SUME EXPERIMENTAL PARAMETERS
AFFECTING MEMANCN IN
CONCEPT LEARNENG

“west: {:60 fine Degree oi: M. A.
MZC .EGAN STATE EIREVERZSETY
Charles Ernest Kenoyer

13.967

 

   
 

  

LIBRARY

Michigan State
. University

may

    

ABSTRACT

SOME EXPERIMENTAL PARAMETERS
AFFECTING MEDIATION IN CONCEPT LEARNING

by Charles Ernest Kenoyer

If it is assumed that organisms are simultaneously
capable of mediated learning and of unmediated learning,
the Kendler association model leads to some new predictions.
In particular, the assumption that relative difficulty
of reversal and nonreversal shifts depends upon the num-
ber of associations necessary in each mode of learning
yields an experimentally verifiable hypothesis that the
number of stimulus dimensions is an important parameter
affecting the tendency of gs toward reversal or nonrever-
sal shifts.

A transfer design was used, in which 28 three-year-
old.§s were given three concept tasks. The first task
differed for the two shift groups in that the relevant
dimension for the reversal group was size, and the rele-
vant dimension for the nonreversal group was color. In
the second concept task, half the stimuli were set aside
so that the correct Option could be selected by either
of two criteria: the opposite value on the initially

relevant dimension, or a value on the initially irrelevant

Charles Ernest Kenoyer

dimension. In the third concept task, the full set of
stimuli was again used; half the gs were required to choose
on the basis of the initially irrelevant dimension (non-
reversal task) and half were required to choose the oppos-
ite value on the initially relevant dimension (reversal
task). Stimuli varied on four dimensions.

The prediction was confirmed that over half the §s
would complete the second task by performing a reversal
shift, as indicated by each.§fs first response on the
third task. Confirmation was also obtained for the hypo-
theses that §s in the reversal task group would complete
the final task in fewer trials than the nonreversal group.

The second major experimental variable was the effect
on performance of telling §§ at the end of the first task
that they had learned the "game" and inviting them to play
another, similar game. Half the §s were given this treat-
ment, while the other half received no indication of a
change in procedure. The significance level for inter-
action of the information variable with type of shift was
.05 < p < .10. Analysis of variance on log trials to the
last error yielded a significant F ratio for the shift
task effect (p < .01) and for its interaction with informa-
tion (p < .05). The prediction that number of errors
per error trial (using the correction method) would be
affected by information was not supported.

It was concluded that gs in this sample had demon-

Charles Ernest Kenoyer

strated mediation, and therefore that the single-stage
associative model was inadequate to describe their behav-

ior. Implications of the results were discussed.

SOME EXPERIMENTAL PARAMETERS

AFFECTING MEDIATION IN CONCEPT LEARNING

BY

Charles Ernest Kenoyer

A THESIS

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

MASTER OF ARTS

Department of Psychology

1967

ACKNOWLEDGMENTS

I want to express my sincere appreciation to
several persons who made substantial contributions to
the process of writing this thesis. I am indebted to
Dr. Frank N. Marzocco, chairman of my thesis committee,
for his encouragement and support during the planning
and execution of the experiment and for his help in the
development of the manuscript. I wish to thank
Dr. John E. Hunter, a member of the committee, for his
contributions to the planning of the experiment and
suggestions for analysis of the data. Thanks are also
due the other members of the committee, to Dr. Terrence M.
Allen for his suggestions for the analysis of data, and
to Dr. James L. Phillips for his critical reading of the
manuscript.

The valuable help of Lawrence H. Nitz in designing
the experimental apparatus is gratefully acknowledged.

Finally, I wish to express my gratitude to my
wife Janice, not only for her encouragement and moral
support, but for her tangible contribution to the work

of transcribing data.

ii

TABLE OF CONTENTS
Page
ACKNOWLEDGMENTS. . . . . . . . . . . . . . . . . . . ii
LIST OF TABLES . . . . . . . . . . . . . . . . . . . iv
LIST OF FIGURES. . . . . . . . . . . . . . . . . . . v

LIST OF APPENDICES . . . . . . . . . . . . . . . . . Vi

INTRODUCTION . . . . . . . . . . . . . . . . . . . . 1
DESIGN AND HYPOTHESES. . . . . . . . . . . . . . . . 16
METHOD . . . . . . . . . . . . . . . . . . . . . . . 23
APPARATUS . . . . . . . . . . . . . . . . . . . . 23
SUBJECTS. . . . . . . . . . . . . . . . . . . . . 28
MATERIALS . . . . . . . . . . . . . . . . . . . . 29
PROCEDURE . . . . . . . . . . . . . . . . . . . . 30
RESULTS. . . . . . . . . . . . . . . . . . . . . . . 36

DISCUSSION . . . . . . . . . . . . . . . . . . . . . 50

REFERENCES . . . . . . . . . . . . . . . . . . . . . 55

APPENDIX A . . . . . . . . . . . . . . . . . . . . . 57

APPENDIX B . . . . . . . . . . . . . . . . . . . . . 60

iii

Table

1.

LIST OF TABLES

Frequencies for Groups Classified by

Information and Inferred Shift .

Trials to Last Error for the Three Concept

Tasks. . . . . . . . . . . . .

Analysis of Variance on Phase 3 Trials to

Last Error 0 O O O O O O O O 0

Trials to Last Error for SS Grouped by

Information and Inferred Shift..

Log Trials to Last Error for the Three

Concept Tasks. . . . . . . . .

Analysis of Variance on Transformed Data .

Number of Error Responses Per Error Trial.

Frequencies for Groups Classified by

Shift Task and Inferred Shift.

iv

Page

37

39

40

42

44
45

47

48

Figure

LIST OF FIGURES

Block Diagram of Experimental Apparatus.

Apparatus-subject Interface. . . . .
Experimenter's Switching Device. . .

Apparatus Configuration (Plan View).

Page

24
25
27

31

LIST OF APPENDICES

Appendix Page
A. Stimulus Sequences for All Tasks . . . . . 57
B. Response Sequences for All Subjects. . . . 60

vi

INTRODUCTION

Several investigators in recent years have used the
concept-learning experiment as a method of evaluating
theories of mediation. In particular, the transfer design
and modifications of it have been used extensively. A
consistent conclusion has been that preschool children
do not make use of mediation in learning concepts, but
this conclusion is perhaps more general than is warranted
by a set of experiments in which some parameters have
been held constant across the whole set. The parameters
in question are the number of stimulus dimensions (attri-
butes) and the information, at the end of the initial task,
that a new task is beginning.

The use of binary dimensions limits the possible
kinds of concept shifts to two: reversal and nonreversal
shifts. In the reversal shift, §_is required in the
training task to learn to classify stimuli according to
their value on a single attribute, and in the test task
to assign stimuli to opposite categories, using the same
attribute as the classification criterion. For instance,
size might be relevant, with large stimuli called correct
in the training phase and small in the test phase.

In the nonreversal shift task, §_is also required to

learn a training task, but the second, or test, task

requires an entirely different classification scheme.
In this part of the transfer task.§_must classify stim—
uli according to a new criterion, an attribute that either
varied irrelevantly or was constant during the training
phase. For example, stimuli could be classified by size
in the training phase and by shape in the test phase.
Simple concepts have typically been used in exper-
iments on concept shifts. Although a defining characteris-
tic of the concept learning experiment is that stimuli
vary with respect to more than one attribute, a simple
concept, as opposed to a conjunctive or disjunctive con-
cept, is identified by only one of these attributes, the
relevant one, and the remaining attributes are irrelevant.
Size, for instance, might be the relevant attribute or
dimension, while shape and color might vary, but in such
a way as to provide no help to §_in his task of classify—
ing stimuli. Some of the ways in which §s have been re—
quired to indicate classification are by sorting stimuli
into groups, stating whether individually presented stim—
uli are or are not instances of a concept, and choosing
the one of a simultaneously presented pair that is an
instance of the concept.
Buss compared the two kinds of transfer in 1953.
His gs were adult humans, and the stimuli were wooden blocks,
the same ones that Buss (1952) had described as similar
to those of the Vygotsky test. The blocks varied with

respect to height, color, tOp surface area, and shape;

there were two values for height and three for each of
the other dimensions. Classification was accomplished
by sorting blocks into two groups. Buss found that the
nonreversal shift required more trials to criterion than
the reversal shift. He attributed this result to partial
reinforcement of the nonreversal response during early
trials of the test phase. Since gs performing a non-
reversal shift performed correctly on half the trials

of the test phase if they continued to classify stimuli
exactly as in the training phase, Buss reasoned that the
training phase response was not extinguished immediately
and so interfered with acquisition of the correct test
phase response. The gs performing a reversal shift, on
the other hand, performed incorrectly on all trials of
the test phase if they maintained the established train-
ing phase response, and so the old response was extin-
guished without delay.

Kendler and D'Amato (1955) offered an alternative
explanation. They agreed that partial reinforcement had
occurred in the nonreversal condition of the Buss exper-
iment, but considered it to be of secondary importance
in determining the results. They hypothesized that the
reason for the relative facility of the reversal shift
was that this condition, but not the other, required the
same mediating response for both tasks. The crucial ex-
periment required the elimination of partial reinforce-

ment of the training phase response, and Kendler and

D'Amato accomplished this by setting aside the stimuli
that could be correctly assigned to the same category in
both tasks, returning them to the stimulus set again only
after §_had reached criterion on the reduced set.

Since the attributes of the stimuli used in the
Kendler-D'Amato study were complex, the following illus-
tration is given in terms of simpler attributes. If all
large stimuli belonged in category A for the training
task, and all black stimuli belonged in category A for
the test task of the nonreversal group, then Se in both
the reversal and the nonreversal groups would be shown
only large white and small black stimuli during the early
test trials. After learning this task to criterion, groups
then received different treatments, in that they were
required to classify differently the stimuli returned to
the set for the final task.

under the modified procedure gs again learned the
reversal task more easily than the nonreversal task.
Kendler and D'Amato concluded that Buss's partial rein—
forcement argument could not be applied to their eXper-
iment and that the extant S-R model for learning in infra-
human organisms (Spence, 1936) was insufficient to explain
their data. They suggested a two-stage model in which
the S-R association was mediated by a covert re3ponse
specific to the relevant stimulus dimension.

Buss was reluctant to abandon the simpler model; he

argued that the Kendler-D'Amato procedure might still allow

the occurrence of partial reinforcement. The partially
reinforced response to which he referred, however, could
only be construed as a covert mediating response. As

H. H. Kendler stated in a note at the end of Buss's paper,
Buss had apparently adopted a mediational model without
being aware of doing so. The experimental results reported
by Buss in the same paper confirmed those of Kendler and
D'Amato; Harrow and Friedman (1958) again obtained con-
firming data under somewhat different conditions.

At about the same time this work was proceeding with
human Se, Kelleher (1956) found that rats learned the
nonreversal shift more quickly than the reversal under
the conditions of his experiment. The contrast leads
naturally to the hypothesis that phylogenetic differences
are responsible for the discrepancy, but it must be recog-
nized that the rat study involved pairs of stimuli that
differed on only two dimensions, while the studies with
adult humans employed more dimensions and, in some in-
stances, more values per dimension.

With the expectation that a transitional stage could
be found in the human developmental process, Kendler and
Kendler (1959) performed a learning experiment on kinder—
garten children. The stimuli in that experiment were
empty water tumblers of two heights, enameled either white
or black. The classification task was to select the cor-

rect tumbler, that which covered a reward. Pairs were

presented simultaneously, the members of any pair differ-
ing from each other on both dimensions. The possible
combinations, therefore, were tall black with short white
and tall white with short black. In a simple two-group
comparison, there was no significant difference in number
of trials to criterion. The gs were then separated into
two categories on the basis of training task performance;
the slow learners were found to perform the nonreversal
shift more rapidly than the reversal shift while fast
learners performed the reversal shift more quickly.

The Kendlers concluded that the group studied was
in transition from unmediated to mediated learning, with
the slow learners still tending to learn through a single-
stage S-R process while fast learners tended to mediate.
Since the transition hypothesis refers to a developmental
change, the conclusion does not follow from the data of
the experiment; moreover, procedural differences as well
as population differences distinguish the Kendlers' ex-
periment from the previous studies with adults. As in
Kelleher's rat study, the number of distinct pairs of
stimuli presented to the Kendlers' gs was smaller than
in any of the adult studies.

