You are on page 1of 8

Journal of Experimental Psychology

1963, Vol. 65, No. 4, 339-346

STIMULUS GENERALIZATION ALONG ONE AND TWO


DIMENSIONS IN PIGEONS1
CHARLES M. BUTTER
University of Michigan

2 groups of pigeons were tested to compare 1-dimensional with 2-


dimensional generalization and to determine whether stimulus dimen-
sions have interacting effects of generalization. Following single stimu-
lus training, both groups were tested for generalization to stimuli varied
in wave length, in angular orientation, and in both dimensions. Both
groups showed less generalization to stimuli varied in 2 dimensions
than to stimuli varied in 1 dimension. Also, the 2 dimensions had inter-
acting effects on generalization. An attempt was made to describe the
mathematical form of this interaction in order to specify the function
relating multidimensional to unidimensional generalization.

Although stimulus generalization generalization decrements to stimuli


has been investigated from many varied in each dimension alone. On
viewpoints, little attention has been the other hand, stimulus interaction
directed toward the problem of how would result if generalization decre-
generalization takes place to stimuli ments to stimuli varied in one dimen-
varied along two or more dimensions sion did not summate. If such an
as compared with generalization to interaction were found, it would be
stimuli varied in a single dimension. of further interest to attempt to
Experimental analysis of this problem determine its mathematical form and
may be considered one approach to thus describe the function relating
the more general question of how multidimensional to unidimensional
different stimulus attributes combine generalization.
to affect perceptual similarity. The The method used in this study is
experiments reported here were under- the one employed by Guttman and
taken (a) to compare the amount of Kalish (1956) to obtain wave-length
generalization to stimuli varied in one generalization gradients in pigeons
and in two dimensions and (&) to following training with a single spec-
determine if two stimuli, each varied tral hue. The present study also
in a single dimension, exhibit inde- employed wave length as one stimulus
pendent or interacting effects on gen- dimension and a visual-spatial dimen-
eralization when combined (Guttman, sion, angular orientation of a band of
1956). Evidence for independent light (Butter & Guttman, 1957), as
effects of such stimuli on generaliza- the second dimension.
tion would be provided if the general- Generalization to stimuli varied
ization decrement to a stimulus varied in one and in several dimensions has
in two dimensions were the sum of been investigated in two recent studies.
1
The second experiment in this report Fink and Patton (1953) found that
forms part of a dissertation submitted to the the amount of generalization decre-
Graduate School of Arts and Sciences, Duke ment of an instrumental tube-drinking
University, 1959, in partial fulfillment of the response in rats was directly related
requirements for the PhD degree in psy- to the number of changed stimulus
chology. The author is indebted to Norman
Guttman for his advice in planning these components. White (1958) found
studies. more generalization decrement in
339
340 CHARLES M. BUTTER

