Date

0-7639

2

LIBRARY
Michigan State
University

This is to certify that the
thesis entitled

AN ANALOG PRDCESS OF REPRESENPA'I‘ION
IN Ell-IE PREON

presented by

JULIE JANELLE NEIVDRIH

has been accepted towards fulfillment
of the requirements for

JLdegree in WJLJ

Major prokssor
Mark Ruling

”30,-“... ‘m'n-nn'n- . ' r ”\ -’ r....-.. ..-....

 

 

MSU

LIBRARIES
W

\—

 

 

RETURNING MATERIALS:
Place in book drop to
remove this checkout from
your record. FINES will
be charged if book is
returned after the date
stamped below.

 

 

 

 

 

 

AN ANALOG PROCESS OF REPRESENTATION
IN THE PIGEON

By

Julie Janelle Neiworth

A THESIS

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

MASTER OF ARTS

Department of Psychology

1985

ABSTRACT

AN ANALOG PROCESS OF REPRESENTATION
IN THE PIGEON

By

Julie Janelle Neiworth

The analog properties of representations in pigeons
were investigated through a series of tests of cognitive
extrapolation of movement. Rotating stimuli which briefly
dissappeared were presented on a video monitor. Pigeons
(n=5) were trained to respond "left" or "right" by pecking
one of two keys to accurate or inaccurate points of
reappearance of the stimuli.' Accurate transfer was
demonstrated when pigeons made correct choices on probe
trials in which the final location of the stimulus was
novel.

To determine if the nature of the skill was cognitive,
several analyses of the choice responses were conducted.
Results indicated that subjects shifted from a
stimulus-response (S-R) learning strategy to a more
cognitive or rule-governed strategy in the final phase of
the experiment. Results of the final phase demonstrated
stimulus control of an analog process of representations in

the pigeon.

To Mark Rosenberg, and
to Mom, Dad, Angie and
Jimmy, whose support has
been essential in the

completion of this thesis.

ii

ACKNOWLEDGEMENTS

I would like to thank Dr. Mark Rilling for his advice
and supervision of this thesis. Dr. Tom Carr, Dr. Ross
Zacks, Dr. Antonio Nunez, Dr. D.F. Kendrick, Marie Balaban,
Trish Neiworth, and Mark Rosenberg provided valuable

comments throughout this research effort.

iii

LIST OF TABLES . .
LIST OF FIGURES . .
INTRODUCTION . . .
METHOD . . . . . .
RESULTS . . . . . .
DISCUSSION . . . .

LIST OF REFERENCES

TABLE OF CONTENTS

iv

.31

. 61

LIST OF TABLES

Table Page

1.

2.

3.

Analyses of variance with repeated measures:
4:00 x 5:00 0 O O O O O O O I O I O O O 0 O 0

Analyses of variance with repeated measures:
4:00 x 5:00 x 6:00 0 I O O O O O O O O I O O O O O .34

Sign test of mean percent correct scores per
bird to first and last five probe sessions in
first probe phase . . . . . . . . . . . . . .

Sign test of mean percent correct scores per
subject from first and last five sessions Of
second probe phase . . . . . . . . . . . . . . .

LIST OF FIGURES

Figure Page

1. Schema of the transformational mappings
between distal objects, proximal stimuli, and
internal representations . . . . . . . . . . . . . . 6

2. An illustration of a perceptual, an imagery and
a violation trial type . . . . . . . . . . . . . . . 14

3. Trial types for 4:00 training and 5:00 probes . . . .17
4. All trial types for training and transfer . . . . . .19

5. Amount of transfer produced by different
learning strategies and different training . . . . . 20

6. (a-c) Predicted response accuracy after 4:00
training, and after 4:00 and 6:00 training, based
on learning strategies . . . . . . . . . . . . . . . 22

7. Stimulus parameters and trial progression . . . . . .28

8. Delay durations and response requirements
counterbalanced within conditions . . . . . . . . . .29

9. (a-b) Response accuracy to probes after 4:00,
and after 4:00 and 6:00 training . . . . . . . . . . 32

