- Email: [email protected]
Effects upon verbal learning of stimulus similarity, number of stimuli per response, and concept formation
JOURNAL OF VERBAL LEARNING AND VERBAL BEHAVIOR 1,
470-476 (1963)
Effects upon Verbal Learning of Stimulus Similarity, Number of Stimuli per Response, and Concept Formation 1 T I M O T H Y A . S M I T H , 2 LYLE V.
JONES, AND
STUART THOMAS
University o/North Carolina, Chapel Hill, North Carolina From the moment a child begins to speak he is confronted with the task of learning the names of objects. The child soon discovers that some objects are grouped together under one name ("these round, red objects are apples"), while for other objects, each has a name of its own ("this red-faced lady is Aunt Minnie"). This paper reports two experiments designed to study the learning of names for objects with varying degrees of uniqueness of association. Several recent studies have investigated the learning of verbal responses associated with sets of stimuli. These studies may be viewed as investigations of paired-associate learning with the number of stimuli associated with a response being varied. Bricker (1955), after varying the number of stimuli per response and correcting performance scores for the effects of guessing, reported that the number of stimuli per response had 1 Part of this study (Experiment I) comprised the unpublished Master's thesis of the first author prepared in 1961 under the direction of the second author. The work was performed while Timothy Smith was a National Science Foundation Graduate Fellow. Stuart Thomas collaborated on Experiment II while a participant in the Undergraduate Research Program of the National Science Foundation in 1962. The study was partially supported by Grant M-1849, National Institutes of Health. We are grateful to Dr. Douwe Yntema and to Dr. Edward Zigler for constructive criticisms of an earlier draft. 2 Now at Florida State University, Tal]ahassee, Florida. 470
little effect on learning to associate verbal responses to stimuli. Hake and Eriksen (1955), Richardson (1958), and Metzger (1958) failed to find a relationship between the number of stimuli per response and learning to make the appropriate response to stimuli. But Yntema and Mueser (1960), utilizing a more complex experimental situation, have presented some evidence that number of stimuli per response can be a determinant of performance in associative learning. Their study focuses upon keeping track of the present states of a set of attributes, the states of which are subject to change during the experimental session. When the same state (response) is a possible value for several attributes (stimuli), performance is substantially poorer than when each state (response) is uniquely associated with an attribute (stimulus). These studies suggest that further investigation is needed into the effects of varying the numbers of stimuli per response upon paired-associate learning. A relatively unexplored facet of pairedassociate learning is the effect of assigning to homogeneous stimuli a common response. I t is one thing to assign to a particular class of round, red objects that are good to e a t the name, apple. However, something quite different is involved when we assign to the noise that a dog makes and to the outer covering of a tree, the common name, bark. When stimuli which are assigned a common
VERBAL LEARNING AND CONCEPT FORMATION name
exhibit
some common
property,
learn-
ing to pair the responses with the stimuli can be
viewed
When
as
groups
a of
concept-formation stimuli
signed to responses, stimuli-response rote
equivocal, indicate
versus
the
Of central the
concept
sponse
on
both
of r o t e
a
are
(1958) concept
s u p e r i o r to t h a t f o u n d
situation.
concern
learning
of
asthe
findings
learning
of Richardson
in the present
effect of the number
associate
(1958)
availability
r e s u l t s in p e r f o r m a n c e in a rote-learning
is one
Metzger's
the results that
task.
randomly
the task of learning
relationships
learning. Although concerning
are
rote at
and two
of stimuli concept
study per
is re-
paired-
levels of stimulus
similarity. EXPERIMENT I
Method Materials. Stimuli consisted of two groups of eight color "caps," selected from the F a r n s w o r t h Munsell 100-Hue test for color discrimination (Farnsworth, 1957). These caps are circular color patches, each m o u n t e d on a plastic cylinder, 10/16 in. in diameter a n d 7/16 in. in height. T h e colors were chosen from the Munsell color system so as to be identical in brightness a n d saturation b u t differe n t in hue. T h e colors are designed so t h a t caps with adjacent numbers, say 19, 20, a n d 21, differ in hue by equal-appearing steps. One set of eight color caps used in the experiment consisted of caps one step apart, specifically n u m b e r s 17, 18, 19, 20, 21, 22, 23, a n d 24. T h e other set was m a d e up of caps five steps apart, n u m b e r s 2, 7, 12, 17, 22, 27, 32, a n d 37. Both sets represent a yellowgreen range of hues. T h e selection of hues provides variation of stimulus similarity, the first set being more similar in hue according to the Munsell color system. Responses were eight nonsense syllables f r o m Glaze's list (1928) selected at r a n d o m with the following restrictions: (a) the association value of each, as reported by Glaze, is less t h a n 50%; (b) the first consonant of each syllable is distinct; (c) no syllables begin or end with x, q, e, or end with h or w (to assure familiar p h o n o l o g y ) ; (d) no syllable begins with y or g (to prevent phonemic association with yellow or g r e e n ) ; (e) the vowels a, e, i, a n d u each appear in two syllables. T h e syllables actually d r a w n were W E Z , KAG, H I F , LEB, NAF, M I P , ZUG, a n d VUT. To provide variation of the n u m b e r of stimuli per response,