The next step in attempting to establish the exist—
ence of the transitional stage was to investigate shift
behavior of preschool children (Kendler, Kendler, and

Wells, 1960). The stimulus situation was similar to that

of Kendler and Kendler (1959), but the stimuli were ena-
meled cookie cutters and the nature of the transfer task
was different. Stimuli for the training series differed
with respect to one dimension only, for all gs. Tasks
for the three experimental groups differed only in the
test phase, a design (Andreas, 1960, p. 476) which happens
to confound type of shift with type of test-phase task.
Stimuli for the test phase varied on a new dimension as
well as on the dimension that had been relevant through-
out the training phase. For the training phase, 96 gs
were evenly divided into four groups on the basis of the
characteristic of the positive stimulus. For two groups,
all stimuli were of medium height; black was positive for
one group and white for the other. For the other two
groups, all stimuli were red; tall positive for one group
and short for the other. In the test phase, stimuli for
the control and reversal gs varied both in height and in
brightness. Pairs differed simultaneously on both dimen-
sions, so that tall white and short black or tall black
and short white cookie cutters were presented. For the
nonreversal group, the positive training-phase cue did
not appear in the test series; a nonreversal §_trained

to white, for example, was presented with stimuli that
differed in size and shape, but not color. Forms were
circles, clubs, and spades; colors were white, red, and

black; and heights were .5 in., l in., and 2 in., obtained

by presenting cutters singly or in stacks. The control
group reached criterion most quickly, followed by the
nonreversal and reversal groups in that order.

Next in the series of studies was a comparison of
several age groups. Kendler, Kendler, and Learnard (1962)
collected data on groups with mean ages of 43, 54, 77,
102, and 127 mo., and a maximum within-group range of
13 mo. Stimuli were black or white squares of two sizes
on a mid-gray background. The design included three phases:
the training phase; the early trials of the test series,
in which the reduced set of stimuli was presented; and
the later trials of the test phase, in which removed stim—
uli were reintroduced. This design, now called the optional
shift design, was identical to the transfer design with
control for partial reinforcement (Kendler and D'Amato,
1956) throughout the first two phases. The third phase
of the optional shift design, however, was distinguish—
able by the following characteristics: (a) a fixed num-
ber of trials, and (b) a change in the pattern of rein-
forcement. The gs were rewarded independently of their
choice response to reintroduced stimuli, while the reward
contingency for the pairs that had been presented through—
out Phase 2 remained the same as on Phase 2 trials. Those
gs who made the same choice for at least 8 out of 10 of
the presentations of the reintroduced stimuli were clas-

sified as reversal or nonreversal, depending upon the choice

reSponse, and those whose responses were split in any
other way were classified as inconsistent. The propor-
tion of gs in the reversal and inconsistent categories
differed significantly over the age range tested, but the
proportion of nonreversals did not differ significantly.
The statistical test employed was not reported.

Jeffrey (1965) conducted a follow-up study consist-
ing of two experiments. Experiment I was concerned with
the effect of pre-training on shift behavior; gs were
4-year-olds and lO-year-olds. Stimuli for the optional
shift task were black or white circles of two sizes, .5
in. thick, and stimuli for the pre-training task were 20
pairs of small toys mounted on masonite squares so as to
cover the reinforcement holes. The gs in the learning-
set training group were trained to a criterion of 9 correct
second-trial responses in 10 sets, then Phase 1 of the
optional shift problem was initiated. gs in the control
group began Phase 1 with no pre-training. Pre-training
facilitated Phase 1 learning but had no significant effect
on learning in Phase 2. Small sample size (eight 4-year-
olds per group) made it necessary to pool inconsistent
and nonreversal categories for a chi square comparison
of Phase 3 performance. The differences in category fre-
quencies were not significant for experimental groups
pooled across age, but the age effect was significant.

Jeffrey observed that the younger children tended to

-10-

verbalize values of both dimensions of the stimuli. He
considered it likely that those §s would maintain the
specific choice response learned in Phase 1 and so exhibit
the nonreversal pattern.

In Experiment II, Jeffrey replicated the design of
Kendler, Kendler, and Learnard except for three changes:
the ages of the gs, the kind of stimuli presented, and
the changing of a stimulus characteristic between Phase 1
and Phase 2. The age groups in Experiment II included
children 4, 6, and 8 years of age and a group of college
students. The 4-year-old group had a mean age of 4.16
years and ranged from 46 to 54 months. The other groups'
means and ranges were not reported. Stimuli like those
employed in Experiment I were again presented, but included
squares as well as circles. Circles were presented in
the first phase and squares in Phases 2 and 3. By chang-
ing the form of the stimuli after Phase 1 was completed,
Jeffrey hoped to reduce the chance that the test-stimuli
would be treated by the gs as the identical Phase 1 forms.
Under the conditions of the Jeffrey study no significant
differences were found among the proportions of gs in
the three shift-type categories for the various age groups.
The constancy is also remarkably apparent in Jeffrey's
table of pr0portions.

The discrepancy between Jeffrey's results and those

of Kendler, Kendler, and Learnard may reasonably be

-11-

attributed to the changes introduced by Jeffrey and cited
above. The difference between the stimulus sets used in
the two studies is not striking, particularly in View of
the considerable variation in stimuli among the previous
studies, all of which found age differences. The age
differences between the youngest groups of the two studies
cannot be discounted: the graph presented by Kendler,
Kendler, and Learnard shows that the most rapid perform-
ance change occurs within the range of their two youngest
groups, and Jeffrey's youngest group falls in that range.

The fact that a change in stimulus form was introduced
into the procedure seems an important departure from the
procedures of all of the preceding studies of classifica-
tion shifts. Jeffrey intended, of course, that the change
have an effect on perfonmance and showed, by means of a
chi square comparison, a significant difference between
the performances of the 4-year-old groups in his two ex-
periments. He concluded that the procedure of Experiment
II had been successful in reducing the tendency to respond
to all dimensions on which stimuli varied.

Beginning with the Kendler-D'Amato experiment, the
studies cited above have been concerned with experimental
hypotheses derived from the model presented by Kendler
and Kendler (1962). The Kendler model includes two alter—
native processes: the single-unit process and the two-

stage mediational process. The single—unit theory is the

-12-

S-R system of Spence (1956); the Kendlers argued that this
theory predicts more rapid learning for the nonreversal
than the reversal shift.

The reason for this is that at the time of
the shift the difference between the strength of
the dominant incorrect habit and the to-be-correct
habit is much greater for the reversal, as com-
pared to the nonreversal shift. Consequently
more training will be required to make the cor-
rect habit dominant in a reversal shift. Accord-
ing to the mediational theory the situation is
entirely different. A reversal shift enables
the subject to utilize the same mediated response.
Only the overt response has to be changed. A
nonreversal shift, on the other hand, requires
the acquisition of a new mediated response, the
cues of which have to be attached to a n§w_overt
response. Because the old mediational sequence
has to be discarded and a new one formed, the
nonreversal shift should be executed more slowly
than a reversal shift. (Kendler and Kendler,
1962, p. 7).

A footnote pointed out that the superiority of the rever-
sal shift may be explained either by the greater number
of new associations necessary for a nonreversal shift or
by the "possibility . . . that a mediating response is
more difficult to extinguish than is an overt response"
(p. 7).

The two important parameters of the Kendlers' model,
then, are the number of associations necessary for the
acquisition of correct performance and the difficulty of
extinguishing overt responses as opposed to mediating
responses. In an investigation of developmental changes
in the mediational process, it is important to take these

parameters into consideration, particularly as they affect

-13-

the relative difficulty of acquisition through the two
processes, single-stage and two-stage. Unless it is
assumed that humans lose their ability to learn without
mediating responses when they acquire the ability to
learn with them, variations in these parameters may be
expected to have an effect upon selection of the process
(single-stage or two-stage) by which the task is learned.
If the mediational sequence is more difficult for a given
task than the set of single—stage associations, then ac—
quisition of that task by the single-stage process should
be completed more rapidly than by the two-stage process,
and so the transfer characteristics of the learning would
fit the single-unit theory.

The Kendlers' diagram of their model made it clear
that they consider a single-stage association to be formed
for each stimulus pair. The number of associations to
be formed in a two-stage sequence, then, is just one more
than the number for a single-stage process, since the
mediating response has to be associated with each stim-
ulus and the stimulus component of the mediator must be
associated with the overt response. Dre—experimental
associations probably reduce the difficulty of the mediated
solution, however. The other, and probably more important,
source of difficulty in the mediational process is the
difficulty of extinguishing irrelevant mediators. If,

as the Kendlers' model implies by its notation (rsize),

-14-

a mediator correSponds to a stimulus dimension, then the
difficulty of extinguishing the inappropriate mediating
responses increases with the number of irrelevant dimen-
sions.

The ratio of the number of stimuli to the number of
dimensions therefore seems a particularly important var—
iable in concept-attainment experiments. Since the num-
ber of stimulus objects or figures that can be constructed
from N binary dimensions is 2N, the ratio is 2N/N, a
strictly increasing function for positive N, and so may
be manipulated by simply varying N. A more detailed quan-
titative development of the relationship between difficulty
of acquisition and the two parameters (number of associa-
tions and difficulty of extinction) would make it possible
to predict the number of dimensions at which the advantage
switches from one mode of learning to the other. The
research effort required for the collection of the data
necessary to do the complete job, however, should not
be spent if a rough test of the assumptions of the above
model can be made more efficiently. In particular, it
seems appropriate at this point to replicate previous
studies with a larger value of N, since there is no evidence
that young children actually behave differently from adults
when N > 2.

The model upon which this study is based, then, is

a modified version of the Kendler model, which acknowledges

-15-

that gs may be able to select, consciously or otherwise,
their mode of learning, and so an individual §_capable

of performing as a two-stage learner under some conditions
may also be capable of performing as a single-stage learner
under other conditions. Considering the two Kendler models
as representing alternative processes both available to

the same organism makes it possible to derive from the
assumptions of the Kendler model the conclusion that the
mode of learning selected depends upon the number of stim-

ulus dimensions.

DESIGN AND HYPOTHESES

Presenting stimuli with a larger number of dimensions
than were used in the studies cited above should, accord-
ing to the argument in the preceding section, increase
the probability of mediation. Four binary dimensions
were used to construct the figures of this study. The
dimensions, and their values, were color, red and green;
size, large and small; shape, square and circular; and
border, present or absent. Figures were presented in
pairs. Combinations were constrained as follows: members
of the pair were alike in shape and in presence or absence
of a border and differed both in size and in color. These
constraints eliminate two dimensions, color and shape, as
possible criteria by which §_could choose, and so serve
to decrease the number of potential mediators. Both the
increased number of dimensions and the constraints on
pairs should increase the ratio of the number of stimuli
to the number of potential mediators, and so increase the
probability that an organism able to mediate will do so.

The transfer design of the Kendler—D'Amato study was
selected for this experiment. As compared to the Optional
shift design (Kendler, Kendler, and Learnard, 1962), the

transfer design has the advantage of giving more learning

-16..

-17-

data in that §s can be run to criterion or at least to
many trials, rather than terminated at some small arbitrary
number of trials. Information about the shift selected
by the g during the trials with a reduced stimulus set
may also be recovered, as readily as in the optional shift
design. It is true that selective reinforcement may be
expected to change response probabilities in Phase 3,
but the effects, in the optional shift design, Of uncon-
ditional reinforcement of either choice response to half
of the set of stimuli are not known either. The only
uncontaminated measure of Ss' state at the end of Phase 2
(the phase in which half of the stimuli are removed) is
the trial 1 response in the next part, Phase 3, and this
measure may be used in the transfer design without the
sacrifice Of information on gs who require many trials
to reach criterion. Although the transfer design is used,
it is convenient to describe the experiment in the terms
coined for the Optional shift design. "Phase 1" refers
to the training task, "Phase 2" to the early trials of the
test task, in which half of the stimuli are set aside,
and "Phase 3" to the latter part of the test task, in which
the removed stimuli are returned to the active stimulus
set.

The stimulus conditions of this study, as explained
above, lead to the expectation that more gs in the pre—

sent experiment than in the Kendler-Kendler-Learnard study

-18-

will perform reversal shifts. Because of procedural dif-
ferences, and particularly the difference between the
dependent variables, such a direct comparison does not
seem appropriate. A more conservative test of the model
is to compare the obtained probability estimate with
P(reversal) = .5. If gs are in the pre-mediational stage
hypothesized by the Kendlers and their associates, less
than half should perform reversal shifts. One prediction
that follows from the assumption that P(reversal) > .5
in this study is that more §§ will make trial 1 responses
in Phase 3 that are consistent with a reversal shift than
will make trial 1 responses consistent with a nonreversal
shift. Those gs in the reversal group who have performed
a reversal shift in Phase 2 have no further learning to
accomplish in Phase 3, while those in the nonreversal group
who have performed a reversal shift then have to perform
a nonreversal shift in Phase 3 in order to respond cor—
rectly. If P(reversal) > .5, it follows that performance
in Phase 3 will be facilitated for more gs in the rever-
sal than in the nonreversal group, and so the prediction
is that the reversal group gs require fewer trials to
criterion than those in the nonreversal group.

Another mediational theory relevant to this experi—
ment is that of Mowrer (1960). Mowrer argues that learn-
ing, even in rats, is accomplished through a mediating

state which he calls expectation (p. 287). The importance

 

-19-

of the expectation construct to learning theory is not

in acquisition, according to Mowrer, but in various forms
of unlearning, including counter—conditioning, of which
concept shifts are a species. This framework suggests
the significance of one of the parameters of the Jeffrey
(1965) study. Jeffrey's alteration of the Optional shift
design in his Experiment II, in which he used circles as
stimuli in Phase 1 and squares in Phases 2 and 3, had at
least two separable effects: it added a new dimension

of variation to the stimulus set for the whole experiment
and it provided a procedural change at just the point
where gs were required to unlearn one task and to learn

a different set of response contingencies given the same
dimensional variations of the old task. If §.continued,
on trial 1 of Phase 2, to respond as in the original task,
Mowrer's theory would lead us to expect a change in S's
expectation, because of the close conjunction of §fs first
error in several trials with a change in the stimulus
situation.

It is possible to manipulate dimension and signal,
the two aspects Of Jeffrey's innovation, separately if
information about task change is given verbally. Half
the gs in each shift group were told at the end of Phase 1
that they had learned that "game" (the Phase 1 task) very
well and were invited to play a similar game. For the

other half Of the §s there was no pause or other break

-20-

in the procedure at the point of transition from Phase 1
to Phase 2. If this information changed.§fs expectation
about the appropriate choice, then the unlearning task
would be avoided (whether extinction, as the Kendlers
regard it to be, or counter-conditioning) and learning
would be facilitated for the informed group.

Another dependent variable Observed in this exper—
iment was §fs tendency to perseverate on a response after
it failed to be reinforced. There is a good deal of
empirical data demonstrating the tendency Of infrahuman
organisms to emit intensified perseverative responses
in operant-transfer situations (Brown, 1961; Mowrer, 1960).
Mowrer discussed the affective aspect Of expectation,
stating (p. 476), "If a rat is not sure whether a given
trial is the one which is to result in reinforcement,
he is not particularly frustrated (surprised, disappointed)
if there is no reinforcement, but if, on a particular
trial, reinforcement i§_confidently expected and does not
materialize, then there will be maximum frustration and
correspondingly greater counter-conditioning (extinction)
of hope and habit strength." Elsewhere (p. 405 ff),
Mowrer argues that the immediate result of this frustra-
tion is anger, which leads to aggression aimed toward
the frustrating Obstacle.

If the information provided in this experiment changed

§fs expectation concerning the stimulus consequences of

-21-

his responses, then Mowrer's theoretical treatment of
expectation implies that the usual response pattern asso-
ciated with cessation of reinforcement should be altered.
A specific aspect of this response pattern that is described
by Mowrer (1960) and Brown (1961) is the intensification
of the previously reinforced response. Although the
stress in both Mowrer's and Brown's discussions was on
such measures as Speed or force of a response, it seems
reasonable to extend Mowrer's explanation Of frustration
behavior during extinction to include the occurrence of
perseveration on the response that is no longer reinforced.
Evidence of an increased response rate at the outset of
extinction appears in data on children in the second year
of life from a study by Weisberg and Fink (1966). The
measure used in this study was the mean number of incor-
rect responses averaged over trials on which at least one
error occurred. Since the correction method was used,
each trial consisted of a string Of incorrect responses
followed by a correct response, where the length of the
string of errors could be zero or greater. The effect
typically fades quickly during extinction, and so the mean
was computed over the first several trials only. Every g
who made at least one error required at least nine trials,
and so it was convenient to compute the measure over the
first nine trials for each §_who made at least one error.

The design of the experiment included two independent

-22-

variables; shift task and information, with two levels

Of each. All Se in the reversal group were required to
learn to select the small figure in Phase 1 and the large
figure in Phase 3. Those in the nonreversal group were
required to select the green figure in Phase 1 and the
large figure in Phase 3. For Phase 2 all pairs were large
red-small green combinations and all Se in both groups
were required to select the large red figure on each trial.
The only difference between the two tasks, then, was the
Phase 1 discrimination. Dependent variables for testing
predictions were number of trials to criterion on Phase

3 and proportion Of gs classified as having performed

a reversal versus a nonreversal shift during Phase 2 for
the shift type variable, and number of errors per error
trial and number of trials to criterion in Phase 2 for

the information variable.

METHOD

Apparatus

A block diagram illustrating the components of the
apparatus and communication among them appears in Fig. 1.
Fig. 2 shows the machine-§_interface. Stimuli were dis-
played by rear projection of 35 mm. slides on a screen
of Opal glass. The screen was located 1/4 in. behind the
window in the front of the cabinet shown in Fig. 2. The
cabinet was constructed of Masonite and wood. The window
consisted of three clear plastic panels 1/4 in. thick.

A 4 x 4 in. square portion of each panel was visible
through the 4 x 12 in. aperture in the front of the cabinet.
The center panel, which was not used in this study, was
blocked so as to be immovable. The left and right panels
were suspended from hinges. The bottom of each panel

was held in position against the front of the cabinet

by the spring action of a.Microswitch. Pressure on either
panel closed an electrical circuit which delivered a single
pluse to the control apparatus. Subjects were instructed
to press a panel to indicate selection Of the stimulus
appearing on the screen directly behind the panel. Spa-
tial separation of stimulus and re3ponse was thus mini-

mized, a condition that has been found (Jeffrey and Cohen,

-23-

-24-

 

 

 

 

 

 

 

 

 

 

 

 

 

 

E
SWITCHING MACHINE-£7.
DEVICE INTERFACE
PROJECTOR CONTROL PROJECTOR
A UNIT B

 

 

 

 

 

 

 

 

 

 

 

DATA
RECORDER

(TAPE PUNCH)

 

 

 

Figure 1. Block diagram of experimental apparatus.

-25-

 

R

RESPONSE PANELS

 

 

 

Figure 2. Apparatus-subject Interface.

-26-

1964) to facilitate learning in children.

Reinforcement contingencies were controlled by E.

A switching device, consisting Of four 5/8 in. doorbell
pushbuttons installed in a l x l x 4 in. rectangular prism
of plastic, was held in Efs hand. This device, shown in
Fig. 3, delivered signals to the control unit. The E
selection was entered in the corresponding memory unit
Of the control device and consequently transmitted con—
tinuously to the recording device until that memory unit
was cleared. The memory unit could be cleared either by
Efs selection of the complementary switch button or by
§fs correct response. For instance, if E pressed the
button specifying reinforcement Of selection Of the left
panel, this selection by §_was cancelled if §_pressed
the button for the right panel or if §_pressed the left
panel.

A correct response by §_also advanced the slide tray
of the Carousel projector, bringing the next stimulus
into position. The correction procedure, which has been
shown in another situation to facilitate learning in
children (Suppes and Ginsberg, 1962), was thus incorporated
into the design of the apparatus. Presentation of the
new stimulus required a new selection by'E as well as
by‘g.

Two slide projectors were used in order to provide

the two sets of stimuli required for the experiment, the

-27-

PROJECTOR
A ON, B OFF

PROJECTOR
B ON, A OFF

    
   
 

LEFT-HAND
PANEL CORRECT

RIGHT-HAND
PANEL CORREC!

 

Figure 3. Experimenter's switching device.

-28—

full set and the set remaining after half of the slides
were removed. A correct response advanced only the pro-
jector that was on. Two Of the buttons on E's switching
device enabled E to switch either projector on, but not
both.

The E's switching device was connected to the control
unit by a 5-wire electrical cable, allowing the plastic
cabinet containing the circuitry to be placed at some
distance from S. Cues provided by relay noise were thus
minimized.

The data recording device was an 8-channel Teletype
paper tape punch. The six data channels used were acti—
vated by the control unit described above. Two channels
indicated which projector lamp was on; two channels in—

dicated E choice; and two, §_choice.

Subjects

Children between 37 and 48 months of age from Uhi—
versity housing and from nursery schools in and near

Lansing, Michigan1 served as gs. Twenty-three gs failed

 

lParticipating schools in the Lansing area were Jack
and Jill Playhouse, MSU Cooperative Nursery,‘Wesley Coopera—
tive Nursery, Carol Lee Nursery, Miss Cheryl's Play Center,
YWCA Meridian COOperative Nursery, and Pennway Cooperative
Nursery. Eleven §s were procured from another nursery
schOOl, but when only five of them were present on the
day of the first experimental session (for the warm-up
task), and only two Of these passed the warm-up task,
research at that location was discontinued.

-29-

the warm-up task; 11 failed the Phase 1 task; five refused
to participate in Phase 1, and one in Phase 2; five gs
were lost due to apparatus malfunction; and four were lost
due to experimenter errors. Twenty-eight gs survived
beyond trial 1 Of Phase 3, and so were included in all
analyses except that Of trials to criterion on Phase 3.
One §_was excluded from the trials analysis due to experi—
menter error during Phase 3, and one due to apparatus
malfunction; 26 gs were therefore included in the trials
analysis. Subjects were assigned to groups without regard

to sex or age; the mean age for the 28 §s was 43.9 months.

Materials

 

All stimuli were presented by projecting 35 mm. slides
on the screen of the interface apparatus. Stimuli for
the initial discrimination task were photographs of pairs
of animal pictures, which were clipped from a child's
picture-story book. The pictures were lifelike rather
than stylized drawings. Each pair consisted Of the correct
Option, which was always the same picture of a dog, and
the incorrect Option, which was either a rabbit, a squirrel,
a parakeet, or a pony. The sequence of stimuli is listed
in Appendix A.

The concept task stimuli were pairs of geometric
figures differing in both size and color. Figures were

squares and circles colored either green or red. A large

-30-

green figure was always paired with a small red figure,
and a small green figure with a large red figure. The
two figures were always alike with respect to shape and
presence or absence Of a border; borders were white and
backgrounds were black. Concept stimuli were produced
by photographing geometric forms cut from red and green
construction paper. Although color-matched light and film
were used, a color shift toward blue was apparent in the
slides, making the reds pale, the greens bluish, and the
backgrounds more nearly navy blue than black. Large
squares were approximately 2 x 2 in. when projected under
the constraints of the apparatus configuration (Fig. 4),
and small squares were approximately 1 x l in. Circles
were approximately 2-1/8 in. and l—l/l6 in. in diameter

for large and small respectively.

Procedure

The‘g visited nursery school classrooms and homes
of gs not enrolled in nursery school in order to estab-
lish rapport. As soon as possible after the rapport
session, usually 1—5 days, the first experimental session
was held. For one nursery school, the first to partici-
pate, repeated apparatus failure On scheduled experimental
days extended the period between first contact and first
experimental session, but the continued coming and going

of‘E probably enhanced rapport. In the first experimental

-31-

 

 

 

 

 

 

 

 

 

 

    
   

 
 
 
  

 

 

   

Figure 4.

Apparatus

 

UNIT

 

 

 

configuration (plan

PRO—
" JECTOR
MACHINE» II
SUBJECT
CHAIR SUBJECT TABLE
INTER- PR0-
J FACE I” ’7 I JECTOR
‘UNcH \
L. _ .1 ,
I
I
CONTROI

 

 

 

view).

 

-32-

session the discrimination task with the animal pictures
was administered. Before presenting instructions and a
demonstration, E offered each s a balloon and a piece of
candy. These were then placed on the table beside the
response device until completion of the game.

The §_gave instructions and a demonstration from
memory in order to facilitate communication with.§, While
there was undoubtedly some variation a typical instruction
ran approximately as follows:

This is the game I've told you about.
Here's how we play the game. YOu see we have
two pictures here. The way we play the game is
to try to pick the right one. we just push like
this (pressing the correct panel, which was on
the left), and if we pick the right picture, it
changes. (The correct panel, again on the left,
was again pressed.) See, every time we pick the
right picture, it changes. If we pick the wrong
one, we can tell we made a mistake (in a manner
implying that the mistake is not at all serious)
because nothing happens; (pressing the correct
panel, again on the left) it only changes when
we pick the right one. Sometimes the right pic-
ture is here (pointing to right panel), so we
don't pay any attention to where it is; we just
look at the picture (pressing the correct panel,
now on the right). Now let's push it with your
hand (taking §fs hand and pressing the correct
panel—-on the right--with it). Okay, now ygu
may do it by yourself.

The §_Operated the response device with no further
instructions, except that.§ answered questions as noncom-
mittally as possible, prompted S3 to continue responding
whenever necessary, attempted to persuade gs to continue
if they indicated a desire to stOp, and provided informa-

tion, when necessary, to enable §s to operate the device

-33-

properly. Some gs attempted to press both panels simul-
taneously at the outset of the first task, and so had to
be instructed to press only one at a time.