10.(a-f) Percent correct scores for each bird after
4:00 and 6:00 training . . . . O . . . . . . . . C . 4O

11.(a-f) Percent correct scores for each bird after
4:00 training . O . . C C . . O . . . . . . . . . O 42

12.(a-e) Percent correct scores for each bird for each
session throughout the experiment . . . . . . . . . .45

vi

INTRODUCTION

Until recently, many psychologists and philOSOphers
accepted DesCartes's division of organisms into two groups:
Humans, who have the capacity to think, and animals, who do
not. According to this point of view, animal behavior, no
matter how elaborate and complex, could always be reduced
to some configuration Of reflexes in which thought plays no
role. The success of stimulus-response (S-R) models of
conditioned behavior had "reinforced" this way of thinking,
for nothing seemed to be gained by trying to explain an
animal's behavior by appeal to its mental life.

This state of affairs has changed markedly during the
past ten years. A variety of studies of spatial memory
(Olton, 1978; Oltcn & Collison, 1979), delayed
matching-to-sample (Roitblat, 1980), recall of photographs
(Wright, Santiago, Urcuioli & Sands, 1984), serial learning
(Straub & Terrace, 1981), natural concept formation
(Herrnstein, Loveland, and Cable, 1976), and abstract
concept formation (Zentall, Edwards, Moore, a Hogan, 1981)
have created unprecedented difficulties for S-R models of
behavior. In each of these studies, the behavior in
question could not be explained by reference to an
immediately available stimulus, or to some covert S-R
mediator. Each of these studies demonstrated that animals
can encode stimuli and relationships between stimuli that

1

are not immediately present. In each case, the animal's
behavior was controlled by representations of stimuli.

Our current understanding of the use and form of
animal representations is based on speculation, with
causality inferred between theoretical descriptions of
memory and observed behavioral phenomena. (See Roitblat,
1982, or a review.) However, a promising technique has been
developed by cognitive scientists to study the
characteristics of humans' representational system. The
technique induces changes of mental representations: the
solution of a problem requires subjects to manipulate their
representations (Shepard & COOper, 1982; Kosslyn, 1980:
Jagacinski, Miller and Johnson, 1983). In this thesis,
behaviors are studied which reflect analog prOperties of
representations of stimuli in pigeons. The aim is to
uncover specific characteristics of pigeons'

representational system.

Accepted Notions about Animal Representations

Any organism that learns something, i.e., that
responds a specific way in the presence of specific
objects, must retain a memory of the object and of the
consequences of responding in a specific way. The memory of
the response-consequent was coined by behaviorists as an

association, a more general term implying that some drive,

3
need or emotion was associated with responding in the
presence of a stimulus. The behaviorists focused on the
response end of the internal process of perceiving and
responding, and in doing so, ignored the stimulus and and
the power of control in its organizational prOperties which
determine which aspects of a stimulus will be perceptually
registered by an organism, and which aspects become the
condition for a particular reinforced response. Since
stimulus properties were neglected, one of the central
problems of learning theory was to explain how organisms
behave apprOpriately to novel stimuli. Stimulus
generalization was one well-established mechanism (Rilling,
1977). However, most behavioristic theories of stimulus
generalization assumed an invariant internal repreSentation
locked to the stimulus in the environment (Schwartz, 1984).
Behavioral plasticity, behaving apprOpriately in the
presence of new stimuli, is a criteria for cognitive
control of behavior (Pylyshyn, 1984) and necessitates a
'plastic' or changeable form of memory.

To solve the problem, researchers separated out the
study of the memory of the object itself. They typically
referred to it as a 'representation' (Roitblat, 1982). A
representation in this sense was a memory of perceptual
information (including size, color, rigidity, movement,
etc.) that describes the object. Still the concept of
representation was invariant because it was inextricably

tied to the percept of the object. So, researchers asserted

4
that plasticity must come from its use (Kosslyn, 1980;
Shepard & COOper, 1982).