471
in some conditions one color cap was assigned to each nonsense syllable, in others two caps to a syllable, a n d in others four caps to a syllable. Concept formation was m a d e possible in the conditions with m o r e t h a n one cap to a syllable by assigning caps adjacent on the Munsell color system to one syllable. T h e rote-learning task was created by randomly assigning n o n a d j a c e n t caps to a syllable. Design. Ss were 60 undergraduate m e n a n d w o m e n enrolled in introductory psychology at the University of N o r t h Carolina a n d at Duke University, r a n d o m l y assigned to 10 experimental groups, each containing 6 Ss. Groups were distinguished by different combinations of treatments: (a) rote learning or concept formation; (b) stimuli of high or low similarity; a n d (c) one, two, or four stimuli (color caps) per response (nonsense syllable). If each t r e a t m e n t h a d been completely crossed with each other treatment, a 2 X 2 X 3 factorial design would have resulted, with 12 experimental groups. However, with only one stimulus per response there is no distinction between rote learning and concept formation, a n d 2 of the 12 groups become extraneous, leaving only 10 distinct groups. Procedure. Lighting was controlled by the use of a darkened room a n d a M a c b e t h Daylight lamp type ALS30. U p o n entering the experimental room S was given a list of eight, f o u r , or two nonsense syllables a n d told he was participating in a s t u d y of the learning of n a m e s for colors. The E then placed the eight color caps one at a time before S a n d told h i m the nonsense syllable which was to serve as the n a m e of each. There followed 100 learning trials, with E m a n u a l l y placing one color cap before S on each trial; S was instructed to produce its n a m e orally, guessing if necessary, b u t responding on every trial. T h e S was allowed as m u c h time as he wished to respond, the color cap resting on an 8 X 11 in. piece of white paper in full view the whole time. The other seven color caps were concealed behind a cardboard screen. Caps were presented in the same r a n d o m order for all conditions. This order was determined by consulting a table of r a n d o m n u m b e r s so t h a t not only were the colors presented in a mixed sequence b u t one color m i g h t be repeated before all the other seven were presented. T h e nature of such a r a n d o m order was explained to S prior to the experiment. Immediately after each of his responses while the stimulus was still in view S was told the color cap's correct n a m e by E. After every 25 trials S was given a s u m of m o n e y based on the n u m b e r of correct answers he h a d achieved in the previous 25-trial block. T h e a m o u n t given for each correct answer was varied from condition to condition so that the average a m o u n t awarded Ss in each condition
472
SMITH~ JONES, AND THOMAS
approximated $1.00 over the total 100 trials. The S was allowed to refer to the printed list of nonsense syllables throughout the experiment, thus eliminating any need for learning the set of possible responses.
9O
Results
7O
The dependent variable was the score X where R is number of correct responses, W is number of wrong responses and n is the number of distinct response categories available. The possible values of n are 8, 4, and 2, depending upon whether 1, 2, or 4 color caps are associated with a given name. Under the assumption that all erroneous responses represent guessing and that the probability of a "correct" guess is I/n, X represents the number of correct responses other than those expected to be guessed correctly. Mean scores for all groups are plotted in Fig. 1. Three analyses of variance were performed: (a) to evaluate effects of stimulus similarity and number of stimuli per response with the rote-learning conditions (the six means represented by circles in Fig. 1); (b) to evaluate effects of stimulus similarity and number of stimuli per response within the concept-formation conditions (the six means represented by triangles in Fig. 1); (c) for conditions with more than one stimulus per response, all three effects were included in one analysis--stimulus similarity, number of stimuli per response, and rote versus concept learning [the eight means for two and four stimuli per response (Fig. 1)].
R--W/(n--1)
Rote Learning. Differences significant beyond the .001 level appear between stimulussimilarity conditions (F ~ 16.30, dj ~1/30), and between differing numbers of stimuli per response (F ~ 10.06, dj ~ 2/30) for rote learning. Their interaction does not reach the .05 significance level (F ~---2.30). Performance clearly is superior for the low stimulus-similarity condition. Under low stimulus similiarity, mean performance scores are 54.45, 31.56, and 19.33 for one, two, and four stimuli per response, respectively. Cor-
80
60
.....~ //Z~
5O
~//////
40
iiiiI ~
20
"~.. •x
/
//
.
"O
I0
Number of stimuli per response FIG. 1.
Mean performance scores (X)
for rote
learning (circles) versus concept learning (triangles), contrasting similar-stimulus conditions (dashed lines) with dissimilar-stimulus conditions (solid lines) for varying numbers of stimuli per response (Experiment I). responding means for high stimulus similarity are 25.14, 13.11, and 13.33. Further, for rote learning, performance tends to deteriorate as a function of increasing numbers of stimuli per response, for stimuli of both high and low similarity. These differences are portrayed graphically as the two curves connecting circles in Fig. 1.