The first session was terminated when §_reached a
criterion Of 8 consecutive correct responses, completed
80 trials, including the demonstration by E, without reach-
ing the criterion, or refused to continue. A second warm-
up session was administered for gs who terminated by
reaching the 75-trial cutoff without reaching criterion.
Any §_who failed to reach criterion on the second attempt
was excluded from the classification task. After learning
the warm—up task to criterion, gs were invited to partici-
pate again in the final session.

In the final session the warm-up task was administered
again, just as in the warm-up session(s). After S reached
criterion, E told S that he had learned that game very
well and invited.§_to play the same kind of game "with
_§hi§ kind of picture," switching to the projector loaded
with the Phase 2 (reduced) stimulus set and pointing to
the screen. After §_agreed, E asked §_to wait a moment
while §_changed the tray on the other projector, loading
it with the full set of stimuli used in Phases 1 and 3.

As E resumed his seat beside s, he switched projectors again,
so that the figures for the first trial of Phase 1 appeared
on the screen.

For the reversal group the Phase 1 task was to learn

-34-

to select the smaller figure; the nonreversal group was
required to select the green figure. When §_reached a
criterion of 8 consecutive correct responses, Phase 2,

the transfer task with reduced stimuli, began. Any S

who failed to reach criterion within 100 trials was ex—
cluded from the remainder Of the experiment. Since it
was impossible under the circumstances to keep an accurate
trial count, §s were run beyond 100 trials, but a 100-
trial cutoff was Observed in processing the data.

When §.reached criterion on Phase 1, the treatment
differed for the informed and uninformed groups. For gs
in the informed group, E switched to the projector holding
the reduced set Of stimuli used in Phase 2, informed'g
that he had learned the game, and invited S to play another
game with the same pictures, adding the qualification,
"But this time we play it a little differently." For the
uninformed group, §_proceeded immediately with Phase 2
without comment. Phase 2 was terminated when S reached
the criterion Of 8 consecutive errorless trials, completed
80 trials without reaching criterion, or refused to con-
tinue. For Phase 2, as for the initial discrimination
task, the count Of 80 trials was kept accurately by leav-
ing the 8lst slot in the slide tray empty.

When §_reached criterion on Phase 2,.§ immediately
switched again to the projector that was loaded with the

stimuli for Phase 1 and 3, with no break in procedure.

-35-

Phase 3 was terminated when §_reached criterion (8 con-
secutive errorless trials), completed 100 trials without
reaching criterion, or refused to continue. As in Phase

1, it was impossible to count 100 trials accurately.

RESULTS

It was predicted from the model described above that
more than half of the gs in this study would perform a
reversal shift during the second phase of the experiment,
in which either a reversal shift or a nonreversal shift
would have led.§ to respond correctly. One of the experi-
mental hypotheses derived from this prediction was that
the first response in Phase 3 would be consistent with a
reversal shift for more than half Of the gs. For the
reversal task group, the first task was to select the small
figure, and so gs in that group who selected the large
green figure were classified as having reversed during
Phase 2. For the nonreversal group, the first task was
to select the green figure, and so §s who selected the
small red figure were classified as having reversed during
Phase 2. For purposes of this analysis, trial 1 Of Phase
3 was taken to be the first trial on which the stimulus
was a large green-small red combination of figures after
§_reached Phase 2 criterion. The apparatus used in this
study provided no way of ensuring that the first figures
di5p1ayed after switching to the full set of stimuli would
be that combination. Some gs were given as many as three

trials between the Phase 2 criterion trial and the first

—36-

-37-

Phase 3 trial. Table 1 shows the group frequencies for

gs classified by information level and type of inferred

shift.
Table l
Frequencies for Groups Classified by
Information and Inferred Shift
Informed Uninformed Marginals

Inferred

9 10 19
Reversal
Inferred

5 4 9
Nonreversal
Marginals l4 14 28

It is apparent that the frequencies are virtually
identical for the two information levels. Since the single-
stage Kendler model is consistent with any reversal-shift
probability less than .5, the appropriate statistical
analysis is a one-tail binomial test on the marginals for
type of inferred shift. The probability of a frequency
of 19 or more reversals in a binomial distribution with
n = 28 and p = .5 is .044. The single-stage model was
rejected, therefore, in favor of an alternative that leads

to the prediction p > .5, such as the two-stage model.

~38—

The second hypothesis from the modified Kendler model
was that gs in the group assigned the reversal task would
complete Phase 3 in fewer trials than those assigned the
nonreversal task. Group means and standard deviations
for trials to the last error on all tasks appear in Table
2. For this analysis, trials are numbered from the trials
immediately following the Phase 2 criterion trial. Although
this method Of counting trials exposes gs to unequal num-
bers Of large green-small red figure combinations, it has
the advantage of equalizing the lengths of the combined
criterion runs for Phases 2 and 3. Unequal criterion runs
would allow the probability Of criterion performance by
chance to be greater for some §S than for others. In a
two-choice task, the probability of such a spurious cri—
terion run is pn, where p is the probability of a correct
choice by chance and n is the number of trials to criter—
ion. The 2 x 2 analysis of variance shown in Table 3
yielded a significant F ratio for type of shift. Both
measures, then, confirm the hypothesis that the probabil-
ity of a reversal shift in Phase 2 was greater than .5.

Even though the majority of the gs apparently per—
formed reversal shifts during Phase 2 (the Optional shift
phase), it would be desirable to be able to explain those
nonreversal shifts that did occur. Under the assumptions
of the Kendler model, nonreversal shifts become more prob-

able if §_fails to use mediating responses. There are

Phase 1

Phase 2
Informed

Uninformed

Phase 3
Informed

uninformed

-39-

Table 2

Trials to Last Error for the

Three Concept Tasks

Reversal

Standard
Deviation

10.85

8.46

1.95

Nonreversal
Standard
Mean Deviation
13.73 16.77
3.75 4.37
6.71 6.34
40.86 35.80

64.17 20.03

-40-

Table 3

Analysis of Variance on
Phase 3 Trials to Last Error

Source df SS MS F
Information (I) 1 411.19 411.19 < l
Shift Type (S) 1 14696.76 14696.76 32.08**
I x S l 1519 1519 3.32*
Error 23 10079 438.22

**
p < .0005

*.05 < p < .10

three ways in which this failure might occur: if §_satis-
fies the requirements of the task before he resorts to

the use of the mediator, if the response is not in st
repertoire, or if the response becomes temporarily inac-
cessible for use as a mediator. Elimination Of the first
of these conditions was a major purpose of constructing

a set Of stimuli varying on four dimensions. The remaining
two conditions lead to hypotheses easily tested with the
data Of this experiment. In view of the two major results
already presented above, the hypothesis that P(reversal)

> .5 was strongly supported. The results were consistent,

therefore, with the assumption that the experimental task

-41-

was constructed so as to be more difficult for Se who did
not mediate than for those who did. Then if any g either
lacked the mediating response or if it became inaccessible,
that S should learn the tasks for both Phase 1 and Phase

2 more slowly than if he could use mediation. A EQEEHEQE
hypothesis, then, was that gs who had performed nonrever-
sal shifts in Phase 2 would require more trials to complete
the earlier Phase 1 and Phase 2 tasks.

The test of this hypothesis was a comparison of mean
numbers Of trials to criterion in the two inferred shift
groups. Those gs who were classified as having performed
a nonreversal shift in Phase 2 should, under the above
hypothesis, require more trials to criterion on the Phase
1 task and the Phase 2 task than those who were classified
as having performed a reversal shift. Means and standard
deviations for §s grouped by inferred shift and by informa-
tion level appear in Table 4. The t test on the mean for
the Phase 1 task yielded a t value less than 1, and the
2 x 2 analysis of variance (inferred shift by information)
on the trial scores for Phase 2 yielded F ratios less than
1 for both main effects, information and inferred shift,
and their interaction.

Other experimental hypotheses dealt with the informa-
tion variable. If information served to eliminate the
unlearning trials immediately following a task shift, then

transfer should have been facilitated for the informed

-42-

Table 4

Trials to Last Error for Se Grouped by
Information and Inferred Shift

Inferred Reversal Inferred Nonreversal
Standard Standard
Mean Deviation Mean Deviation
Phase 1 17.32 65.23 20.00 36.58
Phase 2
Informed 12.40 24.17 15.25 30.23
Uninformed 17.44 1.85 12.40 5.93
Phase 3 Reversal Task
Informed 8.75 8.50 .87 .347
uninformed 0 0 .70 O
Nonreversal Task
Informed 28.80 29.50 1.838 .184

Uninformed 60.25 16.00 1.846 .173

-43-

group in Phase 2. If the experience of a shift without
warning at the beginning Of Phase 2 led to expectation

of a change of task immediately upon the occurrence of

an error in Phase 3, then transfer should be facilitated
for uninformed.§s in Phase 3. Means and standard devia-
tions for trials to criterion in Phase 2 and 3 appear in
Table 2, already referred to above. A 2 x 2 analysis Of
variance for Phase 2 trials to criterion yielded an F
ratio less than 1 for each main effect (shift task and
information) as well as for their interaction. For Phase
3, however, the significance level Of the F ratio for the
interaction was .05 < p < .10, as shown in Table 3.

The data of Table 2 also show a result that requires
further analysis of the data. The group means and the
correSponding standard deviations are closely related,
with only one such pair of quantities, the mean and stan-
dard deviation for the uninformed nonreversal group in
Phase 3, deviating appreciably from the pattern. The
correlation with that pair excluded is .98, and with the
pair included it is .72. The method usually suggested
(e.g., Anderson and Bancroft, 1952, p. 222; Snedecor, 1956,
p. 320) for correcting this particular kind Of heterogeneity
of variance is the log transformation. The transformed
data appear in Table 5, and the analysis appears in Table
6. The main effect due to type of shift task is signifi-
cant at the .01 level, and its interaction with informa-

tion is significant at the .05 level.

-44-

Table 5

Log Trials to Last Error for the
Three Concept Tasks

Phase 1

Phase 2
Informed

Uninformed

Phase 3
Informed

Uninformed

Mean

.632

.627

.901

.714

.200

Reversal

Standard
Deviation

.5413

.5167

.2623

.6013

.3995

Nonreversal
Standard
Mean Deviation
.864 .5710
.554 .3410
.759 .3603
1.338 .6579
1.798 .1296

-45-

Table 6

Analysis of Variance on
Transformed Data

Source df SS MS F
Shift (S) 1 7.961 7.961 33.63**
Information (I) 1 .047 .047 < l
S x I 1 1.531 1.531 6.47*
Error 22 5.208 .226

*

p < .05
**

p < .01

Another important aspect of the strong relationship

between the means and the standard deviations is the kind

Of distribution suggested by this relationship. A well-

known discrete probability distribution for which the mean

and the standard deviation are closely related is the

geometric distribution. The mean and variance for the

distribution
p. 28). For
approximates
for p = .05,

is 19. This

are %-and'%%E , respectively (Parzen, 1960,
fairly small values Of p, the mean closely
the standard deviation, _£%;§L.; for instance,
the mean is 20 and the standard deviation

distribution of trials to the last error is

consistent with a model which assumes that the conditional

-46-

probability, p, Of mastering the acquisition task, given
that mastery has not already occurred, is constant over
trials. A development of the geometric distribution from
such a sequence of Bernoulli trials is given by Feller
(1950, p. 156) in more general terms.

It was also predicted on the basis of a generaliza—
tion from Mowrer's theory and Of data cited by Mowrer
(1960), and Brown (1961), and data presented by‘Weisberg
and Fink (1966) that information would affect gs' tendency
to perseverate on incorrect responses that had been rein-
forced On previous trials. The indicator selected for
this tendency was the mean number of incorrect responses
made on each error trial before §_terminated the trial
with a correct re3ponse. In this experiment any §_who
made one or more errors required at least nine trials to
reach criterion. In order that the same number Of trials
be used to compute the measure for each.§, the number Of
incorrect reSponses in the first nine trials was divided
by the number Of error trials for the first nine trials
only and data for gs who made no errors was excluded from
the analysis. Information group means and standard de-
viations appear in Table 7. Differences were not sig-
nificant by t test.

The shift-task variable was confounded, in this study,
with a difference between Phase 1 tasks. The relevant

dimension for that task was size for one shift-task group

-47-

Table 7

Number of Error Responses
Per Error Trial

Informed Uninformed
Standard Standard
Mean Deviation Mean Deviation
Phase 2 1.81 5.79 1.95 5.22

Phase 3 1.13 1.23 1.33 1.23

and color for the other. A comparison Of the two groups
by t test showed that the difference was not significant,
but p was less than .25. In view Of this difference it
seems necessary to consider whether the 11 §s who failed
to perform the Phase 1 task would have changed the major
outcomes of this experiment if they had participated in
the Phase 3 task. The comparisons Of the two inferred
shift groups in terms of Phase 1 trials to criterion,
discussed above, provides no evidence of a difference
between the two groups. (The frequencies shown in Table
8 also fail to indicate any difference in inferred shift
frequencies for the two shift-task groups.

Another potential source of bias not controlled ex-
perimentally was the possibility that gs selectively

failed the discrimination task in such a way that the

-48-

Table 8

Frequencies for Groups Classified by
Shift Task and Inferred Shift

Reversal Nonreversal Marginals
Task Task

Inferred

9 10 19
Reversal
Inferred

4 5 9
Nonreversal
Marginals 13 15 28

major results of the study were affected. A t comparison
between the two inferred shift groups, however, did not
indicate any difference in the number of trials to criter-
ion on the discrimination task.

Some of E's informal observations during the exper-
iment seem relevant to the problem of planning similar
studies, and so are presented here. The design of the
apparatus used in this study was such that §_was required
to specify the reinforcement contingency at each trial
and to switch from one projector to the other at the end
of each phase of the concept task. If §_switched projectors

quickly enough after the last trial of the criterion sequence,

-49-

the lamp of the projector just switched Off would fade
out before the slide advance process was complete, but
if Efs reaction was slow, the lamp would have a sufficiently
bright afterglow to project an image, and § would see a
double image for a half second or so. Double image pre-
sentations could have distracted §_from the task, or
particularly in the case of the uninformed group, could
have provided a change-Of-procedure cue. Whenever E noted
the occurrence of the double image, he therefore excluded
the §_from the experiment, counting the loss of'§ as due
to experimenter error.

Another observation concerning the design of the
apparatus was that some gs would press both right and
left panels simultaneously. As reported in the procedure
section, §_advised these gs not to press simultaneously,
explaining that the machine would not work properly if
they did so. The effect of such pressing behavior was
intermittent occurrence of the slide advance Operation,
and so it seemed advisable to eliminate the behavior if
possible. The §_did not alter the pre—task instructions
in an attempt to avoid such behavior, however, because he
suSpected that expanding the set of instructions would
result in a loss of Ss' interest in the "game." The simul-
taneous pressing of the two panels was observed to occur
only on the discrimination task, and Efs undocumented
impression was that this response appeared only when S_

had been unsuccessful and had become restless.

DISCUSSION

The two hypotheses concerning proportions of Se per-
forming each type Of shift during Phase 2 (the Optional
shift phase) were supported unequivocally. Trials to
the last error for the two shift tasks differed signifi-
cantly (p < .0005), and the proportions of gs classified
as having performed reversal and nonreversal shifts, re-
spectively, differed significantly (p < .05). Both the
frequency of classification as inferred reversal or non—
reversal shift and trials to the last error supported
the hypothesis that the probability of a reversal shift
was greater than .5.

Comparison Of these results with those of previous
studies with children shows the importance of the number
of stimulus dimensions used in determining the tendency
to mediate. An important implication Of this finding
is the necessity of taking the stimulus dimensions into
account in the design and interpretation of concept-
learning studies and other problems in which mediational
processes are Of interest. The dependency of the media-
tion tendency upon the number of dimensions is a relation—
ship that can be represented as a quantitative function,

P (reversal) = f(N), and so a reasonable next step in the

-50-

_51-

investigation of the problem is the collection of data
for several values of N, the number Of dimensions.

The expectation that the §s classified in the rever—
sal group would display superiority in the acquisition
task of Phase 1 or in the Optional transfer task of Phase
2 was not confirmed by the data. The failure of this
hypothesis, however, gives rise to further questions.

It seems likely that the tendency to perform a reversal
shift in the optional situation is due to some cause other
than failure to employ mediation, but further confirma-
tion Of this assumption is certainly desirable. Another
question for further research is whether the tendency
toward a reversal versus a nonreversal shift is a char-
acteristic of individual gs or simply a parameter Of a
stochastic process. More extensive data on individual

gs is needed to answer this question.

The effect of information, indicated by its inter-
action with shift task in the analysis Of untransformed
scores, was not statistically significant, and so could
be attributed to sampling error. The analysis of the
transformed data, however, showed a significant interaction
between information and type of shift. The absence of
any effect in Phase 2, immediately after the information
was communicated, mitigates against the conclusion that
information was generally facilitative or motivating.

The kind of interaction found in Phase 3 provides a clue

-52-

to the nature of the underlying process. The effect of
information was facilitative for the nonreversal group
and detrimental for the reversal group. A possible inter-
pretation of this result is that some gs understood the
information that a new game was beginning to mean that
the Old choice criterion should be abandoned, and so
adopted a new criterial dimension; that is, they performed
a nonreversal shift. But Table 1 does not support this
assumption, and if all gs who were inferred to have per-
formed a nonreversal shift are excluded from the analysis,
the interaction pattern remains, as is apparent in Table 4.

NO differences in repetition of error responses
appeared in either Phase 2 or Phase 3. This negative
result may be attributed to the absence of any effect of
information on §§' tendency to become frustrated. It is
also possible that the measure selected was not sufficiently
sensitive, since it was an average over the first 9 trials
and it did not take response intensity into account.

The tests for effects due to selective attrition
were all negative, but there remains the possibility that
those gs who failed the discrimination task were qualita-
tively different from those who passed, although the per—
formance Of the latter gs was unrelated to their tendency
toward one or the other type of shift. A developmental
variable Of some kind may well be involved in the selection

Of gs on this task or understanding of behavior appropriate

-53-

to a game. The only kind of difference that would be
relevant to this test of the Kendler model, however, would
be in Ss' tendency to mediate. The discrimination task,
requiring only that‘g consistently select Qgg_picture
(the picture of a dog), should be easily learned through
the single-stage associative process.

Selection on the Phase 1 task is less easily dis-
missed. If the task was very difficult for gs who did
not use mediators and relatively easy for those who did,
then the taSk would select for the ability to mediate.
This assumption is double-edged, however, and implies
that gs who succeeded probably did so through mediation.
The large proportion of gs who succeeded (29 out of 40)
then indicates that gs in the population who survived the
discrimination task use mediation with high probability.

The strong relationship between group means and their
associated standard deviations was an unanticipated result,
but the evidence that the probability function is geometric,
provided by this relationship, is of considerable import-
ance. One set Of assumptions from which the geometric
distribution can be derived is that the sequence of trials
is a sequence Of Bernoulli random variables, in which the
probability, p, of success is constant over trials. The
result is consistent with an all-or-none learning model,
then, in which the probability of solution is constant

for all trials. Confirmation of the assumption of inde-

-54-

pendent Bernoulli trials can be Obtained from other sta-
tistics of the data, but generally requires fairly large
samples. Development of this argument is therefore left
for further research.

The major outcomes of this study, then, indicated
the effectiveness of both Of the manipulated variables,
since the main effect for type of shift and its interaction
with information were both significant. The hypothesis
that most SS in the sample employed mediation in the solu-
tion Of the problem was supported. The contrast between
these results and those of the previous studies already
cited has several important implications. For researchers,
there is a demonstration of the danger of stereotypy of
method and at least a suggestion that underestimation of
gs' abilities can be detrimental to the design of an ex-
periment. TO the extent that the experimental situation
resembled 'real-life' learning situations, the implica-
tion is that even three-year-Old children may use simpli-
fying principles to enhance learning when it is to their
advantage to do so, and the process, even at this low
level of linguistic sophistication, may be altered by

instructions.

REFERENCES

Andreas, B. G. Experimental Psychology. New York: Wiley,
1960.

Brown, J. S. The Motivation of Behavior. New York:
McGraw-Hill, 1961.

Buss, A. H. Some determinants of rigidity in discrimina-
tion-reversal learning. Journal of Experimental
Psychology, 1962, 44, 222—227.

 

. Rigidity as a function of reversal and nonrever—
sal shifts in the learning of successive discrim-
ination. Journal of Experimental Psychology, 1953,
45, 75-81.

. Reversal and nonreversal shifts in concept
formation with partial reinforcement eliminated.
Journal of Experimental PsychologX, 1956, 53,
162-166.

Feller, W. An Introduction to Probability Theory and Its
Applications. New YOrk: Wiley, 1950.

 

Harrow, M., & Friedman, G. B. Comparing reversal and
nonreversal shifts in concept formation with par—
tial reinforcement controlled. Journal of Experi-
mental Psychology, 1958, 55, 592-597.

Jeffrey, W. E. Variables affecting reversal-shifts in
young children. American Journal Of Psychology,
1965..§Z, 577-580.

 

Kelleher, R. T. Discrimination learning as a function
of reversal and nonreversal shifts. Journal of
Experimental Psychology, 1956, 5;, 379-384.

 

Kendler, H. H., & D'Amato, M. F. A comparison of reversal
shifts and nonreversal shifts in human concept
formation behavior. Journal of Experimental
Psychology, 1955, 45, 165-174.

 

Kendler, H. H., & Kendler, T. S. Vertical and horizontal
processes in problem solving. Psychological
Review, 1962, 55, 1-16.

 

-55-

-56-

Kendler, T. S., & Kendler, H. H. Reversal and nonreversal
shifts in kindergarten children. Journal of Ex-
‘perimental Psychology, 1959, 55, 56-60.

 

Kendler, T. S., Kendler, H. H., & Learnard, B. Mediated
responses to size and brightness as a function
of age. American Journal Of PsychologY. 1962,
15, 571-586.

Kendler, T. S., Kendler, H. H., & Wells, D. Reversal and
nonreversal shifts in nursery school children.
Journal of Comparative and Physiological Psychology,
1960, 55, 83-88.

Mowrer, O. Hobart. Learning Theory and Behavior. New
York: Wiley, 1960.

Parzen, Emmanuel. Modern Probability Theory and its Ape
plications. New York: 'Wiley, 1960.

 

Snedecor, G. W} Statistical Methods. Ames, Iowa: Iowa
State University Press, 1956.

Spence, K. W. The nature of discrimination learning in
animals. Psycholpgical Review, 1936, 45, 427-449.

 

. Behavior Theory and Conditioning. New Haven:
Yale University Press, 1956.

 

Suppes, P., & Ginsberg, R. Application Of a stimulus
sampling model to children's concept formation
with and without overt correction responses.
Journal of Experimental Psychology, 1962, 55,
330-336.

 

Weisberg, P., & Fink, E. Fixed ratio and extinction per-
formance Of infants in the second year of life.
Journal Of the Experimental Analysis of Behavior,
1966, 5, 105-109.

ouflm
mom
zoom
HOHHHOOm
mom
uHQQmm
moo
zoom
HOHHHoUm
mom
mom
Comm
uwnnmm
mom
mom
mom
mom
Hmuuflovm
panama
mom

ufimflm

moo
uHQQmm
mom
mom
Ouflm
mom
zoom
mom
moo
UHHm
Hmuuflovm
moo
mom
uHQQmm
Unwm
pannmm
zoom
mom
mom

Hmnuflovm

 

ammo

 

uHQQmm
mom
Hmnuflovm
Comm
mom
zoom
pannmm
moo
mom
mom
panamm
mom
OHHm
pannmm
ouflm
zoom
Hmnnflovm
mom
mom
mom

#Qmfim

mom
Uuflm
mom
mom
Hmnuflovm
moo
mom
zoom
Guam
pannmm
mom
Hmnufloom
mom
mom
mom
mom
mom
Hmonflovm
OHHm
zoom

 

puma

 

panama
HOHHHoUm
mom
mom
zoom
ouflm
mom
mom
moo
zoom
mom
zoom
mom
mom
Guam
mom
mom
HOHHHOUm
zoom
uflnnmm

unmfl

mom
mom
uHQQmm
zoom
mom
mom
Hmuufloom
uHQQmm
OHHm
mom
Hmnnflowm
mom
zoom
HOHHHoUm
mom
HOHHHoOm
Comm
.mOQ
mom
mom

 

ammo

 

mom
mom
comm
mom
zoom
HOHHHoUm
mom
mom
zoom
Hmuuflovm
mom
mOQ
mom
pannmm
mom
OnHm
uHQQmm
moo
m8
moo

poms

xmma oOHumoHEHHOmHQ mom moomovmm OOHSEHum

mMmdB and mom mmozmbdmm mDHDZHBm
¢ XHQZMmm¢

uwnnmm
zoom
mom
HOHHHSOm
mom
mom
pannmm
Hmuuflovm
mom
mom
zoom
Guam
uwQQmm
mom
zoom
mom
mom
uHQQmm
zoom
UHHm

 

ammo

"H UHmm

-57-

-58-

zoom zoom moom
zoom moom zoom zoom zoom moom moom zoom moom moom zoom zoom moom
zoom zoom moom moom zoom zoom moom zoom moom zoom zoom zoom moom
zoom moom mooo moom zoom moom zoom zoom mOOm mooo moom zoom moom
zoom moom zoom mooo moom moom zoom moom zoom zoom zoom moom zoom
moom zoom ooom zoom moom zoom zoom zoom zoom moom zoom moom moom
zoom zoom moom moom moom zoom moom zoom zoom moom moom zoom zoom

.omzu HTUHOQ pom
ommow TEMm mop mo pom HOHOO pom mNHm mpwmommo no no no moowm .Omuoomumoooon on omo

HoQEoE woman oou mo oomumfluomoo woe .Ooflonomoo mo Hflmm 50mm mo HmQEoE puma mop zHoO

HTOHOQ oo «2 THOHHO "U oooum «O HHmEm "m
Hmonon um oumoom «0 Com um momma «m "TOOU
m com A moomom HON ooooovom moaofiflum "N uumm

AUOSGHDGOUV < NHQmem/w

_59-

20mg 20mg
mme Zdwm
206m mOUm

mdom
mUmA

mowq

.o>onm

oOHpHpomon m mom moooovom mop mo mama ooooom mow

zoom moom zoom moom zoom
moom moom zoom moom zoom

moom zoom moom zoom zoom

.ooooooom moHoEHpm moflomoomm

push omummm ommz umuoo mop mo

mofloflumsoo muooEoHo zuHOm mos

mowm
mvma 200m ZUMQ mumq
mdwm 200m 20mg deq

mama 20mg 200w 200m

mama
200m
mdwm

20mg

on» now no meow mop mm moHUOU

m mmmom mom moomovom moHoEHum

"m Doom

xomsamucoov o xmozoooo

APPENDIX B

RESPONSE SEQUENCES FOR ALL SUBJECTS

Explanation of Coding

1. Columns 1 and 2: subject identification.

2. Columns 3 and 4: task identification.

a.

Dn denotes the discrimination task, nth attempt
for that subject.

W denotes the warmup task (column 3 only).

Pl denotes the phase 1 task.

P2 denotes the phase 2 task.

P3 denotes the phase 3 task.

3. Column 5: card sequence number within subject and

task.

4. Columns 6-80: response description.

a.

L denotes left choice before reinforcement was
specified.*

R denotes right choice before reinforcement was
specified.*

1 denotes correct left response.

2 denotes incorrect right response.

3 denotes incorrect left response.

4 denotes correct right response.

 

*L and R codes indicate that the subject responded
before the experimenter could press the button to specify
the correct response.

-60-

-6l-

cc-¢-v_-c~¢-ccccooJ-c.c_n_nc¢-~uon
c-cc-comccoa-nmm-vum-ovcmn-moc-—n_<_mmnccemc¢~cc-cmmomma—cmcmmawm
mocJ-__c_ zoo
«J.J_cc-_-nc.n_.own
cemvmv..J-v-.vmecon-cocummumc—J-aaqoc-commam-vccwnamn
mammcm-vm-mnnomcccm-vm-m4onc-ncvmmmm-JJccmmocm._¢ccmcm-v:mmm-ncc-¢~NN~_J.namn
c._c_¢mmcwwummcm-maun
.cmomcmccm_-amn
mec.-_c_ 3mm
c_-¢¢._.c~_nm_omm
mocwmmmcem-c-n¢cN-occ_v-onmcvoccwc-nnaa<
«Noncmnmmmeccmmmcmmomen—meow-monoconcmmmoccmcv.n¢¢_m~.nvcccmm«nocmmmncwmn_¢~naz¢
q¢_-n¢cmm-mamm—coco-nmcocmc-m.cmmcmmmmc.mommm-nccvcococo—nonmcmcco-ncm-n.n¢_naa<
m¢_¢—_¢¢c-_~aa<
cq-mcmmccm--ncv-..c-mnvvmmvwmoccccumncmm-m-nc~_nvc~_-nz<
_¢¢_-.c~ 3&4
Amoco—.mcmm-oz<
mz-cmommnmmomomn-v_n-c-¢chcoqm¢u-nmoo:mnmcoqmvmnoc¢vmnccmmnaa<
mocmqmov-nmmmncncce-amnvmmcmamcomcvmmmvmcoom-n-ncmua-ncm-:m¢c:-c¢u-omc~¢_.mnaa<
«me.mc¢mmn¢¢v:_nmc_¢vmmccmmomaa<
came—.mo-mcwcmmmaa<
.mcq._mcmnn 3A4
m_¢¢_-.cm-_oa<
«coc-m-ccccmqocoom-moncmvc-ccmm-cmnvmvm-mmmcccmqmm-cc~¢mc-c~¢~_o-nccme_mcoonaz<
.mcqcvo-ccmmemmmocm-Na24
«me—mmcmcmcwmma24
.e¢___c. 3:4
__c¢__-¢.__oz<
.mmnao<
¢_¢c__cm-nccm-ne¢_¢mmm_cco~cmcmcu_.ncvqocmm.c-nvccmn-.mnccvocommec¢c~_n::_mnno<
c-.cmcmocuccvmmummumumwmmao<
ecc.mc:¢c~w¢..mane—cmc-~muo<
-m¢¢..mc.n¢o_.~4 3o<
mac_-¢_c-macammnccvmovmmonccmmcc~_._¢_nnm.004
cacao 44ozu>uozoz ousaouz.
.m.m>4<2< or» z. ammo mam: qpqo moor; whouooom

-62-

camocccc-mmcmmnomo-qmammmmmmmmazo

vomco-C_J_nmvmem_.-azo

.ovm-mtm 32o

Ace—ecc—.nn¢_¢¢~.vmmmmmuvmnummumummc_moo—.maso

mmcoemcoocmqo-J__¢u¢_¢.m¢u:¢a¢4mc¢J--¢cmn14o

-mc¢e4_¢o:mm_¢:o_inocqc¢~mncoqo.vauN-NQJO

:omccmc_-n-_¢__-14o

-¢¢_m.¢- 34o

.vc-Jomoomuo4o

cmo¢¢u_v_cvmomcvmm-.jn¢¢oq:oq¢c-n¢v-_nn¢_m_34o

ncocecmmmmonnccwmmocc—mono.«NM—mnmneemoonoewnmcuoonncm_ntqu-mna¢¢~N-__nvonn_«OJO

-¢¢_-¢__na<o

smcvqccmomcmv_.mcmmemmmmmmxuNm-uaqo

qov-mmccmnnnnnmnmnmnmmm-aqo

mec.__¢- 240

cm-vcm-ovmmmmmoqo

occmm¢mm.nazn

x¢-q-¢-navcmmmayn

-¢_-¢¢_mma¢n

-:c-_-¢- axn

moo-c-oqcuuz¢:qm-q-wo¢¢-c¢uu-mncumnomorn

a¢-_mcc-vm-mmcm-maz<

omv¢¢¢-_¢m_m¢-nccmmmaz<

-o_c-mcqcmmchwow¢¢_¢-.mmvmncm-mu2<

¢_¢¢__vccm-m-c- 324

mmcmecm-nvcm-co_moz<

emco-cmaacacumq-ccvm-J-n-Oz<

ace—nomvcmccmmn.nncmcccccmmomoc—-a¢am~__¢N_v-mqvonquzvomn:M-:¢_macco_mvumnmmoz<v
. aooao 44ozm>uz ouszomzoza

vcacv-cvmm-n¢cmomwazo

comm¢_¢_mmeemmvazo

ccmcmmocmnvcmmamo

.cc__.c. zoo

«cocoo¢mccumm-¢¢.~_omumo_mozo

J-vc-mnaon

mmccmnmcmnncmmnccacmnocmvoonoqonccmmcc_n_ncmmmovwvo.nvme-3mnn.mcc¢ucov-cm-nmnaom

-63-

€-~¢~¢¢-QIU
uQCm-Q~ BIU
¢¢~moan—mvmumfiflfiflfimmmmfltu
~3~¢Q-¢¢M~QNDQ><
umm«mo¢¢c_M¢—Q~A¢CNN¢N«WQM~mfiflvmmnvNNVV—mmCONN~3MN~MN~¢¢¢¢VN491NU¢-~D¢N¢numflQ><
Om"one“~303MNUNNNVNNNNNN—NQ><
C¢~¢-~¢~O¢N-Q><
€-¢¢-~¢Nn 3><
Q¢¢-¢~¢¢N~J~¢¢J~Jm«Cdumfimflfiflflfiflfl-3><
mmafluvmnflaid
~¢¢N¢N¢N¢~fl~vmv~fl~mvawmcmcmflvmmnchvmV~¢~M~D~D~D~vanvm¢V¢~0¢M~0¢M¢¢M-0¢~¢fl03<
mCCNmflunmovmnmnmvuaflCNVNCCNmHQNOVNcmomfimfimQNmCMquuocmnmV~¢~3~fl~0¢~¢fl¢~¢umflm003<
Q—"QCNQ—Q—quum"mun—QNC—vamvuvmmmcmdmunmm¢Nmmflvvwm-M¢¢m~8~O¢M-D¢Q¢N~¢U¢¢Nflfli<
VuflQuflnfluQ—Cmcccmu¢¢——M~HCN_€¢—M~manQQCmcmmmvm¢¢M-fl~¢NQ~¢~"OCCCCNCNC—nmflv¢_fla3<
v~¢~uv€~nfl¢¢¢¢-flfl¢~¢~a1¢~NQ3<
¢Q¢¢—¢m-m¢~fl¢M¢N~«QB<
~€¢-m¢~ 334
~¢¢-~€-M03<
C3-¢~¢¢mflmfl¢¢mqn¢-~03<
H-Q¢¢nmvumflflflflflfln~flafid
~¢~¢~«Q¢«NQO<
vmmmfia€cnaafid
"CC—mm?" 33<
—QQ~——Q~—~OO<
QDOHO J<mam>mQZOZ DMZQORZ—ZD
~¢m~¢€¢€~¢¢~fl~fln¢N~naflo
3am¢¢€¢-MQ~MM¢-MQQNN~NQQD
~6Q—CV—QN~—fl—Q€-~QQO
«ovmmuQ— 3&0
nCng~¢-~OQO
¢¢-~Q€~qcunfllhu
¢¢¢~n¢~¢~flnm¢N€N~NQHU
€n¢m-€€~flnmflku
~¢¢-~¢~ 3P0
m¢Q¢~q¢~¢M¢HMCQme—m—MQNNN-~OPU
¢¢¢~Qv_~—QQ~MQEU

-64-

mammmcmommuum
mcvmmmom sun
«com—mommmoum
cav_¢_mmmvvm_vmmmmam<
covvqmmcmcmmvcummmnand
m¢mmmacmmman<
mcommmom 3nd
mommdq-qoqu-mJ-cadmmtomnomnvdonmmonq
QDOQO JIth>u1 Suzaouzm
moocvmvcmnmmmcammommmmvvmcocwmocutmanmnoumoNNQso
cmovmmmvmnvaqummoccmmommmmm¢¢uocomommcmmm«vmcmmoonmvmficuoummm¢¢u~¢ummocmumcmnaso
commemommmcmmmmaao
mmoomovvmnoooowumqummmcmooummmmuso
moo—mmcm 330
«commomqvu-mm¢¢m_ovmmm030
flamnomnmqocqmwna>3
mmomsmmcmmmaJsmqmvmmmmmmmccmchmoJ-nqmanquuoua-omnm¢uuuujo¢u¢m3~fimsu¢um¢wmna>3
«Commo-oonmcoumeuu>o
mmocommvuum«Castmlovtuomtmnmmomoqumml>3
moo—mmvm 3):
¢¢mmmvmvam0m¢u¢mn¢¢¢mooumtuvmfimcuommmvmm~O>O
:—mnnsomm-Iqmocmcimmoumcmfiaka
mmvmvmmocmmwaho
«momocqmqmmqmommmmmmmaho
commcmmvmommonommcmmmmqm-chmmnmmommvmwmnmJomqmnmqqmoovmquvcuovmommmcmnvmvmmmaho
moon—mom 3pc
¢¢m¢Nm0ho
sommmaommmccvqcmme-m¢v-m~cqmmmmnncovmummmmqv-mcoucm”annousummqvmmmnmvummmmmmoho
vecomcmaonmvccmvmmmc-mqommmmmaso
woos—vomvmvmmmvaoa
mmcvmmmoommmocmHucooumocoqvmouqnmfimnomouomm«130
moon—mom 300
mvmvvmnocmmmmmmmcmmmoco
somoomvmmmuru
mnmnvwcvmcmmvmmcmmmmomovmmqmmommmommcmmmmocmmmvmmmocmvvumommcmmmcomnmvmmmnmmnaru
«commvmomn-mmvomwmmaru

...65..

V~¢N€~¢~MJ¢M¢~MJ~¢N€m0~0_fimvdvm€~¢U¢N~¢¢~fl4mmmflvN~QN<
NuQNQNCQC—M—ufivdmmmmfivmd_fl~fi—Q~Q-¢NQN~QN¢N€N~flmflmfiQNQN~n~0¢N¢M~fl~fiflm¢Nmflmfl~mQN<
DCQNN~¢¢-9~O€~¢N~¢€NQMNOO<
«QCG—anVMQuM~QNMQN—Mmfluqmwflfiuflu"JQ—CfldmkuNCVVa~¢¢N~fl~9~«¢V¢~¢—UDCN¢~¢M¢_MVM33<
~M~¢~QNQN€NQN¢fiflufl~€NQ—fluCNNQ—9—fl~¢~Q~D~OJ¢N¢~¢U¢N¢N~U~¢~TMCmO~¢UQ~fl~Dn¢U~flmUOJ<
QN—fl¢¢N~fl¢~NflvN¢N~uNCUGM—afl—fl—fl¢v¢~quvcmfl¢¢~n~H—¢—flfldvvv¢qO~M~fl¢¢U—MN_33<
nflflflCNQVun«C—CNuflflnflnflflflfivmuflflfl¢¢¢uqflufidfluflflflfi¢vfln“Dflflflfl336UU—a~099€U~OOOOS-33<
. ¢¢nu¢¢-~¢NucN-fl€¢~fl¢Q¢-Q—01QU-Q¢V~E~O~¢M—fl~«034
¢~¢N¢~¢~¢V¢u~fl¢¢Nuflfl~fl~fl~flQNUJ¢~Cu¢¢N~¢N¢mfim~9¢~¢J¢M¢N¢N€N~flqflnN~Qm<
CQuflncNVC—nfluflflQN—Gqflfl~¢Mfl¢M~mvm¢N¢~Mn6N~flfl¢M¢uflfin«MMCCNNufinfla€~9¢V¢V~flnfi~fl~«and
O¥W<h zouhdzuznflum—O NIP DDJ—(k 013 mbUNODDfl

on um<IQ no w_m>J<Z< th Zm sum: #02 maouwamth mam;
oz< n wm<ra ZO aoazm awhzmimamaxm >0 auh<Z_EIMP mans >3 324 GO mhuwfinnm *
cmcmommcwmqmvmmmnvnmnoommmnvmcmmcmmmmnamo
cmmcmmommmvccmn¢_comuommmoqvwcmmvmnommencmcmmmamo
Cmmmmvmmvmaom.Osumn:o_n¢mmn_¢¢mm4monto
"comm—cm ska
m¢mccmmccwmmm¢mmm3m3
«moccomcmmnaau
moo—cmmocmmmuau
«mommmcomnmmaau
ocmevmmcmmmmmcm 330
movaommmcmmmmcmmmoDO
vomqumcmfimmvocNomvmvmmflmflQ¥U
cmmccomau1¢¢vmmomommnmonommmaxu
mm-¢-:a_cmmnmqoommmvono¢mmomoonmfiqmstm.3:mw¢u¢-anmmn.c¢ummmmuxu
mavmmmom azu.
o:¢_.mcmmmavu
oomvomommn3¢¢¢_¢vum.movmnmvmmnaon
mmvm¢_mocmmmaon
cmoqm-vtmnmmaon
mocmmmcm 30D
m_c¢_mm¢mmmmaoo
ommmcvmmcmmmmmfiuun
mmcmommvommmuun

-66-

mom—mvvmnmmmcvmomemenccmmONNJmoqum¢¢NNNN~mJmnnmmmmacmmwmmqcavqmvmmmmmvmmNmo<o
"uncommmmmmocmNmmmmmmmvmmomenmcmommmoqoummmJ—mqflmenqavamnnovoumm_O¢NJ_mmmmo<o
vmcwomcdmcwv*1~¢Ndnflmfimflmouwowo~O¢M3~OUJ¢_flfi~nn_¢NNUQ>U
mncmvmwcucNomomma—noncomnmniuvsmnfl.nw_:m:_nmsuvmsuu¢amfimtosvd¢mnm:tmfi-nfimvm«unzu
czmmnemmvomomvucmnunnovcm«nonna—vwzmcmcmuvcmwmcmfim4Nm3>O
m0fl_c~¢omvmcmomnmnnmemvmnmncqumnmnmcmvmflmn¢u¢mflcmcsomemtocucmfim3uoummfimommma>u
IQNuvaGU—V—QNMVNMUMNNM—flmfifloomfinfi«”3152:.1ovaoflvmfimfifimfiflfimcvflumflvuuoau
NMVMOOCQN—m—nnnzaachmmnchmemnvqunoumsnowmouuuctJmfimficumnc:Mmmo:—mmcuzmnmualu
. own—910
«MGNCN—flzavvmnacvw—n. .fifichmflm ~4~flfifi¢¢flufia~fl~flfiflflfifivdduumficsumflcwivvm m mfi¢UMmflfiovumalu
coon—m«nowcwmnncuovmmfimficmmncnmmmmace—fivocmmnmfi¢mn¢uqummonocuummmnficvmmnfinnmmalu
‘ c—OQMG—MQCN— mflvcvu _ _ mflvcmcuvvmficvommn: mmIUU—Osiwcsivumlz _ m :JVNLOZU
cumcmcmmmmom—cmmmmnmmcchmmcovcmnmficwm¢N¢UJmfi_Occwdwwduxmwum mcdummmummmemmmozu
cave—mnmnomnmomamcmcnmmnmnmaqmam«memo—ONGNNOOO
Nun—n—Q—ncmcmcmcmv_mmnmcmvmmmvmvmnmm—o—canmccmoomvmmmvmcchmnmoMNaomnmnmonmmoau
coco—cmcmomnmmmoutmmnmmcmcm«moutnmcmvmnmmmamflumanu
cmemacmcmcm—n—acmcmoncwvmmnmownJvmvmouzoumowvu¢umnmomsun mnmsu:vu~m_nn¢umnfl_manu
Gouno—vmnfiodsmnm«nomnmfimwosuuvuomncmcmcwuamu
«cucu—«cmcmomzmccmnmnmvmvmnmcm¢mnmnmomsmfimcuvumt¢d¢umfi_fi¢dm::ofimtuimfimfimtwmmuamu
omcmmomu
Nome—meow"NmnnmmavcmmvnwuumunummmnmaJmflnfimoutmfioqmfin::.JTUdvmfimctm¢¢1u1¢uaoumamu
mflmmmmvmmcammmmmaocmmmmnmmmVNNmnncamm.vmcm¢¢NNZ¢mnzmv_a¢¢~anoowmvama—mmoumnmmama
cvmmcmcommmocmmmvmmmuomu
vmvcmuficNmflflomvam "fivvmuflflmfimm “UQJMV—vmnovuvmflvvcwm m ~03N4~0¢¢¢N3¢4N€N~Ouu
mm—monom—mcvmmmmmanmmvmmmmmomcmnvouwmomnnmcmnvmucumnmnnstuuuuumuummmncmnnmnmmouu
owmvncmmommmwsmv-momommmmnnooommmo>n
. ommmoxn
chJmcchNmmmaccaommJJ—vacmnmommovmmccmmmom—ooaoacowmmmmmmnvmvam~m¢¢m~_vmmmaxn
QNNN—«mmmccmmoccmmcmmmcommmmomvmmmmvmnovmmmmmnm¢umnovmummmonotmmsoummm—cumnmmoxm
cach—cmmmcmvmnmnvcmmmmmmmmmnnmmmcmvmommvvmmmoomuuwmmnmomummnOOJvamain
mflzvuomovcuuM—mmvum:uvmfimmoM3Mmmummvmmuaan
ammoouwmmucouvumflJmnmnoumncunmochmnccuvummfimo.wIMt-fimthmfiam~3mfim«330
¢u~n¢¢umflvummfiotummficcumm.mmfilsucmfiOUJmntuvvunio
mnccmmmnmchmmvcmomcqumn—"Commccmmmmmmcmmmvmmavommncuvsmmnvm¢¢mmnccmmmmcmmmvain
cm—cmflvmmcmmnmoovmunmovwmmmmmmochcmnvmchmnc¢mnmnmmommncovmmozn
ach—mmnmacmcmmm~0¢NN¢NN~MJ~M~¢N~M¢~M.mccmomcmcmmnmcmacmvmmmacmavmnmmmmow—m.~02

-67-

«JavmmflfiqmnvvmdNufivmvflcqmuflmmwmmm«ODD
O~H~Mm¢~fl¢NQN¢N¢N¢N~mfl¢mflvmuwmmwm-¢N-¢¢N~QM~Q-Q¢N~fl¢~fl¢~vummmuanNNNNQnflmmaDm
VNVuQNQ—QV—Mflflvflvmd~fl¢N-flfl¢¢un~nmflvfloflN~mAE¢U~D~DCN¢N¢~033M¢J¢M4¢M~H~¢u3N¢~mlDD

¢¢~¢¢Ma~vfluflvt€vmfiaHJ—M¢¢-O¢¢MM-~¢V~¢Vanfi¢uccuomn3~vvtm~¢V¢nma¢n §DD
"—ACC—NNDDD
{uchNaflCVVan~¢U~¢QMQ¢M¢-~¢QN-D¢UQM~D~J~¢M~¢MH~¢¢~¢U¢¢U~«Qatucum~6UQ-~¢-UDDD
¢~¢¢mvmfi¢¢afinfll¢fl¢UUU-mmMTUCAJIHCCJ~11¢U-~OC~O¢U~JDD
N¢Q¢¢N~«QQVNJa«MINCCM—flmflmflcflflm¢~0J~¢¢~N¢N¢Utumfim31U~fi¢utuqfi~1¢Uid~afimfifivmfla~330
~¢JvU—~90~H¢N—O~¢Nflmfllfl
¢¢n¢¢fla€fl¢~¢~fin~¢~¢VQV¢VJU¢¢MV~M~¢UHVafiJ—0~¢~¢U¢a¢~9¢~0~3¢U¢atx~1U~W¢~O~D-U~Qlfl
CQVQ~¢—€V¢~1Q«GQNNQNCN—flchaflq¢¢nMA«DCMflunflmfiC—OmflcvfltumC¢13~1¢M~0~fl~fi¢~¢fl¢~«QQD
«CC—«uvn 3am
Q~v¢dna¢~fi¢¢m«flQQ—mmvmumoafl

o¥m<b m MWQIQ NIH demdu 013 WhUNOflDm

CNNuMCO—Mvummvvmmqflvflvwm~n~fl¢¢~3~03¢umfi¢vflmmemwNAQ>Q
unccccqmnmmnch—M—NQCMNmu«MQN—MQNmJnmvvwmMCQMCN~D~W¢MN~$¢¢N-9¢M¢U-~3¢N~flmn~3>D
Quivmfinflvvmauflfl¢¢~m9m0n¢fl¢~¢~¢N¢U~D¢U¢VmfimJ~cmnfifllvuvMUaQXQ
#NQ—flJnmuflGNQ—m~6NN¢~D~M~Q~¢M1~MO~¢MNM¢~¢M¢NV¢~O~¢~¢¢U~I«O¢HJTMUNO~MaflDC—Oflm«QXO
Qmmflvqumflquvltfluuvmm¢¢NJmmumflvaDZQ
Qufl¢~n¢QNJmma¢u-Q~0¢¢¢¢M¢-«Evfiflmmccm—~¢~MQU-¢¢U~U¢1¢-Tm¢U¢mm¢UU¢J~O-¢m~UOZO
GunfiQufl¢~3¢-¢Nvmfl “Om~¢M¢~¢~1¢Qmfitvumme¢Umvtmi¢dltuuqHO~9¢UJ¢UU~JZO
~u€¢N-M~¢M~MQN~mfimvmcummdwchflmflmfivdmfifi¢d uuuuquum“OOUqumlflfifimmfifitflfldmfifinnozo
Qmacvamw—QJ¢~_€QNN~aHmocmflow:“OOCUUNNmnOMONNC—fi_N—OZO
MnflQN—mmvvvcca~M~MQ¢~M~MMQQ~D-0M¢M~M¢~0~fl¢¢~fl¢¢¢fl~1mfldmfioqmfi«nfivc~fl-9¢~Mfl-DED
mQQQanmnflvmwmmflmmNCQNAnflvNNCMNnJmfln«admflmflflflmmm~O¥O
QN—MQQN~M¢N~M¢QN-M¢QN—gflJ—umcvmmflvfluflvfimfivvmvm_~¢N«N~ODO
€¢13€¢¢N-M~fl¢¢flmmnmvmvmuunamVNnVM—~m¢¢m0flvvvuvufimfitdmfitdtudmfiiflcumnfimvfinmna#010
€~¢NQMQ~¢~¢M¢U~M~QUQ~Z~D~OO~3V3~¢~¢UQm¢N¢~O~O_¢U~OOD
JuflQNQNNfiGNCnO—m—¢M¢~9~GV¢MnO—Ofl~¢mflwfluflmflflvm3vV¢M~M«anCUJvJAmfl¢fl¢-O~¢J~D-300
CNN—MMMQQJ—VNuvau«TQM—-~fl¢¢~¢~fl¢flflq¢¢UJ-~¢N~9Ofl¢¢¢¢ZQQVNMNUNNN"330
NJJnunJ—CQQNN—«QQN—n«CnflcmaCCNmMO¢€¢N~J~0v~00¢M30M~J~OCCUUQVJUUU4~039~0O3¢~q~333
CVN~m_QUD
N¢NJ¢NN~nnmnnflflflflnfl¢N~fl¢¢V~annfiflflnMCUCVuua«MVC—0€M~fl¢vum€vq~94fi¢dUUJVJ~OOfiflN~OUO
MNMMVNQQQQM—nfl—NQCVq~M¢€NJ-“MC—QN~fl—HCVQ~MCCH¢U~0_Qufi¢¢m0~flfi€fl¢U~0~O~fl¢VNaa«DUO
€~M¢QNH~O<O