With few empirical studies on animal representations,
speculation ensued concerning the possible use of
representations in animals. The most pOpular speculation
was that animals use some imagery process to represent
objects, probably because images seem to be the lowest form
of memory, i.e., they are solely based on prOperties of
percepts. Psychologists such as Premack (1983) and Jerison
(1973) have used evolutionary arguments to imply that lower
animals use images to represent the world. A contemporary
ethologist, Griffin (1981) states:

Mental images obviously vary widely in the

fidelity with which they represent the

surrounding universe, but they must exist in some

form in any conscious organism. (p 6)

These speculations were not met with arguments. Curiously,
imagery in animals has been assumed to be the way in which

animals represent the world, but it is assumed by default

without empirical justification.
Animal Representation as an Analog Process

Within the framework of cognitive psychology and
evolutionary theory, Shepard and COOper (1982) present the
strongest logical case describing a representational system
in animals. They begin by assuming that animals probably
represent perceptual information which describes objects.

Next, they assume that organisms that are able to

5

manipulate those representations in order to "anticipate
the consequences of structure-preserving transformations"
have an evolutionary advantage. Shepard and Cooper do not
say that animals retain 'pictures' in their heads which
they can move around or transform in order to anticipate
change. Shepard (1984) states:

What is internalized at the deepest and most

abstract level is not any particular object or

transformation (which are arbitrary with respect

to orientation and path) but the set of

constraints that in three-dimensional Euclidean

space govern the possible projections and

transformations of an object. (p 442)
The process Shepard and COOper describe is an analog
process (not necessarily an imagerial one). In an analog
process, the relational structure of external events is
essentially preserved in the corresponding relational
structure of the representation. This relational structure
could be preserved in digital form, in picture form, or in
some logical form. In any case, there must be a one-to-one
relation (or isomorphism, to use Shepard's term) between
stages of the internal process and stages of the
corresponding external process. Moreover, intermediate
stages of the external process are represented even if
those external stages are not physically presented. As in
the diagram presented in Figure 1, organisms can observe
stage A', and then stage C' and retain an internal
representation which includes the transformations which

occurred between A' and C' (i.e., state 3'). Thus an analog

process preserves stimulus changes and allows the organism

6
DISTAL PROXIMAL

INTERNAL
STIMULUS STIMULUS REPRESENTATION
External Surface Deep
Object

Structure Structure

o e4 e
e

i Kc;

Figure 1. Scheme of transformational mappings between
distal Objects, proximal stimuli, and internal

representations. Adapted from Kubovg 8. Pomerantz,
MMOW. LEA. 1981. p. 295.

 

 

 

 

 

7
to respond apprOpriately to new stimuli which obey the same

rules of change.

Evidence of Analog Processes

Not only is the analog process more logical than were
the imagery speculations, but it is also testable. Shepard
and COOper, and their colleagues have develOped a
quantitative methodology for the study of analog
transformations of mental representations in humans. The
most well-known are studies of 'mental rotation'. After
discrimination training between stimuli and their mirror
images, a test stimulus which has physically changed in
orientation from its trained presentation is presented to a
subject. The dependent variable is reaction time, or time
taken to respond if the stimulus matched the standard or
mirror image form of the trained stimulus. The
well-established finding is that mean reaction times
increased linearly as a function of the angular departure
from the expected or trained orientation. An analog process
is obviously in use, since the subjects' reaction times
revealed a one-to-one correspondence between internal
process and external presentation of orientation.

From this and other studies on humans, it has been
established that humans can represent transformations of
objects in analog form. The question which follows is

whether other species of animals have a similar cognitive

ability.

Pigeons have a very well-develOped visual system
(Grands & Maxwell, 1979) and have demonstrated the ability
to remember features of complex naturalistic photographs
(Herrnstein & De Villiers, 1980) as well as features of
complex computer-generated stimuli (Cerella, 1980). These
demonstrated abilities indicate that pigeons have an
accurate memory system for visual stimuli. The possibility
that they could also represent visual stimulus prOperties
in some analog form seems great. But, there has only been
one study to test the analog process in pigeons, and it
produced negative results. Hollard and Delius (1982)
borrowed a mental rotation methodology (Kubovy & Podgarny,
1981) to test pigeons. They employed a matching-to-sample
paradigm. First, a single sample stimulus was presented to
pigeons. Next, two stimuli were presented simultaneously.
Pigeons were reinforced for pecking the comparison stimulus
which matched the sample stimulus. One test stimulus
matched the sample, and the other was a mirror image of the
sample. On 80 percent of the trials, both test stimuli were
presented at angular orientations which differed from the
sample stimulus. The task required pigeons to discriminate
between two simultaneously presented stimuli which differed
in orientation from the sample, and then to peck the
comparison stimulus which matched.