Concept Formation. Both stimulus similarity and number of stimuli per response significantly affect performance in the concept-formation conditions. Both of these main-effect F-values (16.20, df = 1/30; 46.34, df ~-~2/30) are significant beyond the .001 level. Again their interaction is not significant (F = 2.65). As was the case for rote learning, the low stimulus-similarity
VERBAL LEARNING AND CONCEPT FORMATION conditions show superiority over the high stimulus-similarity conditions (Fig. 1), having means of 54.45, 81.1I, and 89.33 for one, two, and four stimuli per response, respectively. T h e corresponding high stimulussimilarity means are 25.14, 71.78, and 79.33. I n direct contrast to the finding for rote learning, performance is enhanced with increasing numbers of stimuli per response when concept formation is a possible aid to learning.
Rote Learning versus Concept Formation. This analysis was performed in order to make a comparison between the rote-learning and concept-formation conditions. Excluded are data obtained under the condition of only one stimulus per response. The results of the analysis show significant differences only between stimulus-similarity conditions ( F ~ 8.45, df ~ 1/40, P ~ .01) and between concept-formation and rote-learning conditions ( F ~ 262.99, d / ~ 1/40, P < .001). Appraisal of the subgroup means (Fig. 1, ignoring data for one stimulus per response) is an aid to the interpretation of these differences. Consistent with results of the earlier analyses, performance is better for conditions of low stimulus similarity than high stimulus similarity. I n addition, there is evidence that, when a single response applies to a group of two or four stimuli, performance is greatly enhanced if the group consists of homogeneous stimuli (conceptformation condition) rather than heterogeneous stimuli (rote-learning condition). EXPERIMENT I I
Meth~od It will be noted that Experiment I failed to centrol response time, for S was allowed as long as he wished to respond. In an attempt to assess the importance of this factor, another experiment was performed. Twenty-four undergraduates enrolled in introductory psychology at the University of North Carolina were the Ss. The materials and lighting were the same as in the previous experiment, with the exception that a black shutter operated by a small motor was used to regulate the exposure time
473
of the color caps. It was placed on a table top in a position intervening between S and the sheet of white paper where the color cap was placed manually by E. With the shutter closed, S was unable to view the cap. Two variations of response time were studied. In forced-response conditions an electronic timer raised the shutter and exposed the stimulus for 2 sec., after which S was given 2 sec. during which to respond. The color cap was then again shown to S for 2 sec. while he was told its correct name by E. In free-response conditions the color caps were exposed by raising the shutter, but no limit was placed either on the stimulus-viewing or response time, the shutter remaining open until S responded, as in Experiment I. The response time for each S on every trial was recorded. As in Experiment I, S was told the color cap's correct name immediately following each of his responses. In Experiment II, rote learning exclusively was studied, with stimuli of low similarity. Conditions of one and four stimuli per response were considered. Instructions and incidental experimental procedure were the same as in the main experiment, including the feature of payment for correct answers.
Results Again the dependent variable was the score 1). The analysis of variance for the four groups shows significant effects for response time ( F ~ 6.54, d / ~ 1/20, P < .05), n u m b e r of stimuli per response ( F ~ 13.95, d / ~ 1/20, P < .01), and their interaction ( F ~ 6.54, d / ~ 1/20, P < .05). As was found for rote learning in Experiment I, performance deteriorates with increasing numbers of stimuli per response. W i t h free response, the means for one and four stimuli per response were 52.40 and 20.00, respectively, almost identical to the means for the corresponding conditions in Experiment I. W i t h forced response, the means were 26.07 and 20.00 for one and four stimuli per response, respectively. The difference between the one- and four-stimuli-per-response conditions is considerably more marked in freethan forced-response conditions; the forcedresponse mean difference is not significant ( t ~ .814, d ] ~ 10, P > .40). This is the source of the significant interaction. When allowed free response time, tile mean time
X -~ R - - W / ( n - -
474
SMITtt, JONES, AND THOMAS
for response was 6.74 sec. for one stimulus per response and 3.42 sec. for four stimuli per response. DISCUSSION
In this study, performance on a pairedassociate rote-learning task is found to be better for stimuli of low similarity than for highly similar stimuli. This result is consistent with the findings of Gibson (1942), Underwood (1951), and others. A plausible interpretation of this effect, first advanced by Gibson (1940), is based on stimulus generalization. For highly similar stimuli a greater degree of generalization, and hence a higher rate of errors is anticipated than for stimuli of lower similarity. That is precisely what is found here. Stimulus generalization can also be used to explain the difference in performance between rote-learning and concept-formation conditions, when more than one stimulus is assigned to each response. In the conceptformation conditions, where stimuli adjacent on a dimension of stimulus similarity were assigned the same response, responses which are determined by stimulus generalization often are "correct" responses, as defined by E. In the rote-learning conditions, different responses are more likely to be associated with adjacent stimuli and stimulus generalization would interfere with correct performance. In view of the large consistent difference found between rote- and concept-learning conditions in this study (Fig. 1), it is reasonable to consider differences between this study and the similar ones of Metzger (1958) and Richardson (1958), where findings pertinent to this issue were considerably less striking. Three possible critical differences should be noted. First, the present design included 100 trials for each S, as compared with 200 and 400 trials in the studies of Metzger and Richardson, respectively.
