Word imagery, frequency, and
meaningfulness in short-tenn Inemory*
ALLAN P AlVIO and PADRIC C. SMYTHEt
University of Western Ontario, London, Ont., Canada
Word imagery (I), frequency (F), and meaningfulness (m) were independently
varied in different lists, and the effect of each investigated in the Brown·Peterson
SThl paradigm. Consistent with previous findings, overall correct recall was
much better for high-I words (66%) than for low-I words (39%). Conversely,
low-F words were recalled significantly better than high-F words (65% vs 43%).
The effect of m was not significant. Although performance declined generally
over trials, there was no evidence of any differential PI buildup for words that
were high and ones that were low on any given attribute. Alternative
interpretations were discussed.
This study compared the effects of
noun imagery (I), frequency (F), and
meaningfulness (m) in the distractor
short-term memory (STM) paradigm
(Brown, 1958; Peterson & Peterson,
1959). Borkowski & Eisner (1968)
found that concrete (high imagery)
words were recalled better than were
abstract (low imagery) words in such a
task. They obtained no effect of m
unless it covaried with concreteness.
The present study sought to confirm
their findings and extend the analysis
to include word frequency. A
comparison of imagery and frequency
in STM is especially important because
of the contrasting effects of these
variables in different tasks. Imagery is
consistently positive in its relation to
performance in a variety of learning
and memory tasks (see Paivio, 1969,
1971), whereas frequency is
inconsistent and sometimes clearly
negative in its effect (e.g., Gorman,
1961; Paivio & Madigan, 1970;
Shepard, 1967). Their relative effects
are reversed, however, in perceptual
recognition, where frequency is
positively related and imagery is
unrelated to ease of tachistoscopic
recognition (Paivio & O'Neill, 1970).
Thus, imagery is facilitative in tasks
involving learning and considerable
memory load but not in immediate
perception, whereas frequency
facilitates perception but its effect is
negative in certain memory situations.
The use of an STM paradigm is
pertinent to such findings because, in
terms of such factors as memory load,
it can be regarded as falling somewhere
between perceptual recognition and
the learning and memory tasks :n
which I and F have been most ofte.~
*This research was supported by grants to
the first author from the National Research
Council of Canada (AP A-S 7) and the
University of Western Ontario Research
Fund. The artwork and photography for
Fig. 1 were through the courtesy of the
London Psychiatric Hospital.
tNow at the Board of Education,
London, Ont.
Psychon. Sci., 1971, Vol. 22 (6)
investigated. That is, the number of
words presented on a single trial in the
Brown-Peterson task is typically below
immediate memory span. The task is
also relevant theoretically because of
its apparent sensitivity to proactive
interference (PI) effects. Interference
has been suggested as the basis of both
the negative effect of frequency in
certain tasks (e.g., Gorman, 1961;
Shepard, 1967) and the inferior recall
of abstract as compared to
high-imagery words in STM (Wickens
& Engle, 1970). To the extent that
these interpretations are correct, the
con trasting positive and negative
effects of I and F should be revealed
clearly in STM as a differential
performance decline following the first
trial. That is, PI should be greater for
low-I than for high-I items, and for
high-F than for low-F items.
The above analysis is also applicable
to m, since high- and low-m words are
operationally distinguished in terms of
number of associations, which has
been proposed as a basis for
differential interference via common
mediating associations, Le., associative
overlap (e.g., Underwood & Schulz,
1960). Borkowski & Eisner (1968)
found no evidence that m was
effective independent of concreteness,
however, and we obtained similar
results in a modified (unpublished)
replication of Experiment 2 in
Borkowski and Eisner. In the
replication, m and I were varied within
different sets of items, each comprised
of four words. The retention interval
was 18 sec. While recall was
considerably greater for high-I than for
low-I words (72% as compared to 41 %
correct over all trials), the effect of m
did not approach significance. Two
additional features of the data were
in t eresting. First, there was no
evidence of any differential buildup of
PI for high-I and low-I words, since
imagery did not interact with trials.
Second, low-m words tended to be
somewhat better recalled than high-m
words, suggesting a possible negative
e f fee 1 w hi c h warranted further
investigation.
METHOD
The design of the experiment was
patterned after Borkowski & Eisner
(1968) in that a given item attribute
was varied within, a four-word set
presented on each of six trials.