Needless to say, 1000's of sessions were required to

train the pigeons how to respond accurately. Many of the

9

orientations were trained as well (0, 45 and 90 degrees).
Results of testing indicated that pigeons performed
accurately to stimuli oriented to novel 135 degrees and 180
degrees, and they performed accurately to novel stimuli
presented in trained (0, 45, or 90 degrees) and novel (135
or 180 degrees) orientations. Pigeons' reaction times to
respond, however, were consistently the same across trials.

Because the reaction times function did not increase
linearly with angular disparity, Hollard and Delius
concluded that pigeons do not 'mentally rotate' stimuli in
the continuous fashion that humans do. Upon reading this,
most assume that the study provides evidence that pigeons
do not have the type of representational system that
Shepard and COOper discuss. The study demonstrates that
pigeons CAN represent stimulus prOperties in an analog
fashion, and they can use their analog representational
system to solve the task when novel stimuli and novel
orientations are employed. How else could they perform so
accurately, unless they coded a one-to-one correspondence
between represented stimulus orientations at standard or
mirror image form? Pigeons did not produce the reaction
time function that reveals a 'mental rotation' process.
However, pigeons had to solve the problem through the use
of some analog process, as Hollard and Delius admit in
their discussion:

Pigeons solved the problem differently (from

humans) and more efficiently, presumably through

a parallel mode of information processing. (p
806)

10
By 'parallel mode', they mean something which produces the
same response accuracy as mental rotation does, but not the
same timing considerations; i.e., some other type of analog
process.

The reaction time function was the only negative
finding of this study. However, this is due, in part, to
the fact that pigeons are easily distracted from a task by
Outside noises, temporary changes in the experimental
environment, and so forth. The primary reason that reaction
time is never a valuable variable in Operant study is that
it is not measured appropriately. In cOgnitive tasks,
reaction times are meaningful in reflecting time for
internal processing only if subjects are asked to respond
as quickly as possible. A time constraint could easily be
placed on responding such that pigeons are forced to
respond as quickly as possible. But in most pigeon
experiments, including Hollard and Delius's, reaction times
are simply measured with no time constraints. Because of
the methodological and measurement flaws, the question of
analog processes of representation in the pigeon remains

Open.
New Methodology to Study Analog Processes
Since Hollard and Delius failed to obtain_clear

evidence for an analog process (i.e., mental rotation) in

pigeons using Shepard and COOper's procedure, I decided to

11
employ a different methodology. There is an area of human
cognitive psychology in which the experimenter presents
moving stimuli to the subject for the purpose of drawing
inferences about humans' representation of movement. An
experiment by Jagacinski, Miller and Johnson (1983) is a
prototype for testing this ability. The basic task was to
estimate stimulus location at various times after a moving
stimulus dissappeared. This ability required "cognitive
extrapolation” based on information gained about the
stimulus while it was present. The stimulus moved from left
to right and dissappeared. In its absence, subjects had to
make some mental manipulation of the remembered information
about location, direction, and speed of motion in order to
accurately predict location. Subjects pressed a button when
they thought a dissappeared moving stimulus would have
reached a particular spatial location. Their response was a
yes/no judgment. With feedback, humans made accurate mental
transformations of stimuli, and this cognitive skill was
accurately applied to novel situations. The task involved
an analog process through which subjects represented
stimulus movement.

This procedure might be reasonable for testing
pigeons. Movement is not implied by static displays of
changed stimuli, as in the mental rotation task, so
excessive training is not necessary. Examples of movement
are presented so that any 'mental extrapolation' is based

on representations of perceived physcial transformations.