The initial advantage that concept formation has over rote learning may disappear as learning progresses. Second, Ss in the present study were paid for correct responses. This pay probably served to maintain the S's attention to the task. As a result, correct responses accidentally discovered (on the basis of stimulus generalization) might then be incorporated into the S's repertory faster than if no payments had been made. Third, both Metzger and Richardson presented the stimulus for 2 sec. within which the S was required to respond. Experiment I I reported above indicates that time allowed for response is an important variable. Of considerable interest, in the present study, is the finding for rote learning that performance deteriorates with increases in the number of stimuli assigned the same responses. Richardson (1958), Hake and Eriksen (1955), and Metzger (1958) found that, after paired-associate scores were corrected for guessing, the number of stimuli per response was not a significant variable. Richardson and Metzger both required their Ss to respond within 2 sec. while Hake and Eriksen required a response immediately following the appearance of the stimulus, making their experiments comparable to the forced-response conditions of Experiment II. Their findings are thus consistent with those of the present study when time is taken into account. Bricker (1955) permitted the Ss to respond at their own rate (with an upper limit of 12 sec.) and found that the number of stimuli per response was a significant variable. Other data gathered by the present authors suggest than an 8-see. presentation (viewing the stimulus for 4 sec.; having another 4 sec. to respond) is sufficiently long for the one-stimulus-per-response condition to match free-response performance (Mean 49.3) without appreciably changing the performance in the four-stimuli-per-response condition (Mean = 21). In a separate un-
VERBAL LEARNING AND CONCEPT FORMATION
published study, Stuart Thomas tested the effect of having provided a printed list of the nonsense-syllable "names" to the Ss. He found that while the list facilitated learning, response time and number of stimuli per response were still significant variables even when Ss were required to learn the names, as well as their connections with the stimuli. In general it would seem that given sufficient time, performance in the one-stimulus-perresponse condition is clearly superior. While increasing the number of stimuli assigned to a given response hinders performance in the rote-learning conditions, it was found to enhance performance in the concept-formation conditions. It is considered that all concept-formation tasks involve components of rote learning. The S not only must discover that each set of interrelated stimuli is associated with a single response, but he still must learn to associate the correct response to that stimulus set. As the number of distinct stimulus sets decreases, the relative prominence of this rote-learning component is diminished. Increasing the number of stimuli assigned to a given response, in the present design, decreases the number of stimulus sets, diminishes the rotelearning component of the task, and thus enhances performance, since fewer distinct stimulus-response associations need be established and retained. SUMMARY Two experiments are reported. In Experiment I, paired-associate learning was studied with ten groups of six college students each. Stimuli were color caps selected from the Farnsworth-Munsell color test; responses were nonsense syllables. Two sets of eight color caps were used, one of high, the other of low interstimulus similarity. In some conditions one stimulus was assigned to a response; in others there were two stimuli per response; in others, four. In some conditions
475
similar colors were assigned to the same nonsense syllable (allowing concept formation); in others colors were randomly assigned to a syllable (forcing rote learning). There were 100 learning trials, and each correct response was rewarded by payment of cash. A correction procedure was used. Analysis of the mean numbers of correct responses (after correction for guessing) allows the following conclusions: (a) both rote and concept learning is better when stimuli are of low similarity; (b) rote learning is much more difficult than concept learning; (c) increases in the numbers of stimuli per response hinder rote learning but aid concept learning. The results of Experiment I I suggest that the influence of the number of stimuli per response on rote learning is not independent of response time. The mean performance difference between one and four stimuli per response is small and nonsignificant when Ss are allowed only 2 sec. to respond, compared with a larger and significant difference when Ss are free to determine response time. This finding provides an explanation of some discrepant results of earlier studies. REI~FA~ENCES BI~ICKER, P. D. The identification of redundant stimulus patterns. J. exp. Psychol., 1955, 49, 73-81. FARNSWOETII, D. The Farnsworth-Munsell lO0-hue test manual. Baltimore, Maryland: Munsell Color Co., 1957.
GIBSON', E. J. A systematic application of the concepts of generalization and differentiation to verbal learning. Psychol. Rev., 1940, 47, 196229. GIBSOn, E. J. Intra-list generalization as a factor in verbal learning. J. exp. Psychol., 1942, 30, 185-200. GLAZE, J. A. The association value of non-sense syllables. J. genet. Psychol., 1928, 35, 255-269. HAKE, H. W , AND ERIIKSEN, C. W. Effect of number of permissible response categories on learning of a constant number of visual stimuli. J. exp. Psychol., 1955, 50, 161-167.
476
Si~ITH~ JONES~ AND TI-IOMAS
METZGER, l~. A comparison between rote learning and concept formation. J. exp. Psychol., 1958, 56, 226-231. RIC]~ARDSO~¢, J. The relationship of stimulus similarity and number of responses. J. exp. Psychol., 1958, 66, 478-484. U~DERWOOB, B. J. Studies of distributed practice:
II. Learning and retention of paired-adjective lists with two levels of intra-list similarity. ]. exp. Psychol., 1951, 42, 153-161. YNTEhfA, D. B., AND MUESER, G. E. Remembering the present states of a number of variables. J. exp. Psychol., 1960, 60, 18-22. (Received November 1, 1962)
470-476 (1963)
Effects upon Verbal Learning of Stimulus Similarity, Number of Stimuli per Response, and Concept Formation 1 T I M O T H Y A . S M I T H , 2 LYLE V.
JONES, AND
STUART THOMAS
University o/North Carolina, Chapel Hill, North Carolina From the moment a child begins to speak he is confronted with the task of learning the names of objects. The child soon discovers that some objects are grouped together under one name ("these round, red objects are apples"), while for other objects, each has a name of its own ("this red-faced lady is Aunt Minnie"). This paper reports two experiments designed to study the learning of names for objects with varying degrees of uniqueness of association. Several recent studies have investigated the learning of verbal responses associated with sets of stimuli. These studies may be viewed as investigations of paired-associate learning with the number of stimuli associated with a response being varied. Bricker (1955), after varying the number of stimuli per response and correcting performance scores for the effects of guessing, reported that the number of stimuli per response had 1 Part of this study (Experiment I) comprised the unpublished Master's thesis of the first author prepared in 1961 under the direction of the second author. The work was performed while Timothy Smith was a National Science Foundation Graduate Fellow. Stuart Thomas collaborated on Experiment II while a participant in the Undergraduate Research Program of the National Science Foundation in 1962. The study was partially supported by Grant M-1849, National Institutes of Health. We are grateful to Dr. Douwe Yntema and to Dr. Edward Zigler for constructive criticisms of an earlier draft. 2 Now at Florida State University, Tal]ahassee, Florida. 470
little effect on learning to associate verbal responses to stimuli. Hake and Eriksen (1955), Richardson (1958), and Metzger (1958) failed to find a relationship between the number of stimuli per response and learning to make the appropriate response to stimuli. But Yntema and Mueser (1960), utilizing a more complex experimental situation, have presented some evidence that number of stimuli per response can be a determinant of performance in associative learning. Their study focuses upon keeping track of the present states of a set of attributes, the states of which are subject to change during the experimental session. When the same state (response) is a possible value for several attributes (stimuli), performance is substantially poorer than when each state (response) is uniquely associated with an attribute (stimulus). These studies suggest that further investigation is needed into the effects of varying the numbers of stimuli per response upon paired-associate learning. A relatively unexplored facet of pairedassociate learning is the effect of assigning to homogeneous stimuli a common response. I t is one thing to assign to a particular class of round, red objects that are good to e a t the name, apple. However, something quite different is involved when we assign to the noise that a dog makes and to the outer covering of a tree, the common name, bark. When stimuli which are assigned a common
VERBAL LEARNING AND CONCEPT FORMATION name
exhibit
some common
property,
learn-
ing to pair the responses with the stimuli can be
viewed
When
as
groups
a of
concept-formation stimuli
signed to responses, stimuli-response rote
equivocal, indicate
versus
the
Of central the
concept
sponse
on
both
of r o t e
a
are
(1958) concept
s u p e r i o r to t h a t f o u n d
situation.
concern
learning
of
asthe
findings
learning
of Richardson
in the present
effect of the number
associate
(1958)
availability
r e s u l t s in p e r f o r m a n c e in a rote-learning
is one
Metzger's
the results that
task.
randomly
the task of learning
relationships
learning. Although concerning
are
rote at
and two
of stimuli concept
study per
is re-
paired-
levels of stimulus
similarity. EXPERIMENT I
Method Materials. Stimuli consisted of two groups of eight color "caps," selected from the F a r n s w o r t h Munsell 100-Hue test for color discrimination (Farnsworth, 1957). These caps are circular color patches, each m o u n t e d on a plastic cylinder, 10/16 in. in diameter a n d 7/16 in. in height. T h e colors were chosen from the Munsell color system so as to be identical in brightness a n d saturation b u t differe n t in hue. T h e colors are designed so t h a t caps with adjacent numbers, say 19, 20, a n d 21, differ in hue by equal-appearing steps. One set of eight color caps used in the experiment consisted of caps one step apart, specifically n u m b e r s 17, 18, 19, 20, 21, 22, 23, a n d 24. T h e other set was m a d e up of caps five steps apart, n u m b e r s 2, 7, 12, 17, 22, 27, 32, a n d 37. Both sets represent a yellowgreen range of hues. T h e selection of hues provides variation of stimulus similarity, the first set being more similar in hue according to the Munsell color system. Responses were eight nonsense syllables f r o m Glaze's list (1928) selected at r a n d o m with the following restrictions: (a) the association value of each, as reported by Glaze, is less t h a n 50%; (b) the first consonant of each syllable is distinct; (c) no syllables begin or end with x, q, e, or end with h or w (to assure familiar p h o n o l o g y ) ; (d) no syllable begins with y or g (to prevent phonemic association with yellow or g r e e n ) ; (e) the vowels a, e, i, a n d u each appear in two syllables. T h e syllables actually d r a w n were W E Z , KAG, H I F , LEB, NAF, M I P , ZUG, a n d VUT. To provide variation of the n u m b e r of stimuli per response,
471