However, only one attribute was
varied for a given group of Ss. Thus
one group was presented sets of items
comprised of high-I and low-I nouns
that were equated on m and F;
another had m varied with I and F
controlled; a third had F varied with I
and m controlled. A constant 18-sec
retention interval was used, since
Borkowski & Eisner (1968) found that
item attribute and retention interval
did not interact.
Items
The items were nouns selected from
the Paivio, Yuille, & Madigan (1968)
norms. For each of the three variables,
24 nouns were selected such that 12
nouns were high and 12 were low on
the critical independent variable while
the other two attributes were
controlled at a medium value. For the
imagery-varied sets, the mean attribute
values for high-I and low-I nouns,
respectively, were 6.33 and 2.97 for I,
5.78 and 5.53 for m, and 34.92 and
39.25 for F (where AA and A
Thorndike and Lorge words are
assigned values of 100 and 50). For
the m-varied sets, high- and low-m
words had mean values of 6.56 and
4.70 on m, 4.79 and 4.51 on I, and
27_50 and 23.58 on F. The high- and
low-F words had mean values of 7.50
and 2.25 on F, 4.65 and 4.66 on I, and
5.68 and 5.68 on m. Mean word length
was held approximately constant over
the different lists. The items were
typed on cards in sets of four, such
that each quadruplet contained two
words of each level of a given variable,
alternating in the sequence of high,
low, high, low (HLHL) or LHLH. An
example of HLHL imagery quadruplet
is BOULDER, ECONOMY, HOUND,
ATTITUDE; a comparable m-varied
set is DRAMA, FATIGUE, PANIC,
HATRED; and an F-varied set is
POWER, TIDBIT, VOLUME,
REFLEX.
Subjects and Procedure
The Ss were 60 volunteers from
introductory psychology classes, 20 Ss
(11 men) being assigned randomly to
each of the three experimental lists.
Each quadruplet was presented for
3 sec, followed by an 18-sec retention
interval, during which S counted
forward from a three-digit number.
This was followed by a recall interval
of 4 sec. One-half of the Ss in each list
condition were presented the
quadruplets in a HLHL sequence, the
other half in a LHLH sequence. The
order of the six quadruplets in a given
333
~o
r
,. '.' Ill" i 11
1 " 11 a b I (>
i r g r a lui t Oll S
('xplanati o lls or th e imagery an d
\
rrequ e ncy e ('fects.
\
A further possibility is that Ss '
....
~
\
rehearsal strategies systematically
\
~60
\
...........
AlOF
\
....
favored particular items within a set.
Him
\
\
a: SO
\
Q,
No relevant evidence is available from
\
this study, but in a related companion
~A,o,
40
_ Hlr
I~
experiment, to be reported separately,
lo
'
y
...........
lO[
.! 30
which involved homogeneous triads of
~
pictures, concrete words, and abstract
20
words, we examined the counting task
~
. f
<. I I ,
.
To
for possible differences in performance
Z 3 4 5 6
I
2 3 4 5 6
that might suggest differential
2 3 4 5 6
rehearsal as a function of concreteness.
T R
A l
S
No such evidence was obtained,
Fig. 1. Percentage of items recalled correctly on each trial as a function of although imagery-concreteness again
high and low levels of imagery, meaningfulness, and frequency.
had a positive effect.
Finally, it may be that the absence
of differential PI occurs only when
list were also systematically varied for effect of m was not significant, and item sets are mixed with respect to
different Ss, thereby controlling for the positive direction of the trend levels of the independent variable, as
the effect of any particular quadruplet contrasted with the negative trend in the Borkowski & Eisner (1968)
over trials.
noted in the introductory experiment. experiments and the present one.