children to stimuli varied in two TABLE 1


dimensions than to stimuli varied in NOMINAL AND ACTUAL PEAK TRANSMISSION
either dimension alone. However, VALUES OF MONOCHROMATIC
no conclusions were reached regard- INTERFERENCE FILTERS
ing the quantitative manner in
Nominal Value (Mju) Actual Value (Mp)
which stimuli combined to affect
generalization. 520 522
In the present study, two groups Group 1 550 548
580 579
of pigeons were tested. The 5s in
Group 1 were tested for generalization 530 532"
to a small number of stimuli, while 540 539
Group 2 550 548
those in Group 2 were presented with 560 561
a larger number of stimuli in order 570 571
to obtain gradients of generalization. a
Upon completion of the experiment, spectrophpto-
metric analysis indicated that the peak transmissions
METHOD value of the 530-m/z filter had changed to 538 mji.
Because it is likely that this change occurred during the
course of the experiment, responses to this wave-length
Apparatus.—The apparatus was an auto- stimulus are not included in the presentation of the
matically-controlled Skinner box, designed data.
for key pecking and similar to apparatuses
used in other wave-length generalization Subjects.—The 5s were 37 experimentally
studies (Guttman & Kalish, 19S6). The naive white Carnean pigeons, 10 of which
special feature of the present apparatus, how- were assigned to Group 1, and 27 to Group 2.
ever, was that only a narrow band of light was All 5s were housed in individual cages and
transmitted to the key, and this band of light maintained at 75% of their ad libitum body
could be rotated about its center to provide weight throughout the experiment.
various degrees of angular orientation. A Procedure.—Following magazine training,
circular translucent key, set within a frame, all 5s were trained to key peck by the method
was masked so that a band running through of successive approximations and were then
its center, 1 in. in length and xs in. in width, given 50 continuous reinforcements on each of
was exposed to the light beam. The key 2 successive days. During this stage of
could be rotated about its center within the training, an overhead light was on in the
frame by means of spur gears engaged by a apparatus, but in subsequent stages, the only
rod. When 5s were key pecking, the rod was illumination in the box was provided by the
moved back to disengage the gears on the key. light from the key. In training sessions, the
The angular orientation of the band of light band of light on the key was illuminated by
was set by means of a pointer attached to the light of 550 mju and was in a vertical position
other end of the rod. Settings, calibrated by (90°). Following the completion of key peck
a protractor, were accurate to 1°. Mono- training, 5s were trained on a variable-interval
chromatic interference niters (Bausch and (VI) reinforcement schedule with a mean in-
Lomb Company), which could be mounted in terval of 1 min. Training sessions were ad-
a holder in the light path, had band widths ministered over 10 days and consisted of 30
varying between 7 and 9 mju. The nominal 55-sec. light-on periods, during which the key
and actual peak transmission values of the was illuminated and key pecks were rein-
niters are shown in Table I.2 In subsequent forced, alternating with 15-sec. light-off
reference to wave-length stimuli, nominal periods, during which no key pecks were
values will be used. However, in graphs, reinforced.
wave-length values are spaced according to On the day following the completion of VI
their actual values. A wratten K-2 filter was training, 5s were tested for stimulus gen-
used along with the 570- and 580-mM niters eralization. Immediately before testing, 5s
in order to exclude lower wave lengths in the were reinforced for pecking at the training
first-order spectrum. stimulus for six 30-sec. periods. Subsequent
generalization testing was carried out in
2
Nominal values are those given in the extinction.
manufacturer's catalogue, while actual values In generalization testing, 5s in Group 1
are those determined spectrophotometrically were presented with stimuli differing in one
by the manufacturer and by E. and in two dimensions from the training
STIMULUS GENERALIZATION IN PIGEONS 341

stimulus by pairing three values of wave function of wave length, with angular
length, 520, 550, and 580 myu, in all possible orientation as the parameter, and as
combinations with three values of angular
orientation of the band of light, 40° (rotated a function of angular orientation,
50° to the left of 90°), 90°, and 140° (rotated with wave length as the parameter.
50° to the right of 90°). In addition, 0° It is evident that generalization on
(horizontal) was paired with 550 mp, so that the two tests is highly similar, differ-
a total of 10 different stimulus combinations ing only in the overall response level.
was presented. These 10 stimulus com-
binations were presented 12 times, and within Responses on the two tests were
each of the 12 blocks, the stimulus combina- pooled in analyses of the results.
tions were arranged in a different random Generalization decrement occurred
order. Each stimulus presentation lasted for to stimuli varied in either dimension;
30 sec. and was followed by a 15-sec.dark
period, during which stimuli were changed. analysis of variance indicated that
On the following day, Group 1 5s received a both wave length and angular orien-
second generalization test, administered in the tation effects are highly significant
same manner as on the first day of testing. (see Table 2). Further, there is an
The 5s in Group 2 were presented on the asymmetry in wave-length generali-
first day of generalization testing with all
possible combinations of five values of wave zation ; significantly more responses
length, 530, 540, 550, 560, and 570 mM, and were given to 580 m/j. than to 520
five values of angular orientation, 30°, 60°, m/i, t (9) = 7.90, p < .001. Angular
90°, 120°, and 150°. Each of these 25 stimu- orientation generalization, on the
lus combinations was presented six times, and
within each of the six blocks, stimulus com- other hand, was more nearly sym-
binations were arranged in a different random metrical ; among the 10 5s, there was
order. Three blocks of stimuli were presented no consistent tendency to generalize
on the second day of generalization testing. more to one value of angular orienta-
In all other respects, testing was conducted tion than to the other.
in the same manner as with Group 1.
In order to compare the amount of
generalization to stimuli varied in one
RESULTS and in two dimensions, differences
Group 1.—Mean responses to stim- between mean responses to these two
uli presented in the two generalization classes of stimuli were evaluated by a
tests are plotted in Fig. 1 and 2 as a t test for correlated scores. The
FIRST TEST

520 550 5800° 40° 90° 140*


WAVE LENGTH - ANGULAR ORIENTATION - DEGREES

FIG. 1. Mean total responses to wave length and angular orientation


stimuli on the first generalization test (Group 1),
342 CHARLES M. BUTTER

SECOND TEST
100

BO-\

<n
60-
£50 Mu.

40-

20-

0°-550 M)L

520 550 5800' 40° 90° 140°


WAVE LENGTH - MJJ. ANGULAR ORIENTATION-DEGREES

FIG. 2. Mean total responses to wave length and angular orientation


stimuli on the second generalization test (Group 1).

results of this test showed significantly parallel, but rather interact, as indi-
less generalization to stimuli varied in cated by the significant Wave Length
two dimensions than to stimuli varied X Angular Orientation interaction
in one dimension, i (9) =6.73, p<.001. (see Table 2). Furthermore, it ap-
The results may be further ex- pears from the graphs depicting the
amined in order to determine whether results that this interaction occurs in
stimuli on the wave-length and angu- such a way that the gradients are
lar orientation dimensions exert in- multiples of each other, i.e., the form
dependent or interacting effects on of the gradients suggests that relative
generalization. If the effects of these generalization along one dimension
stimuli are independent, then the is equal at different levels of the
family of gradients along one dimen- second dimension. Thus, for exam-
sion should be parallel. It appears, ple, if Rt = responses to the training
however, that the gradients are not stimulus, Ra = responses to a stimu-
lus varied only in angular orientation,
TABLE 2
Rw = responses to a stimulus varied
ANALYSIS OF VARIANCE OF TOTAL RESPONSES only in wave length, and Raw = re-
ON BOTH TESTS TO WAVE LENGTH
AND ANGULAR ORIENTATION sponses to a stimulus varied in both
STIMULI: GROUP 1 angular orientation and wave length,
then:
Source if MS F Ra Rwa
Total
Wave length (W)
89
1 109,751.5 37.75***
Rt Rw
Angular orientation (A) 2 148,678.5 12.49** and
W XA 4 13,058.6 5.79*
5s in groups 9 Ra X Rw
W XSs 18 2,899.2 Rwa =
A X5s 18 11,902.6 Rt
W X A X 5s 36 2,257.1
Dividing both sides of the equation
Note.—Since there were repeated observations on
5s, the significance levels of F ratios were tested by by Rt:
means of the conservative procedure suggested by
Geisser and Greenhouse (1958).
* p < .05. Rwa = Ra X Rw
**t <-01.
*** p < .001, Rt: Rt X Rt
STIMULUS GENERALIZATION IN PIGEONS 343

TABLE 3 are shown in Table 3. It is evident


MEAN RESPONSES OBTAINED AND PREDICTED that predicted responses closely ap-
BY THE MULTIPLICATIVE RULE : GROUP 1 proximate obtained responses; the
overall mean error of prediction was
Mean
Obtained
Mean
Predicted PVE»
7.9%. Included in the same table
Stimulus
Responses Responses are values of PVE (percent variance
580 mp, 40° 84.1 91.6 .92 explained by predicted responses);
580 mit, 140° 89.5 86.5 .93 the high values of this statistic indi-
520 m», 40° 46.4 44.4 .90 cate that responses of individual 5s
520 nip, 140° 49.3 41.9 .92
were predicted with a high degree of
a
accuracy.
Proportion of Variance Explained = 1 ~-
Group 2.—The results of generali-
where p = predicted responses and o = obtained re-
sponses (Levine, 19S9). zation testing, shown in Fig. 3 and 4,
are similar to those of Group 1.
According to the last equation, rela- Analysis of variance indicates that
tive generalization to the stimulus both wave length and angular orien-
varied in two dimensions equals the tation effects were significant (see
product of the values of relative Table 4). Also, wave-length generali-
generalization to the two stimuli each zation was markedly asymmetrical,
of which is varied in a single dimension. i.e., significantly more responses were
Mean responses to stimuli varied given to 560 m/i than to 540 m/j,,
in two dimensions and responses pre- I (26) = 7.94, p < .001. As in the
dicted by this "multiplicative rule" previous experiment no consistent

FIRST TEST

150-

ffi
K.

540 550 560 30° 60* 90° 120' 150*

WAVE LENGTH-Mu ANGULAR ORIENTATION-DEGREES

FIG. 3. Mean total responses to wave length and angular orientation stimuli
on the first generalization test (Group 2),
344 CHARLES M. BUTTER

SECOND TEST

550 M«
24-

16-

550 560 570 30« 60° 90° !»•


WAVE LENGTH-MO ANGULAR ORIENTATION-DEGREES

FIG. 4. Mean total responses to wave length and angular orientation stimuli
on the second generalization test (Group 2).

asymmetries were found in angular Table 5. Prediction accuracy is


orientation generalization. Further, similar to that found previously;
significant differences were again the mean error of prediction was 9.7%.
found between mean responses to However, it is evident that the ma-
stimuli varied in one and in two di- jority of values of PVE are markedly
mensions ; fewer responses were given lower than those found in the results
to stimuli varied in two dimensions
than to stimuli varied in one dimen- TABLE 4
sion, t (26) = 7.54, p < .001. ANALYSIS OF VARIANCE OF TOTAL RESPONSES
Finally, a significant Wave Length ON BOTH TESTS TO WAVE LENGTH
AND ANGULAR ORIENTATION
X Angular Orientation interaction re- STIMULI : GROUP 2
sults from the tendency for gradients
along one dimension to broaden at Source df MS P
successive levels of the second di- Total .5.19
mension (see Table 4), although in- Wave length (W)
Angular orientation (A)
3
4
35,477.4 9.35**
222,794.4 31.32***
spection of Fig. 3 and 4 reveals that Ss in groups 26
W XA 12 3,403.4 3.27**
this tendency is not entirely con- W XSs 7X 3,792.8
A XSs 104 7,114.3
sistent. The multiplicative rule de- W X A X 5s 312 1,039.7
scribed previously was used to predict
responses to stimuli varied in two Note.—'The conservative procedure of Geisser and
Greenhouse (1958) was used in evaluating the sig-
dimensions, and mean obtained and nificance of the F ratios.
**p <.01.
predicted responses are shown in ***f <,OQ1,
STIMULUS GENERALIZATION IN PIGEONS 345

TABLE 5 both groups. This finding is not con-


MEAN RESPONSES OBTAINED AND PREDICTED sistent with the view that multidimen-
BY THE MULTIPLICATIVE RULE : GROUP 2 sional generalization is the result of the
algebraic summation of generalization
Mean Mean decrements to all discriminable stimulus
Stimulus Obtained Predicted PVE changes. The form of the generalization
Responses Responses
gradients obtained in Group 1 suggested,
540 mp, 30° 22.1 23.5 .46 rather, that relative generalization to a
540 mn, 60° 57.6 68.1 .30 stimulus varied in two dimensions equals
540 HIM, 120° 73.3 67.6 .39 the product of the values of relative
540 m/j, 150° 21.5 21.7 .44
560 m/i, 30° 38.5 36.3 .56 generalization to stimuli varied in a
560 m/i, 60° 87.2 104.9 .52 single dimension alone. In perceptual
560 mn, 120° 91.9 104.1 .74 terms, this multiplicative rule states that
560 mn, 150° 28.9 33.4 .88
570 niM, 30° 30.5 27.9 .90 the similarity between stimuli differing
570 HIM, 60° 89.3 80.7 .16 in several dimensions is a joint function
570 HIM, 120° 76.0 80.1 .52 of the perceived similarity between
570 m M> 150° 27.2 25.7 .37 stimuli within each of the relevant di-
mensions. The multiplicative rule pre-
dicted mean responses and responses of
of Group 1, reflecting the poorer individual 5s in Group 1 with a high
prediction of individual Ss' responses. degree of success. However, predictions
of individual Ss' responses in Group 2
DISCUSSION were much less accurate, a finding which
may be accounted for by the fact that
The results of this study indicate that each stimulus was presented less fre-
the operant situation described by Gutt- quently to these 5s than to those in
man and Kalish (1956) in their study of Group 2, thereby tending to reduce re-
stimulus generalization can also be em- liability of measurement. It is also
ployed to obtain gradients along two possible that response interactions be-
dimensions from individual 5s. More- tween stimulus presentations contributed
over, these gradients are similar to those variability to the results and thus to
generated in tests in which stimuli are predictions of responses. The lower ac-
varied only in a single dimension. Thus, curacy of predictions of the Group 2 data
in both, the wave-length gradients may also be due, at least in part, to lack
around 550 m/i were asymmetrical in of complete orthogonality of the dimen-
the same direction as those reported in sions, i.e., probabilities of response to
previous studies (Blough, 1961; Gutt- each aspect of the stimulus may not be
man & Kalish, 1956; Hanson, 1959), independent. However, since no con-
while angular orientation gradients were stant errors of prediction were found in
more symmetrical, resembling those de- mean responses or in responses of indi-
scribed earlier for this dimension (Butter vidual 5s, it appears unlikely that depar-
& Guttman, 1957). ture from orthogonality had any sig-
The results of both groups demon- nificant effect on the results. Finally,
strate that generalization to stimuli one should recognize the possibility that
varied in two dimensions is less than multidimensional generalization cannot
generalization to stimuli varied in either be described by any simple quantitative
dimension alone. This finding is con-
rule that is applicable to a variety of
sistent with previous ones (Fink &
Patton, 1953; White, 1958) and suggests stimulus dimensions and experimental
that perceptual similarity can be altered conditions. Future studies using other
more readily by multidimensional than dimensions and methods of testing may
by unidimensional stimulus changes. provide more information about the
A significant interaction between di- manner in which multidimensional gen-
mensions was found in the results of eralization occurs.
346 CHARLES M. BUTTER

REFERENCES GUTTMAN, N. The pigeon and the spectrum


and other perplexities. Psychol. Rep.,
BLOUGH, D. S. The shape of some wave- 1956, 2, 449-460.
length generalization gradients. /. exp. GUTTMAN, N., & KALISH, H. I. Discrimin-
Anal. Behav., 1961, 4, 31-40. ability and stimulus generalization. /. exp.
BUTTER, C. M., & GUTTMAN, N. Stimulus Psychol., 1956, SI, 79-88.
generalization and discrimination along the HANSON, H. M. Effects of discrimination
dimension of angular orientation. Amer. training on stimulus generalization. J. exp.
Psychologist, 1957, 12, 449. (Abstract) Psychol., 1959, 58, 321-334.
FINK, J. B., & PATTON, R. M. Decrement of LEVINE, M. A model of hypothesis behavior
a learned drinking response accompanying in discrimination learning set. Psychol.
changes in several stimulus characteristics. Rev., 1959, 66, 353-366.
J. comp. physiol. Psychol., 1953, 46, 23-27. WHITE, S. W. Generalization of an in-
GEISSER, S., & GREENHOUSE, S. W. An strumental response with variations in two
extension of Box's results on the use of the attributes of the CS. /. exp. Psychol.,
1958, 56, 339-343.
F distribution in multivariate analysis.
Ann. math. Statist., 1958, 29, 885-891. (Received May 14, 1962)

You might also like