-68..

ommmomcmcmmmmnmmommvommmommommmmlJo
emoNNNJmmomooNNm”mommmmmqmommmmmamommmmmONN.mmmoccmmmvaomuamvmomflmmmo_¢NNNommuJo
ammochmmmvommmam 340

mqumoqm"mane—fioomvvomcmmN_OJQ
muMQQNNN—_M¢¢N—_“mommfivmwmHMQQNNJ~EQN¢¢NMN~J_flfl¢N~«Hovwmm~31UUQNM.D~O~¢NM—fl~«DJQ
_J_¢¢_€¢¢m~¢_13_mfi¢¢uuummmmvumnmmemo

«wammevmmDDIQ

cuwmnvcwusmmnmannuomncmouuum:~«mnmvflumoflmasu
€~fl~N-flflfln¢fl¢N¢MJJmMDmmMJmmammmcvqm"manwammfivuucvumDOW¢UU¢~¢M¢~¢~M~D~¢¢N—mmaiu
mvcmmmom 330

mmoommmcmmmmasu

cmccmmJ_¢Jm 3>U

chJmmmcmmmo>O

J-¢€-~€~fl RDU

mmmmvomomomvmmnn¢muvmnmmeQONMNNNmnvmenU
vcommcmammnmnvmuvmmn—cmommmnocmcummmvmvmvmouucmflfimCucucmflmikuumfimfimnnmqnmmmmsnu
mtwmvomonvmvmnmomaau
omvmmmmmomomvlmcondmnmomomomomomn-n~nm¢N¢_qmvuvm¢mmn¢M301mfimcmnm¢wuvumnmflmonmaau
NCanmflmvuflm¢N¢N¢~¢M¢UU¢mow—Omnm¢~U¢N¢U¢M¢mflmvm¢mflmfim¢um¢U¢~¢~¢~fi~3U€~¢AV~¢~NHQDU
m—mmnmmoomQN—QNomomvaONOmmmom«commnmvuvmmomJmnmtoucmW¢¢U143o3¢mfloOmnvumvmvm«Q30
movmmmom 330

qmcmccmmcnqmmanum¢V3momqu~¢NcanOU
mm—mcmcmowomomommmmommvmmmouommmmnmvmmqmmmmvmom¢¢mowmmmomqmcmfimcmqmnmmmvwmmm«GOO
mmmcomvmmmo4vumqu¢m¢mnmm«amoummm«COO

~¢¢m_m¢~ 300

maaavcmmmomnmammoau

oumflmnmcvmmnflfloqcmmovucmvnmuuu
N—QNMM"Mmommmvmvmdmmchmmmvmqmmmmmmvmvm¢~vm_fl¢~momvaN—VMummcnmmfimflmiwvmcmvwmauu
NCNunQN~¢N¢¢NN~MNQNmJaflQCNJuflmnufivmmfimfl~9¢N~fl~VVMQUHOM¢U¢M¢MM~C~Hflm1¢4~¢U¢N«“QUU
"comm—om BOO

vuacmoczovmHaemaovumnnm¢acumm«mnowmmmmouu

mama—scummmmommnvuowvnmdnu
mflmchN_¢M¢~¢mmmnmmccuvmmmcvmomoflomvmnmoumcmJomfinmcqumfimcumovamm«Jan—cmomconmanu
NammcmnvmvmoNNmMJmmovnvmmmmcwmfloem"mm«nVNm.mvocummmnqqumncmuomvummmmcmouu_nmmanu
. moo—mmom Emu
moommmvmmmonu

-69-

~33€33-~mm33m~33mmn~3~Ffi3-flfl¢3mmc33~m3mfi33-33U-Dnmmfififi3NN~3U3N_ fi~0~fl-33<

Q~Q3Q3N~m~3N~D333N3d3Nn~N~DU<

~N33N—ufl3NN3~m~3N~M¢N~AMMONNNG3N~N3¢QN-3N~flmma33Nm~M3N3N—-S3N~03N3N~Mmflmnm~0U<

33.433nm3dafl33mm33Nm_~93U~N~UO<<

3~NQ3~3~3N3~u93N3N~m~O~M3N3N~3V~O33d~00fi033UUMNN3N~"a O3N—D<<

Nafl33333N-~3N3N-M€3~m~3N~QNm~3N3N~3N3N3N~N—33~3N3N~0—~3N3N-~3Un3N3N~flnm~a«344
ohmimJ m0< IDQQD JIP Zdih ZNQJO Man? 013 WPUMODDW

voq¢om3wumuqmuumumumuuuuMn-oxo
NNN~JJJJJJ~mJJmM33—deJdmmoommmmmmon3NJJJ~JJJ—JJ—ONNJJmnaqoocmme~M33JJmmmN—Dxu
33NN~_~M3NNNNNN—mommmm3ommmmfiv3cN-33Nmn33NNmm~uunnnm¢33uuuuummflmnfi3NNN~flmnmmoxu
J3m3m33m33NmQx<
NJm—mmcacccmcommmmcomm~33N~mama—cmmmmmvmcmvmomdemcmmqommm33Nmmmmommmmmmnmmmmox<
cowawhmmzmzo< m4: xmdh hawuzou th ZTIB kzmmmd MIN; OI3 mhuwomDm
3mm3NNm3N~m33mmaDO
mme33N~ODD
~3—3NN3N~3N~M33M3Nmnmfionmmn33mmmmamcmnfiJmmo-nm33M3Jmn3u:datdm3_mumm3~3N3NJmmaDQ
m33~mmcm 330
mvvmmm34~__030
-mu-qmnoucu3M31oos-nn3umnjmawucmmua40
”cmmvvvmmcuomvu3_3mmmmonomm-QJO
m33mmm3m 34m
ammovmmmvmmmDJn
.Mmdk N mdeQ th OMdeu 013 mhumfimDm

~3Mm_3N3~MN_MMm0mO
33N-~M3N~M~333~fl3333N~3-~N3~33333~N~3N~fl3~0-3m033—fl3mfid~J~fi33J~fl3U~fl3N~NM~QflO
3N—fl—Mafln33nfl3NHQQ—3—m333Nn~33N-33N~fl-33N~mun33~3~333m33333um3-9~93~3U3N-Qflo
naivuua-vq .3an

-m3~33333NN-~33un~N~Qn0
NMMNN-~MMQNN~H3N~«M33NN~3N3U3UN~"OWMDGDOQNUUN~9333UUUUUM-893M3~O~fl0~030~fi-303
3N3N3N-flfi3J~«O33UNuflmmflfiJ3U3NNMNNm-flfl3N~fl~QIQ
NuCNQNQ~3¢N3N3NU~3~N-3~3N3N3J3N3NmJ~3NmQuflmfl~3m3N3~3~3m~3U3~3~3q3~fl~M~N~3NN~110
Qunflcn3NNN¢~3N63NN33N~N~3N-333Naflm~33U~fi~flu3unfl~fl3vu3fiun333U3Nm3m4~3N~QU3N~«Q10
—33—~_3— 3&0

~33—-3——~0&O

-70-

N.mom—acmmcmmommcmmcmnnmon—cm¢mnn-cmc_ccuu¢_nn-¢-«now-an—auuuucu-n-mnmnnoc_-ozn
emu-cmcnoonn
_c_c~¢_n_¢~¢~mnmn-nmncwcocococoncmcm-nom-coma-cmouecmcmcm-n-qocmcu-nmncmvN-N_oom
«memonn¢_Accouovmnnnmccccmmmoncmuumnnneeummo_nnnncvuwuuuuuuon-.nncumum-nnnnmoono
¢_-ncmmnvcm-oc-¢chAJ4
cmqo¢cumncvmmoceucuu-cmncu_.cuccumcu-n-cq_ccu-nq-.-n:¢¢u-an:-q-::z:vce-nccmwnAJ<
.nm_¢c~_non¢_n~mcmum_nvecmmonce—n-n-ncmn-nzcu-nv¢u-nmnc-¢u¢m-mnquc-n:c--ncuna44
coccvmnc~¢u_n_c¢¢~_nccvumn_neonvcu.-m_n¢¢_cuonon¢q¢¢uv-q¢qu-on¢cuuo-n¢uu-m-waoo
soccccsm-cu-cmm-ncvo-MAJ<

-c-_c-cqmm-ao<

.cmcc~_m 344

co_oc_c-m¢¢-mccc-oc-nccuo-nnqcm-_cm-mno-OJ<

-c_cco-qcm-o-c-mn-O¥<

q-cc--ccm-na-<

-qqcco-¢m-mJ¢-cmcm-ma-<

-vq-m-¢-nvuc¢u-m-¢c-nquc-¢w-Jo-¢ncmmm-a-<

-c-¢c-Jo_ 3-4

sac-.s-cqu-cv-_-v-mmmm-o-<

coccc-¢-nar<

-¢o:-¢¢muumumummmcmm-uar<

-¢q-¢--v~¢-m-ar<

cJ--c-qc-m-m- 2:4

c-m-Or<

~3—3333N3N— -33N~ unmmochwm—Jmn3—m3mm ~mmvd3mom3mvwm m3Nm3a3vN—n. ~MM33N~ m mnm3mm~ mar-4
-n-o-c¢-n-ccoa-m-cmcmmnau<
«coo—cmmnmcmcmc-cmm-c-mcmc_.mcmcmocmm-¢mcmoacm-q¢m-m-ccmq-q-m-c-mcvmc-mc-mau<
mc-q-cc-mau<

e--c-cc-au<

mv-co-- zoo

-c-¢c-och--cN-m-ou<

ovc-_¢- nouq

c-co-em-nccmo-cc__m-c~¢oau-¢¢J-ncqv-momco-c:mcc¢m-uooo<
omn¢~¢~..¢~¢-_.mvmncuommcmc-ocmcocmn.ncu-nvuvuoo-guano-n-n-wsn-ncucuon-n-n-oo<
oncmnccccccmcmm.mccmm-cco"mococcmcm-cmmmqmn_qccmcom-sm-om-o-o-ncmv-cm-vmuoao<
meocam¢¢omwon_cv~ocue_n.m-n¢v_m-nmc-m.neucmmvuuocucucucuuu-s-on-wnovu-n¢ucum.ao<
-_<c4__-¢J-nce-_o :94

-71-

PZDJU 0&40 3-~3~3-fl33~3fl33-H3N.N3N3NHO~O3N3U~O_O—QmDMflm~DNC
oh~E_J moq EMBOJ MIF ZdIk HMOZDO> wads OI? WPUMDNDW

m3—33mm33mmmm4~3_m_~QNU

J~3N3mmm3mmmSJU

«~33~m~fl3~MM~HDOU
~—~3—33333am~fl—13N3—m~33N~NJHM~3~3~N~N3N3m3NNN3N3~N~3~3JU3HO31~3U3~M~DO~3~mm"QED

N3~fl~m3m3aQN~3>D
3333N-~fl33NN-3N3N~a~3~3NJ~M~3N3—M33NN3—mmm3—33UNNN-3333NNN3NNUNm—Jmfi3mmmm~3>D

~33~mntummflh0
N3-flJ3~3NN3-3N3N~J-13N3~3&~3N3NNU~?3NN3~H~M~3NMN—N~Oflfl
33333.0—M~M3N3~J~3N3N—~M~M3~3Nmmm~3N3Nmfl3N3N3N~N~3N~M3N3Nm~33UNm_CD~3NN~NMM-OWT

3~33~3~3N3~“NQNN3ummmmfln3NJ3~3~3N3~D3N3~JJ1~M~3—93N33N3N3Nmozm

NIH

m m...
._ lllll.|u
n l

N. “A:
HNMMWI
vmmmmo

um

NH

1293

 

IIHIIWIHHIIIIII