RESULTS
Thus m obviously has no reliable Using a design in which one group of
The results for the six trials for high effect in this task. The effect of Ss received only high-I items and
and low attribute levels are plotted for frequency was clearly negative, which another group only low-I items,
each variable in Fig. 1. It can be seen agrees with other studies (see the Wickens & Engle (1970) and Paivio &
that high-I words were recalled better introduction) in which rare words Begg (1970) found evidence of PI
than low-I words; high-m were were found to be better remembered differences, such that high-l and low-I
somewhat better than low m; and than frequent words. These findings items did not differ on the first trial,
low-F were clearly superior to high-F. appear to be consistent with the view but the decline on subseq uent trials
The results for each variable were that low-imagery (abstract) words and was relatively greater for low-I items.
analyzed by a 2 by 2 by 6 analysis of high-frequency words, relative to their On the basis of their findings for
variance, with word sequence (HLHL respective high- and low-value imagery, Wickens & Engle (1970)
vs LHLH), attribute level, and trials as counterparts, are particularly suggested that semantic overlap was
factors. None of the main effects or susceptible to interference from other greater among abstract than among
interactions involving sequence items. Note, however, that the effect concrete items, and that this was the
approached significance, so this began with the first set of items in basis of differential PI for the two
variable will not be discussed further. each case and was not reflected in any types of items. The explanation is
The analysis of the I-varied differential PI buildup over trials, untenable, however, inasmuch as
experiment showed that the positive su ggest i ng that any differential P a i vio & Begg ( 1970) obtained
effect of I was highly significant, interfering effect must have occurred superior recall of high-imagery items
F( 1 ,18) '" 45_64, p < .001. Overall within, rather than between even with overlap controlled. A more
correct recall was 66% for high-I words quadruplets_ The possible sources of likely alternative is that imagery
as compared to 39% for low-! words. such interference include protects high-I items from any
The trials effect, F(5,90) = 3.58, (1) associative or semantic intraverbal interference that might
p < .01, indicated the typical relationships among the items within a develop if only (or mainly) the verbal
performance decline over trials, but quadruplet, (2) extralist intrusions, code were available, as in the case of
the Imagery by Trials interaction was and (3) differential retroactive abstract nouns. Alternatively, imagery
nonsignificant, F(5,90) = 0.24. The interference from the distractor (i.e., may provide an additional memory
analysis of the m-varied list results counting) task. The first of these code that facilitates the retrieval of
yielded a significant trials effect, alternatives is logically unlikely, since high-I items as units in the
F(5,90) =2.64, p < .05, but within-set associative relationships Brown-Peterson paradigm, although it
nonsignificant effects for m, should facilitate rather than hinder apparently does not facilitate retrieval
F(1,18) = 3.76, and the m by Trials recall through better organization or of order information in STM, as
interaction, F( 5,90) = 0.33 . For the chunking of items that are evidenced by the absence of an
F-varied lists, the negative effect of F semantically or associatively related. imagery effect on immediate memory
was
highly
significant, Such facilitation in STM has been span (Paivio & Csapo, 1969).
F(1,18)= 17.89, p < .001, overall demonstrated by Paivio & Begg
recall being 65% as compared to 43% (1970). Thus, if interitem associations
for low-F and high-F words, indeed were stronger among low-I than
REFERENCES
respectively. The decline over trials among high-I items, or among high-F BORKOWSKI, J. G., & EISNER. H_ C.
Meaningfulness and abstractness in
was also significant, F(5,90) = 4.06, and low-F items, it should have been
short-term memory. J o urnal of
P < .05, but the interaction of trials reflected in better recall of low-I and
Experimental Psychology. 1968, 76,
57-61.
and attribute was again nonsignificant, high-F items, at least on the first trial,
J. Some tests of the decay theory
F = 0.88.
rather than the poorer recall that was BROWN.
of immediate memory. Quarterly Journal
DISCUSSION
observed for such words. No evidence
of Experimental Psychology. 1958, 10,
Th e strong positive effect of is available from this study or any
12-2l.
imagery replicated Borkowski & Eisner other one known to us on differential GORMAN. A. M. Recognition memory for
nouns as a function of abstractness and
(1968) and the exploratory study
interference from extralist items or
frequency . J o urnal of Experimental
mentioned in the introduction . The
from the counting task, so these
Psyc hology . 1961.61. 23-29 .
0
__ 70
~~I
... ,
334
,
\~/"
Psychon. Sci., 1971, Vol. 22 (6)
1':\1\ 0100
0\
\ 1 . 1i ~ , d
; ; : 1,;;':'1")" '0 HI
d:-'~ Ol" i ~t \ i\ O l'
I . :arn l ng
, 1 11(..1
Illt ' 1l1 0l"\°0
PS\ c h o )ogic:al
Hnil'\\", 1 ~HHL 711, :.!-l1-2G::!.