in some conditions one color cap was assigned to each nonsense syllable, in others two caps to a syllable, a n d in others four caps to a syllable. Concept formation was m a d e possible in the conditions with m o r e t h a n one cap to a syllable by assigning caps adjacent on the Munsell color system to one syllable. T h e rote-learning task was created by randomly assigning n o n a d j a c e n t caps to a syllable. Design. Ss were 60 undergraduate m e n a n d w o m e n enrolled in introductory psychology at the University of N o r t h Carolina a n d at Duke University, r a n d o m l y assigned to 10 experimental groups, each containing 6 Ss. Groups were distinguished by different combinations of treatments: (a) rote learning or concept formation; (b) stimuli of high or low similarity; a n d (c) one, two, or four stimuli (color caps) per response (nonsense syllable). If each t r e a t m e n t h a d been completely crossed with each other treatment, a 2 X 2 X 3 factorial design would have resulted, with 12 experimental groups. However, with only one stimulus per response there is no distinction between rote learning and concept formation, a n d 2 of the 12 groups become extraneous, leaving only 10 distinct groups. Procedure. Lighting was controlled by the use of a darkened room a n d a M a c b e t h Daylight lamp type ALS30. U p o n entering the experimental room S was given a list of eight, f o u r , or two nonsense syllables a n d told he was participating in a s t u d y of the learning of n a m e s for colors. The E then placed the eight color caps one at a time before S a n d told h i m the nonsense syllable which was to serve as the n a m e of each. There followed 100 learning trials, with E m a n u a l l y placing one color cap before S on each trial; S was instructed to produce its n a m e orally, guessing if necessary, b u t responding on every trial. T h e S was allowed as m u c h time as he wished to respond, the color cap resting on an 8 X 11 in. piece of white paper in full view the whole time. The other seven color caps were concealed behind a cardboard screen. Caps were presented in the same r a n d o m order for all conditions. This order was determined by consulting a table of r a n d o m n u m b e r s so t h a t not only were the colors presented in a mixed sequence b u t one color m i g h t be repeated before all the other seven were presented. T h e nature of such a r a n d o m order was explained to S prior to the experiment. Immediately after each of his responses while the stimulus was still in view S was told the color cap's correct n a m e by E. After every 25 trials S was given a s u m of m o n e y based on the n u m b e r of correct answers he h a d achieved in the previous 25-trial block. T h e a m o u n t given for each correct answer was varied from condition to condition so that the average a m o u n t awarded Ss in each condition
472
SMITH~ JONES, AND THOMAS
approximated $1.00 over the total 100 trials. The S was allowed to refer to the printed list of nonsense syllables throughout the experiment, thus eliminating any need for learning the set of possible responses.
9O
Results
7O
The dependent variable was the score X where R is number of correct responses, W is number of wrong responses and n is the number of distinct response categories available. The possible values of n are 8, 4, and 2, depending upon whether 1, 2, or 4 color caps are associated with a given name. Under the assumption that all erroneous responses represent guessing and that the probability of a "correct" guess is I/n, X represents the number of correct responses other than those expected to be guessed correctly. Mean scores for all groups are plotted in Fig. 1. Three analyses of variance were performed: (a) to evaluate effects of stimulus similarity and number of stimuli per response with the rote-learning conditions (the six means represented by circles in Fig. 1); (b) to evaluate effects of stimulus similarity and number of stimuli per response within the concept-formation conditions (the six means represented by triangles in Fig. 1); (c) for conditions with more than one stimulus per response, all three effects were included in one analysis--stimulus similarity, number of stimuli per response, and rote versus concept learning [the eight means for two and four stimuli per response (Fig. 1)].
R--W/(n--1)
Rote Learning. Differences significant beyond the .001 level appear between stimulussimilarity conditions (F ~ 16.30, dj ~1/30), and between differing numbers of stimuli per response (F ~ 10.06, dj ~ 2/30) for rote learning. Their interaction does not reach the .05 significance level (F ~---2.30). Performance clearly is superior for the low stimulus-similarity condition. Under low stimulus similiarity, mean performance scores are 54.45, 31.56, and 19.33 for one, two, and four stimuli per response, respectively. Cor-
80
60
.....~ //Z~
5O
~//////
40
iiiiI ~
20
"~.. •x
/
//
.
"O
I0
Number of stimuli per response FIG. 1.
Mean performance scores (X)
for rote
learning (circles) versus concept learning (triangles), contrasting similar-stimulus conditions (dashed lines) with dissimilar-stimulus conditions (solid lines) for varying numbers of stimuli per response (Experiment I). responding means for high stimulus similarity are 25.14, 13.11, and 13.33. Further, for rote learning, performance tends to deteriorate as a function of increasing numbers of stimuli per response, for stimuli of both high and low similarity. These differences are portrayed graphically as the two curves connecting circles in Fig. 1.
Concept Formation. Both stimulus similarity and number of stimuli per response significantly affect performance in the concept-formation conditions. Both of these main-effect F-values (16.20, df = 1/30; 46.34, df ~-~2/30) are significant beyond the .001 level. Again their interaction is not significant (F = 2.65). As was the case for rote learning, the low stimulus-similarity
VERBAL LEARNING AND CONCEPT FORMATION conditions show superiority over the high stimulus-similarity conditions (Fig. 1), having means of 54.45, 81.1I, and 89.33 for one, two, and four stimuli per response, respectively. T h e corresponding high stimulussimilarity means are 25.14, 71.78, and 79.33. I n direct contrast to the finding for rote learning, performance is enhanced with increasing numbers of stimuli per response when concept formation is a possible aid to learning.
Rote Learning versus Concept Formation. This analysis was performed in order to make a comparison between the rote-learning and concept-formation conditions. Excluded are data obtained under the condition of only one stimulus per response. The results of the analysis show significant differences only between stimulus-similarity conditions ( F ~ 8.45, df ~ 1/40, P ~ .01) and between concept-formation and rote-learning conditions ( F ~ 262.99, d / ~ 1/40, P < .001). Appraisal of the subgroup means (Fig. 1, ignoring data for one stimulus per response) is an aid to the interpretation of these differences. Consistent with results of the earlier analyses, performance is better for conditions of low stimulus similarity than high stimulus similarity. I n addition, there is evidence that, when a single response applies to a group of two or four stimuli, performance is greatly enhanced if the group consists of homogeneous stimuli (conceptformation condition) rather than heterogeneous stimuli (rote-learning condition). EXPERIMENT I I
Meth~od It will be noted that Experiment I failed to centrol response time, for S was allowed as long as he wished to respond. In an attempt to assess the importance of this factor, another experiment was performed. Twenty-four undergraduates enrolled in introductory psychology at the University of North Carolina were the Ss. The materials and lighting were the same as in the previous experiment, with the exception that a black shutter operated by a small motor was used to regulate the exposure time
473
of the color caps. It was placed on a table top in a position intervening between S and the sheet of white paper where the color cap was placed manually by E. With the shutter closed, S was unable to view the cap. Two variations of response time were studied. In forced-response conditions an electronic timer raised the shutter and exposed the stimulus for 2 sec., after which S was given 2 sec. during which to respond. The color cap was then again shown to S for 2 sec. while he was told its correct name by E. In free-response conditions the color caps were exposed by raising the shutter, but no limit was placed either on the stimulus-viewing or response time, the shutter remaining open until S responded, as in Experiment I. The response time for each S on every trial was recorded. As in Experiment I, S was told the color cap's correct name immediately following each of his responses. In Experiment II, rote learning exclusively was studied, with stimuli of low similarity. Conditions of one and four stimuli per response were considered. Instructions and incidental experimental procedure were the same as in the main experiment, including the feature of payment for correct answers.
Results Again the dependent variable was the score 1). The analysis of variance for the four groups shows significant effects for response time ( F ~ 6.54, d / ~ 1/20, P < .05), n u m b e r of stimuli per response ( F ~ 13.95, d / ~ 1/20, P < .01), and their interaction ( F ~ 6.54, d / ~ 1/20, P < .05). As was found for rote learning in Experiment I, performance deteriorates with increasing numbers of stimuli per response. W i t h free response, the means for one and four stimuli per response were 52.40 and 20.00, respectively, almost identical to the means for the corresponding conditions in Experiment I. W i t h forced response, the means were 26.07 and 20.00 for one and four stimuli per response, respectively. The difference between the one- and four-stimuli-per-response conditions is considerably more marked in freethan forced-response conditions; the forcedresponse mean difference is not significant ( t ~ .814, d ] ~ 10, P > .40). This is the source of the significant interaction. When allowed free response time, tile mean time
X -~ R - - W / ( n - -
474
SMITtt, JONES, AND THOMAS
for response was 6.74 sec. for one stimulus per response and 3.42 sec. for four stimuli per response. DISCUSSION
In this study, performance on a pairedassociate rote-learning task is found to be better for stimuli of low similarity than for highly similar stimuli. This result is consistent with the findings of Gibson (1942), Underwood (1951), and others. A plausible interpretation of this effect, first advanced by Gibson (1940), is based on stimulus generalization. For highly similar stimuli a greater degree of generalization, and hence a higher rate of errors is anticipated than for stimuli of lower similarity. That is precisely what is found here. Stimulus generalization can also be used to explain the difference in performance between rote-learning and concept-formation conditions, when more than one stimulus is assigned to each response. In the conceptformation conditions, where stimuli adjacent on a dimension of stimulus similarity were assigned the same response, responses which are determined by stimulus generalization often are "correct" responses, as defined by E. In the rote-learning conditions, different responses are more likely to be associated with adjacent stimuli and stimulus generalization would interfere with correct performance. In view of the large consistent difference found between rote- and concept-learning conditions in this study (Fig. 1), it is reasonable to consider differences between this study and the similar ones of Metzger (1958) and Richardson (1958), where findings pertinent to this issue were considerably less striking. Three possible critical differences should be noted. First, the present design included 100 trials for each S, as compared with 200 and 400 trials in the studies of Metzger and Richardson, respectively.
The initial advantage that concept formation has over rote learning may disappear as learning progresses. Second, Ss in the present study were paid for correct responses. This pay probably served to maintain the S's attention to the task. As a result, correct responses accidentally discovered (on the basis of stimulus generalization) might then be incorporated into the S's repertory faster than if no payments had been made. Third, both Metzger and Richardson presented the stimulus for 2 sec. within which the S was required to respond. Experiment I I reported above indicates that time allowed for response is an important variable. Of considerable interest, in the present study, is the finding for rote learning that performance deteriorates with increases in the number of stimuli assigned the same responses. Richardson (1958), Hake and Eriksen (1955), and Metzger (1958) found that, after paired-associate scores were corrected for guessing, the number of stimuli per response was not a significant variable. Richardson and Metzger both required their Ss to respond within 2 sec. while Hake and Eriksen required a response immediately following the appearance of the stimulus, making their experiments comparable to the forced-response conditions of Experiment II. Their findings are thus consistent with those of the present study when time is taken into account. Bricker (1955) permitted the Ss to respond at their own rate (with an upper limit of 12 sec.) and found that the number of stimuli per response was a significant variable. Other data gathered by the present authors suggest than an 8-see. presentation (viewing the stimulus for 4 sec.; having another 4 sec. to respond) is sufficiently long for the one-stimulus-per-response condition to match free-response performance (Mean 49.3) without appreciably changing the performance in the four-stimuli-per-response condition (Mean = 21). In a separate un-
VERBAL LEARNING AND CONCEPT FORMATION
published study, Stuart Thomas tested the effect of having provided a printed list of the nonsense-syllable "names" to the Ss. He found that while the list facilitated learning, response time and number of stimuli per response were still significant variables even when Ss were required to learn the names, as well as their connections with the stimuli. In general it would seem that given sufficient time, performance in the one-stimulus-perresponse condition is clearly superior. While increasing the number of stimuli assigned to a given response hinders performance in the rote-learning conditions, it was found to enhance performance in the concept-formation conditions. It is considered that all concept-formation tasks involve components of rote learning. The S not only must discover that each set of interrelated stimuli is associated with a single response, but he still must learn to associate the correct response to that stimulus set. As the number of distinct stimulus sets decreases, the relative prominence of this rote-learning component is diminished. Increasing the number of stimuli assigned to a given response, in the present design, decreases the number of stimulus sets, diminishes the rotelearning component of the task, and thus enhances performance, since fewer distinct stimulus-response associations need be established and retained. SUMMARY Two experiments are reported. In Experiment I, paired-associate learning was studied with ten groups of six college students each. Stimuli were color caps selected from the Farnsworth-Munsell color test; responses were nonsense syllables. Two sets of eight color caps were used, one of high, the other of low interstimulus similarity. In some conditions one stimulus was assigned to a response; in others there were two stimuli per response; in others, four. In some conditions
475
similar colors were assigned to the same nonsense syllable (allowing concept formation); in others colors were randomly assigned to a syllable (forcing rote learning). There were 100 learning trials, and each correct response was rewarded by payment of cash. A correction procedure was used. Analysis of the mean numbers of correct responses (after correction for guessing) allows the following conclusions: (a) both rote and concept learning is better when stimuli are of low similarity; (b) rote learning is much more difficult than concept learning; (c) increases in the numbers of stimuli per response hinder rote learning but aid concept learning. The results of Experiment I I suggest that the influence of the number of stimuli per response on rote learning is not independent of response time. The mean performance difference between one and four stimuli per response is small and nonsignificant when Ss are allowed only 2 sec. to respond, compared with a larger and significant difference when Ss are free to determine response time. This finding provides an explanation of some discrepant results of earlier studies. REI~FA~ENCES BI~ICKER, P. D. The identification of redundant stimulus patterns. J. exp. Psychol., 1955, 49, 73-81. FARNSWOETII, D. The Farnsworth-Munsell lO0-hue test manual. Baltimore, Maryland: Munsell Color Co., 1957.
GIBSON', E. J. A systematic application of the concepts of generalization and differentiation to verbal learning. Psychol. Rev., 1940, 47, 196229. GIBSOn, E. J. Intra-list generalization as a factor in verbal learning. J. exp. Psychol., 1942, 30, 185-200. GLAZE, J. A. The association value of non-sense syllables. J. genet. Psychol., 1928, 35, 255-269. HAKE, H. W , AND ERIIKSEN, C. W. Effect of number of permissible response categories on learning of a constant number of visual stimuli. J. exp. Psychol., 1955, 50, 161-167.
476
Si~ITH~ JONES~ AND TI-IOMAS
METZGER, l~. A comparison between rote learning and concept formation. J. exp. Psychol., 1958, 56, 226-231. RIC]~ARDSO~¢, J. The relationship of stimulus similarity and number of responses. J. exp. Psychol., 1958, 66, 478-484. U~DERWOOB, B. J. Studies of distributed practice:
II. Learning and retention of paired-adjective lists with two levels of intra-list similarity. ]. exp. Psychol., 1951, 42, 153-161. YNTEhfA, D. B., AND MUESER, G. E. Remembering the present states of a number of variables. J. exp. Psychol., 1960, 60, 18-22. (Received November 1, 1962)