. rest period during which time
could be done without increasing the com- 5s were free to leave the room.
plexity of the experimental design. Experiment III.—The purpose was to
Finally, as also will be shown, there is determine the effects of L on Ri. In Exp.
no difference between visual and auditory Ilia the lists were 15-2, 30-2, 45-2, 60-2, and
presentation. Therefore, while visual presen- 7S-2. A different group was used for each
tation was used under the SC, auditory list, and each list was presented once. Au-
presentation was used instead in any experi- ditory presentation was used. One 75-word
ment which required a presentation time of list was recorded on a tape recorder and,
1 sec./word or less. With auditory presenta- for each group, the tape was simply stopped
tion all homonyms were deleted from the after the appropriate number of words had
population of words before the samples were been read. There were 14, 55, 19, 29, and 25
drawn. 5s for the five groups, respectively. The 55
In the following descriptions of specific 5s who were given a 30-2 list were also given
experiments, it is to be understood that in all one trial on a second 30-2 list as soon as the
experiments making use of repeated presen- first list was finished. In Exp. Illb seven
tations the procedural details were as de- additional groups were tested, and each
scribed under SC unless specific mention to group was given one trial on each of two to
the contrary is made. In addition to the four lists. The lists and the order in which
they were presented were as follows: first
2
Although the two experiments using group ( t f - l O ) : 400-1, 5-1, 5-2; second
individual testing are described as Exp. V group (TV = 17): 100-1, 200-2, 6-1, 6-2;
and VI, they were actually the first experi- third group (N = 15): 200-1, 100-2, 8-1, 8-2;
ments to be performed. The numbering of fourth group (N - 24): 45-1, 150-1, 10-1,
the experiments is based on expository con- 10-2; fifth group (N = 30): 75-1, 15-1, 30-1;
venience, not chronology. sixth group (N = 9): 60-1, 60-2, 30-2;
IMMEDIATE RETENTION OF UNRELATED WORDS 225
seventh group (N = 13): 60-1, 150-2, Thus, test list learned the first day was 30-2. There
for Exp. Ilia and Illb combined, there were were 22 5s in Exp. VII.
at a FT of 2 sec./word Lists 5, 6, 8, 10,15,30, Experiment VIII.—The purpose was to
45, 60, 75, 100, 150, and 200 words in length. determine the effect of PT on Ri and b. The
For a PT of 1 sec./word there were all these procedure was identical with that of Exp.
lengths plus a 400-wojd list. In all a total of VII except that the experimental variable
260 5s participated in Exp. III. was PT and not L. The eight lists were
Experiment IV.—-The purpose was to 30-4, 30-f, 30-1, 30-1J, 30-1J, 30-2, 30-3, and
determine the effects of L on RI using the 30-4. The "fast" (1J sec./word or less) and
method of whole presentation. Three lists "slow" (1J sec./word or more) rates were
(25, 50, and 75 words long) were mimeo- counterbalanced as in Exp. VII, and the
graphed. Three different groups of 18 5s first test list learned the first day was 30-2,
each were used. Each group was given one Auditory presentation by a tape recorder
trial on each list, and the order in which the was used. A total of 18 5s participated in
three groups learned the three lists was one or more sessions, but only 13 5s par-
counterbalanced. For each list 5s were ticipated in all four sessions.
given 30 sec. to study the list in any way they Experiment IX.—-The purpose was to
wished, and then they were tested for recall. determine the effect of frequency of usage on
Experiment V.—The purpose was to deter- Ri and b. The procedure was identical with
mine if the results of individual 5s learning that of Exp. VII except that the experimental
single lists in free-recall verbal learning are variable was frequency of usage and not L.
adequately described by an exponential func- Frequency of usage was varied by dividing
tion. There were 18 5s tested individually the 30,000 words of the Thorndike-Lorge list
on a 30-2 list. Both practice and test list into eight subgroups, each subgroup con-
were carried to a criterion of one perfect trial taining approximately the same number of
except that the experimental session was words. The frequency of usage of the eight
limited to 1 hr.; if 5 did not reach criterion subgroups were 100-{-/million, 50-99/million,
on the test list within the allotted time the 10-49/million, 4-9/million, 2-3/million, I/
experiment was terminated. Written recall million, 5-17/18 million, and 4/18 million.
was not used; 5 said aloud all the words he The "high" (4-9/million or more) and the
could recall and E recorded them on a check "low" (2-3/million or less) groups were
list. counterbalanced as in Exp. VII, and the
Experiment VI.—-The purpose was to 100+/million (comparable to the 1,000 most
determine the effect of L on b and c. One 5 common words) was the first test list learned
was tested individually for 22 sessions, and the first day. All lists were 30-2. A total of
at each session a different list was learned. 22 5s participated in one or more sessions,
There were six lists of 20-2, four lists each of but only 20 5s participated in all four sessions.
30-2, 40-2, and 50-2, and two lists each of 75-2 Experiment X.—-The purpose was to
and 100-2. The order in which these 22 lists determine the effect of emotional value on Ri
were learned was approximately counter- and b.3 Each of 50 5s (friends and sorority
balanced. Both practice and test list were sisters of E) rated 30 words from Stagner's
learned to a criterion of one perfect trial. As list (Stagner, 1933) on a 7-point scale of
in Exp. V, E recorded the words recalled emotional value. For each 5 two lists were
on a check list. The 5 was an undergraduate constructed, one of the 12 highest-rated and
psychology major who volunteered for the one of the 12 lowest-rated words. Each 5
experiment, and she was paid for her services. then learned both lists to a criterion of one
Experiment VII,—The purpose was to perfect trial. The 5s were tested individually,
determine the effect of L on b and, in part, and the order in which the two lists was
to replicate Exp. VI with additional 5s. learned was counterbalanced. No practice
This experiment consisted of four sessions. list was used, E recorded the words that 5 was
On each session there was one trial of a 10-2 able to recall, and PT was 3 sec./word.
practice list, four trials of a test list, the filler Experiment XI.—The purpose was to
task described under Exp. II, and then four replicate Exp. II, to determine Ri as a joint
trials of a second test list. The eight test lists function of L and PT, and to study the effects
were 15-2, 20-2, 25-2, 30-2, 35-2, 40-2, 50-2, on .Ri and b of variations in L and PT holding
and 60-2. The order was counterbalanced in 3
that each session one "long" (35 words or This experiment was designed and con-
more) and one "short" (30 words or less) ducted by Cynthia Marvin in a course in
list was used, and the order of long and short experimental psychology taught by the
alternated on successive sessions. The first author.
226 BENNET B. MURDOCK, JR.
total presentation time (t) constant. In w = be — bR where A./? was the
Exp. XIa (Session 1) the first half of Exp. II upper envelope of the curve and R
was replicated exactly except that the order
of the lists was changed from 1,2,3, to 3, 1,2. was the midpoint of each AR interval.
In Exp. Xlb (Session 2) following a practice The mean values for c were 21.0, 30.5,
list RI was determined for 18 different lists. 40.5, 51.2, 80.2, and 101.3 words for
The lists were of three different lengths (IS, the 20, 30, 40, 50, 75, and 100-word
45, and 75 words) and six different presenta-
tion times (?, J, 1, lj, 2, and 3 sec./word). lists, respectively. The coefficient of
The order in which the lists were presented determination (r2) was 99.7%; thus,
was determined by a table of random num- practically all the variability in c
bers. Auditory presentation was used, and can be attributed to L. (b) The same
there was a 2-min. rest after the sixth and least-squares solution was used on the
after the twelfth list. In Exp. XIc (Session 3)
there was a practice list, a 60-1 list, a filler data of each of the 18 SB in Exp. V.
task, a 20-3 list, a filler task, and a 40-1J For all 5s the mean value of c was
list. Auditory presentation was used; for 31.0 words, and this did not differ
all three sessions N = 16. significantly from the expected value
Experiment XII.—The purpose was to
replicate Exp. II and XIa, and to get further
of 30 words (t = 1.03, P > .05).
information on the effects of L on RI. In (c) Bruner, Miller, and Zimmerman
Exp. XI la (Session 1), following a practice (1955) used the method of free-recall
list, 5s were tested for RI after four lists of verbal learning with four lists of 8,
15-2, 30-2, 45-2, and 60-2 each. The order of 16, 32, or 64 words. Their recall
these 16 lists was counterbalanced in blocks
of four, and auditory presentation was used. data are plotted in percentages in
In Exp. Xllb (Session 2), Exp. II and the top half of their Fig. 1 (p. 188).
XIa were replicated exactly except that All groups reached a criterion of at
completely different lists were used. For least 80% correct, and their curves
the two sessions the Na were 16 and 17, suggest that, with a sufficient number
respectively.
of trials, all groups would reach or
come close to 100% recall.
RESULTS AND DISCUSSION
The fact that the asymptote c is the
Instead of presenting the results same as L means that one of the param-
of each experiment separately, the eters in the exponential equation can be
conclusions that have been drawn identified with an experimental variable
will be stated and, following each which is under the control of E. Also,
conclusion, its evidence. This method it greatly simplifies the work of fitting
of presentation will be used because the exponential to the data. Knowing
the value of the length, L, of a list,
in most cases each conclusion is substituting L for c, and writing the
based on more than one experiment. equation in the logarithmic form gives
In some cases corroborative evidence the equation ln[(L — R}/L~} = — bn.
from other studies in the literature Thus, a plot of ln[(L — R)/L~\ against
is available; such evidence will be n gives a straight line with a slope of b
cited where appropriate. and an intercept of zero.
Conclusion 1.-—The numerical value A standard method has been used to
of the asymptote c is the number of solve for &. For SC there are actually
words in the list. Evidence: (a) For five points to determine the curve.
each of the 22 lists learned by 5 in These are the number of words recalled
after 0, 1, 2, 3, and 4 trials (when n = 0
Exp. VI a least-squares solution of then R = 0). It is necessary to use the
the exponential was used which gave least-squares formula for a straight line
a numerical value for both b and c. that passes through the intersection
Specifically, the least-squares solu- of the two axes (Bennett & Franklin,
tion was used on the approximation 1954, p. 232).
IMMEDIATE RETENTION OF UNRELATED WORDS 227
A least-squares solution for the ex- 1.4 words from the best-fitting ex-
ponential in log form does not minimize ponential, (b) For the 18 5s of
the sums of squares of deviations for Exp. V, b was determined in the same
the exponential form. In fact the log manner (i.e., first four trials only),
form introduces a systematic bias, weight-
ing the last trial most heavily and the but the nR.n was computed from the
Oth trial least heavily (Murdock & Cook, total number of trials given each 5.
1960). In an attempt to correct for this The median <rR.n was 1.7 words and
bias, Trials 1, 2, 3, and 4 were weighted the semi-interquartile range was 0.4
by a factor of 4, 3, 2, and 1, respectively. words.
Although this weighting is arbitrary and
does not result in a true least-squares As no mathematical function ever
solution for the exponential it is at least describes experimental data perfectly,
objective and simple to use; also, the the question as to whether a given set
data simply do not warrant a more of data is "adequately" described by a
precise treatment. particular function is essentially a matter
It is to be understood, then, that in all of judgment. In the present case the
the analyses to follow, the parameter c standard error of estimate has been used
in the exponential is assumed to have as a measure of goodness of fit. As has
the numerical value of L. Furthermore, been shown, the deviations between
in all the analyses to follow the numerical the obtained data and the mathe-
value of 6 was determined by the matical function average approximately
weighted least-squares method described 1.5 words; this would seem to be an
in the preceding paragraph. Finally, adequate fit.
in all cases a separate numerical value Another way of testing the appro-
of b was always calculated for each list priateness of a given function is to
learned by each S. Thus, when mean compare its goodness of fit with that
values are presented, they will be mean of other possible functions. In the
b values and not the b obtained from the present case the most reasonable com-
mean trial-by-trial values. parison is with the stochastic model of
Bush and Hosteller (1955) which has
Conclusion 2.—The data for indi- also been suggested for free-recall verbal
vidual 5s learning single lists is learning. For the Bush and Mosteller
adequately described by an exponen- model it is necessary that 5s reach a
tial function. Evidence: (a) There criterion of one perfect trial in order to
were a total of 101 5s who learned obtain an estimate of ai. There were 13
under SC: 29 5s from Exp. I, 14 5s 5s in Exp. V who reached criterion
from Exp. II, 22 5s from Exp. VII, within the allotted time, and for each 5
the necessary parameters of the Bush
20 5s from Exp. IX, and 16 5s from and Mosteller model were determined.
Exp. XI. In all cases 5s were given The value of pt> was determined by the
four trials on a 30-2 list and, except ratio (N — I ) / ( N o — 1) and, for 01
for the practice list, it was the first ?2 was the mean number of nonrecalls.
list they had learned. After the value As a measure of goodness of fit for these
of b had been determined for each 5 13 5s, the median <?R.n for the Bush and
a standard error of estimate, (crB.n) Mosteller model was 2.2 words whereas
was computed for each 5. The for the exponential function the median
distribution of a-R.n for the 101 5s 0R.n was 1.6 words. A t test showed that
was positively skewed with a median the exponential function gave a signifi-
cantly smaller <rjj.n than did the stochas-
of 1.4 words and a semi-interquartile tic model (t = 4.71, P < .01). There-
range of O.S words. Thus, on the fore, insofar as it was possible to make
average, two-thirds of the obtained a comparative test, it would seem that
values did not differ by more than the exponential provides a more accurate
228 BENNET B. MURDOCK, JR.
- —
Conclusion 3.—There is no differ- 15-
_
ence between group testing with
written recall and individual testing 10-
warm-up effect. (c) As a further ship which holds for perceptual recogni-
check, Exp. XI Ib was an additional tion thresholds (Howes & Solomon,
replication of the first half of Exp. II 1951) also holds for learning. The
except that three completely different results are also in essential agreement
30-2 lists were used. An analysis with those of Hall (1954) for free-recall
verbal learning.
of variance showed that the effect
of lists was not statistically significant Conclusion 10.—There is no warm-
(F < 1.00). (d) In Exp. X a number up effect for Rit the number of items
of 5s reached criterion so quickly that recalled after one presentation. Evi-
it was not possible to calculate a dence: (a) There were 55 5s in Exp.
numerical value for b. Instead, trials Ilia who were given one trial on each
to criterion was used as the measure of two successive 30-2 lists. The
of learning and the mean difference mean RI difference was —0.3 words,
between the first and second list was and this difference was not significant
a nonsignificant 0.3 trials (t = 0.83, (t = 0.76, P > .05). (b) In Exp.
P > .05). XI la, 5s were given one trial on each
With the large number of different lists of 16 lists. These 16 lists were
used it was inevitable that some lists divided into four quarters and the
would be appreciably harder or easier differences among the four quarters
than average. The List 2 of Exp. II were not significant (F = 1.44, P > .05).
seemed to be the only list of all those (c) An analysis of variance of. RI for
used that was sufficiently deviant so as the three lists of Exp. XI Ib showed
to result in statistically significant dif- that the differences among lists were
ferences in learning.
not significant (F = 1.14, P > .05).
Conclusion 8.—The reliability of b
is .77. Evidence: (a) The correlation The data of Exp. Xlb cannot be used
coefficient of .77 is based on 76 5s: to test for warm-up because the order
in which the lists were presented was
29 5s from Exp. I, 14 5s from Exp. II, randomized, not counterbalanced. How-
16 5s from Exp. XIa, and 17 5s from ever, in addition to the above evidence
Exp. Xllb. The correlation . was there is an unpublished experiment by
based on the first two 30-2 lists learned A. J. Babick and the author which is
by each 5 except for the 14 5s of Exp. relevant. This experiment tested RI
II. In Exp. II the second list was the for 18 successive 25-1 lists; each of 18 5s
deviant list; therefore, the third 30-2 was tested individually. The 18 lists
list was substituted for the second were divided into six groups of three
30-2 list. lists each, and an analysis of variance
Conclusion 9.—The relationship be- showed that the differences among lists
were not significant (F = 2.24, P > .05).
tween the rate constant b and fre-
quency of usage (FU) as measured by Conclusion 11.—RI = kt + m where
the Thorndike-Lorge word count is / = L X PT. Evidence: The evi-
given by the formula, b = .048 log dence for Conclusion 11 is shown in
FU + 0.31. Evidence: For the data Table 3 which gives the values of r2,
of Exp. IX the coefficient of deter- k, and m for the six relevant experi-
mination (r2) was 66%, and the slope ments (considering Exp. Ilia and III
of the line differed significantly from as separate experiments even though
zero (t = 3.43, df = 6, P < .05). five points from the former are
The linear relationship between b and included in the 25 points of the
log FU suggests that the same relation- latter). One source of evidence for a
IMMEDIATE RETENTION OF UNRELATED WORDS 231
DEESE, J. Serial organization in the recall MURDOCK, B. B., JR., & COOK, C. D. On
of disconnected items. Psychol. Rep., fitting the exponential. Psychol. Rep.,
1957, 3, 577-582. 1960, 6, 63-69.
DEESE, J., & KAUFMAN, R. A. Serial effects PETERS, H. N. The relationship between
in recall of unorganized and sequentially familiarity of words and their memory
organized verbal material. J. exp. Psychol., value. Amer. J. Psychol., 1936, 48, 572-
1957, 54, 180-187. 584.
EDWARDS, A. L. Statistical methods for the SNEDECOR, G. W. Statistical methods (5th
behavioral sciences. New York: Rinehart, ed.) Ames: Iowa State Coll. Press, 1956.
1954. STAGNER, R. Factors influencing the memory
GREENBERG, R., & UNDERWOOD, B. J. value of words in a series. /. exp. Psychol.,
Retention as a function of stage of practice. 1933, 16, 129-137.
J. exp. Psychol, 1950, 40, 452-457. THORNDIKE, E. L., & LORGE, I. The teacher's
HALL, J. F. Learning as a function of word- word book of 30,000 words. New York:
frequency. Amer. J. Psychol., 1954, 67, Teachers College, Columbia Univer., 1944.
138-140. UNDERWOOD, B. J. Psychological research.
New York: Appleton-Century-Crofts, 1957.
HOWES, D. H., & SOLOMON, R. L. Visual WOODWORTH, R. S. Experimental psychology.
duration threshold as a function of word New York: Holt, 1938.
probability. /. exp. Psychol., 1951, 41,
401-410. (Received September 21, 1959)