PAIVIO. A. InHtRer), and p(>l'hai nn) c('sscs.
New York: Holt, Rin~hart an d Winston,
197L
PAIVIO, A . . & BEGG, I. Imagery and
associative overlap in short-term memory .
Research Bulletin No . 163, Department
of Psychology , University of Western
Ontario, 1970.
PAIVIO, 1\., & CSAPO, K. Concrete-image
and verbal memory codes. Journal of
Experimental Psychology, 1969, 80,
279-285.
PAIVIO, A., & MADIGAN, S . Noun
imagery and frequency in paired-associate
and free recall learning. Canadian Journal
of Psychology, 1970, 24, 353-361.
PAIVIO, A. , & O'NEILL, B. J. Visual
recognition thresholds and dimensiOns of
word
meaning.
Perception &
Psychophysics, 1970,8,273-275 .
PAIVIO, A., YUILLE, J C., & MADIGAN,
S.
Concreteness. imagery, and
meaningfulness values for 925 nouns.
Journal of Experimental Psychology
Monograph Supplement, 1968, 76(1,
Pt. 2).
PETERSON, L. R., & PETERSON, M. J.
Short-term retention of individual verbal
items. Journal of Experimental
Psychology, 1959,58,193-198
SHEPARD, R . N. Recognition memory for
words, sentences, and pictures. Journal of
Verbal Leaming & Verbal Behavior,
1967,6,156-163.
UNDERWOOD, B. J .• & SCHULZ. R. W.
Mea n ingfulness and
verbal learning.
Chicago: Lippincott. 1960.
WICKENS. D. D., & ENGLE, R. W. imagery
and abstractness in short-term memory.
Journal of Experimental Psychology.
1970.84. 268-272.
Psychon . Sci., 1971 , Vol. 22 (6)
The effects of test events and
spaced practice on free recall learning
ESTHER U. COKE
Bell Telephone Laboratories, Murray Hill, N.J. 07974
Two types of events, testing and a task unrelated to learning, were substituted
for some practice trials during free recall learning. Three practice trials spaced by
either of the substitute activities produced as much learning as seven massed
practice trials. Statistically significant increases in recall over three criterion tests
were not accompanied by increases in the stability of the recall sequence over
successive criterion tests_ The results of this experiment suggest that
improvement in free recall performance due to repeated tests may be in part due
to spaced practice.
Learning in an individual S is
frequently measured by test events
interspersed throughout practice. This
method of evaluating the learning
process assumes that testing does not
influence subsequent behavior. The
results of a number of studies
(Rothkopf & Coke, 1963; Rothkopf,
1966; Lachman & Laughery, 1968)
indicate that this assumption is not
always valid.
Tulving (1967) found that test trials
could influence recall performance. He
substituted two tests for two out of
every three practice trials during free
recall learning. This reduction in
exposure to a word list did not affect
the amount of material recalled.
Tulving suggested that tests were as
effective as practice because S was able
to organize the material into larger
memory units during a test.
The test trials in Tulving's
experiment also interposed a time
interval between successive practice
trials. This spacing of practice might
account for Tulving's results. The
distribution of practice trials has been
shown to affect the relationship
between the number of exposures to
material and its acquisition
(Underwood, 1962 ; Rothkopf & Coke,
1963).
In the present experiment, test trials
or an activity of equal duration, but
unrelated to learning, replaced some
practice trials during free recall
learning. According to Tulving's
hypothesis about subjective
organization, test trials may be as
effective as practice trials in increasing
recall. Further, if Tulving's
characterization of the problem is
correct, the substitution of an
100
oJ
oJ
80
c
u
w
CIt
...
SO
0
~
~
oJ
.40
CD
c
III
0
CIt
Q.
zJ
-•
P - CONDITION
T- CONDITION
,....--~
R - CONDITION
....... --.o
U- CONDITION
~
O~'~~------~------~~
CRT (I)
CRT(Z)
CRTIl)
CRIHRION
R[CA~~
HST
Fig.!. The probability of recalling a
word from the list as a function of the
criterion recall test for each of the
four experimental conditions.
335
© Copyright 2026 Paperzz