12
If these representations are retained in some analog
fashion, pigeons should be able to accurately extrapolate

stimulus movement to novel locations.
Specific Plan

The present experiment investigated pigeons' ability
to use analog transformations of representations Of
stimuli. The methodology was a synthesis of the
Jagacinski et a1. (1983) study of movement extrapolation
and of Shepard and COOper's (1982) research on mental
rotation. The basic idea was to determine if after
discrimination training of movement to two locations,
transfer is Obtained at a new intermediate location. The
aim was also to discover if two boundary location
conditions must be trained before an analog process is
applied to a novel location.

The task was a conditional discrimination with
movement and location used as discriminative stimuli. The
stimulus was a lighted rectangle which rotated in a
clockwise fashion on the screen of a video monitor. Three
different types of trials were trained: 1) a perceptual
type in which the rotating stimulus moved continuously to a
certain point and was always physically present, 2) an
imagery type in which the rotating stimulus disappeared at
a specific point and then reappeared at the point it would

have reached if it continued moving while it was absent,

13

and 2) violation types in which the rotating stimulus
dissappeared and reappeared at a location which was
incompatible with the time it was absent and its speed of
motion. According to Shepard and COOper (1982) mental
transformations are performed in the absence of external
counterparts, i.e. during the delay. An example of each of
the trial types is presented in Figure 2.

A trial began with the stimulus presented at 12:00.
After certain contingencies were met (an FR-12 on the
middle response key of a three-key display in front of the
monitor), the stimulus started its rotation. For all
perceptual trials the stimulus moved continuously and then
stopped at a specific location. For all imagery and
violation trials, the stimulus moved to 3:00, dissappeared,
and then reappeared after some time. For imagery trials,
the stimulus reappeared at the location it would have
rotated to on a perceptual trial. For violation trials, the
point of reappearance was incorrect assuming that the
stimulus continuously moved during its absence. Left key
responses were reinforced for perceptual and imagery types.
In both cases, the stimulus moved at some constant rate
even though it may have dissappeared for some time. Right
key responses were reinforced for violation type trials,
indicating that the point of stimulus reappearance was

incompatible with constant stimulus motion.

14

400 Training
perceptual

Tne‘stimuius rotates from 12:00
to 4:00. It is alwags present.
L8“. keg responses are reinforced

 

imagery

The stimulus rotates from 12:00 to
3:00. It then dissappears, and is absent
during the time it would take to rotate
from 3:00 to 4:00. After this delay, it

reappears at 4:00. Left keg responses
are reinforced.

 

violation

The stimulus rotates from 12:00 to _
3:00. It then dissappears, and is absent
during the time it would take to

rotate from 3:00 to 6:00. It then
reappears at 4:00. Right keg responses
are reinforced.

 

Figure 2. An illustration of a perceptual, an imagery,
and a violation trial type.

15

Birds were first given discrimination training with
all trials ending at 4:00. After the 4:00 task was
mastered they were next given training at 6:00. In the 4:00
condition, there was a perceptual trial type in which the
stimulus moved continuously from 12:00 to 4:00, an imagery
type in which the stimulus dissappeared at 3:00 and
reappeared after a short delay at 4:00, and a violation
type in which the stimulus dissappeared at 3:00 and
reappeared after a long delay at 4:00.(F0r an illustration,
refer to Figure 2.) Contingencies of reinforcement were
based on the response requirements previously described.

After acquisition of 80% accuracy at 4:00 with the
different trial types (perceptual-imagery-vielation), a
transfer test was given. The question of interest was
whether the birds accurately responded to novel stimulus
presentations ending at 5:00. Only a few examples of these
probe trials were presented within typical training
sessions of the 4:00 condition. This strategy was used in
an effort to prevent the birds from learning 5:00
contingencies, since these were re-tested later after both
boundary conditions (4:00 and 6:00) were trained. In the
5:00 condition, four different trial types were tested (see
Figure 3). In the perceptual type, the stimulus moved from
12:00 to 5:00 continuously. In the imagery type the
stimulus moved from 12:00 to 3:00, dissappeared for a
"medium" delay and then reappeared at 5:00. