Professional Documents
Culture Documents
BY K, S. LASHLEY
Harvard University
AND
MARJORIE WADE
Radcliffe College
The results of these two tests are the tests. The difference (12 trials, 1
typical of eight others dealing with sur- error) is insignificant. As in the ex-
face brightness, direction of lines, num- periments with rats, there is not the
ber, and various similarities of form, slightest indication that continued train-
which it is unnecessary to report in de- ing with one stimulus forms any
tail. In every case a differential reac- stronger association with that stimulus
tion was established more quickly when than with another distant from it upon
the training involved extinction of the the same stimulus dimension.
initial reaction to a single stimulus than Tests with chimpanzees. Two adults
when that reaction was reinforced. The and two about 4 years old were used.
differences are statistically unreliable None of these animals had been trained
but are consistent in all 10 experiments. in any discriminative task before. The
PAVLOVIAN THEORY OF GENERALIZATION 77
TABLE 2
TRAINING SCORES FOR CHIMPANZEES IN TESTS OF STIMULUS GENERALIZATION
Two adults and two four year old animals, not previously trained in discrimination tests, were
first given 200 trials with one stimulus, then trained to 20 errorless trials with two. ++ origi-
nal reaction reinforced; •)— original reaction extinguished. F, failed to improve in SOO trials.
Figures Animals
tests were arranged as with the mon- ences between the cords used to pull in
keys except that the stimulus objects the boxes, and when those differences
were painted fronts of boxes to be were controlled he failed to show im-
pulled to the cage by stout cords from provement after SOO trials with each of
a distance of ISO cm. In initial train- two pairs of stimuli differing in color
ing the chimpanzee was presented with and form. His entire training score is
a single food box bearing a constant pat- therefore suspect. Vera, the other
tern, pulled it in, and got food from it adult, learned to discriminate colors but
200 times. A second box with a dif- failed all tests in which discrimination
ferent figure was then added and dif- of form was required.
ferential training continued to 20 con- The averages for all scores of all ani-
secutive errorless trials. The chimpan- mals (exclusive of failures) are given
zees are more difficult to motivate and in table 2. Larger numbers of trials
more erratic than the monkeys and the and errors were required when the origi-
range of their training scores is so great nal reaction was extinguished than when
as to make any conclusions from them it was reinforced but the differences,
questionable. In the first three tests 43.0 ± 123.4 trials and 14.3 ± 47.7 er-
the stimuli differed both in form and rors, are entirely unreliable. The vari-
color, with the object of presenting ability and inconsistency of the scores
optimal conditions for association with justifies the conclusion that any strength
specific characters of the stimuli father of association established by 200 posi-
than for generalization; the three later tive reactions to a single stimulus is in-
tests .involved differences of number or significant in comparison with other
size. Training scores are given in factors which modify the rate of learn-
table 2. ing.
In tests following the series pre- Test with human subjects. The hu-
sented in table 2, Shorty was found man adult has a vast fund of general
to be reacting on the basis of differ- concepts and tends to classify and rank
78 K. S. LASHLEY AND MARJORIE WADE
any new object in one or another gen- duce a generalization along the dimen-
eral category. Tests involving familiar sion of similarity of the profiles. With
stimulus dimensions are therefore worth- human subjects as with animals, re-
less for evaluating the theory of ir- peated stimulation is not in itself suffi-
radiation. For a significant test of cient to establish a stimulus dimension.
stimulus generalization a dimension The dimension arises as a result of at-
must be chosen which is unfamiliar to tention to and comparison of two ob-
the subjects, so that the formation of a jects.
new generalization can be observed. The results of all these tests are con-
Subjects were given a penny of cur- sistent and are incompatible with the
rent issue and a 'buffalo' nickel, asked theory of irradiation of the effects of
to examine the coins carefully and to conditioning. The theory postulates as-
enumerate all the resemblances or di- sociation with all aspects or dimensions
mensions of similarity by which the of the stimulus and a maximum strength
two could be compared and classified. of association with the stimulus actu-
Diameter, size, thickness, weight, color, ally used in training. It therefore im-
value, dates of issue, amount of wear, plies more rapid learning in those tests
and similar habitual categories of clas- where the reaction to the original stimu-
sification were readily specified by all lus is reinforced than in those in which
subjects. No one, however, discovered it is extinguished. No such result has
the scale of similarity represented by the been obtained. On the contrary, train-
profiles enlarged in figure 1 until his at- ing with one .stimulus fails to produce
tention was directed to it. The subjects a significantly greater strength of asso-
had, of course, looked at coins of these ciation with that stimulus than with
issues hundreds of times before but such others on the same stimulus dimension.
repeated exposures had failed to pro- Under conditions which rule out pre-
PAVLOVIAN THEORY OF GENERALIZATION 79
existing habits of relational thinking, form of the gradient will not account
no evidence for any graduated spread of for our results.
effects of training is obtained. Lack of attention. It may be ob-
jected that none of the tests reported
POSSIBLE OBJECTIONS TO THE above is relevant to the problem be-
EXPERIMENTAL TESTS cause, during the initial training with
It might be argued that the amount one stimulus, the subjects did not at-
of training with one stimulus was in- tend to those aspects of the stimuli
sufficient to establish a measurable dif- which were later differentiated. Spence
ference in habit strength between it and (24) has advanced such a criticism
other stimuli on the same dimension be- against somewhat similar experiments
cause the slope of the irradiation gradi- of Krechevsky's. All of the tests re-
ent is very slight. However, on the quired some manipulation of the stimu-
assumptions of neo-Pavlovian theory lus objects by the subjects under such
the 200 trials of initial training would conditions that the stimuli were visually
establish a greater difference in habit fixated and therefore excited the retina.
strength between the positive and nega- The theory as originally proposed as-
tive stimuli than that produced by the sumes that all afferent excitations from
differential training which actually pro- the stimulus are associated with the re-
duced errorless discrimination.4 The action. If this assumption is modified
4
We have not attempted to reduce these
to the form, "all aspects of the stimulus
'habit strengths' to the equations of the neo- which are attended to during primary
Pavlovian system. The spurious character of conditioning are associated with the re-
its quantitative and mathematical treatment of action," then the theory has to ac-
learning is illustrated by the definition of its count for differences in attention during
units of measurement, the hab and wat, in
terms of percentage of the practice required training with one stimulus and during
by a 'standard' organism to reach the physio- differential training with two, when dif-
logical limit of learning. Such units are com- ferent aspects of the stimuli are asso-
pletely devoid of meaning. The physiological ciated.
limit is in no case determinable and is a con-
cept of questionable value. Standardization is Application of the theory to the prob-
dependent upon one of three procedures. The lem of attention has been sketched by
standard may be an arbitrarily selected object Spence (23, p. 432). "Moreover, the
which can be preserved unchanged, like the animal learns many other responses in
bronze bar representing the British standard addition to the final, selective approach-
yard. This method is feasible for a physical
unit, but a standard rat might deteriorate ing reaction. Prominent and impor-
seriously in the course of years. The standard tant among these are what have been
may be calibrated in terms of some unvarying termed, for want of a better name,
natural phenomenon, as the standard ther- 'preparatory' responses. These latter
mometer by the boiling and freezing points of
water. No such natural constants exist for consist of the responses which lead to
learning. The only remaining alternative in- the reception of the appropriate as-
volves the assumption of a correlation be- pects of the total environmental com-
tween different indices of the magnitude of plex on the animal's sensorium, e.g.,
the variable being standardized. In endocrine
therapy a standard preparation may be de- the orientation and fixation of the head
fined in mouse units because this measure cor- and eyes toward the critical stimuli.
relates satisfactorily with therapeutic effects. That is, the animal learns to 'look at'
For learning, the correlation between different one aspect of the situation rather than
learning processes is so low as to preclude any another because of the fact that this
such method; as well define the standard rat
for discriminative learning by length of tail response has always been followed
as by performance in the maze. within a short temporal interval by the
80 K. S. LASHLEY AND MARJORIE WADE
in the cortex, and the direction of gen- ferent generalization continuum,' corre-
eralization by the spread of excitation. sponding to what we have called the or-
In discarding his physiological theory ganic continuum. This, however, leaves
his followers have been forced to make his discussion of generalization com-
the assumption that degree of similarity pletely circular. "It is held that the
is a direct function of the quantitative number and nature of the various pri-
physical relations of the stimulus ob- mary generalization gradients are caused
jects. This assumption is not explicitly jointly by the nature of the stimulus
stated but is implied in all discussions energy and the nature of the receptor
where degree of similarity is expressed response" (9, p. 198). No theory is
as distance on the physical continuum proposed, however, to account for the
(8, 21). If it were true, there would be production of a 'gradient of afferent
no problem of stimulus equivalence and generalization' by the receptor response.
the phenomena would be without theo- The statement quoted therefore merely
retical interest. The assumption is, constitutes an assumption that, through
however, negated by every item of evi- some mysterious process, the effects of
dence. Degree of similarity is a prod- training irradiate to stimuli which are
uct of the activity of the organism, not similar for the organism. Similarity is
a physical property. When the physical not defined other than as a product of
dimension is quantitatively continuous, this irradiation. Nevertheless, this pa-
as the frequency of sound or light ralogism is advanced as .a satisfactory
waves, confusions of discrimination may solution of fundamental psychological
be more frequent for remote than for problems. "In this way are explained
adjacent points on the continuum; both the paradox of the occurrence of
violet is more similar to red than is super-threshold learning where the con-
green; the individual with absolute pitch ditioned stimulus is never exactly re-
more frequently confuses notes an oc- peated, and the paradox of reaction
tave apart than notes within the octave. evocation where the evoking stimulus
Similarities in experience may exist for has never been associated with the re-
which no objective continuum is dis- action evoked" (9, p. 199).
coverable, as appears in the classifica- Identification of degree of similarity
tion of odors and in equivalencies across with relative position of the stimuli
sensory modalities. The relation of the upon the physical dimension or the pos-
physical to the organic continuum is tulation of a 'gradient of afferent gen-
not direct even in the cases of intensity eralization' introduces the conception
and extensity, used as illustrations by that the organism reacts (projects the
Spence and Hull; Weber's law and the effects of training) to relative position
phenomena of size constancy are some and thus assumes the very reaction to
of the many evidences that the physi- relations to which Spence and Hull ob-
cal dimension is always transformed by ject in field theory. It actually goes
organic processes into a different di- even farther from their claimed objec-
mension. This lack of correspondence tivism than does field theory, for the
between the physical and organic di- latter requires at least two stimuli to
mensions constitutes the problem of determine reaction to a relation, whereas
stimulus equivalence and generalization. the theory of irradiation supposes that
In a more recent discussion Hull has such a reaction is determined only by
recognized the problem of the relation the subsistent or potential relations of
of the physical dimension to the 'af- a single stimulus.
84 K. S. LASHLEY AND MARJORIE WADE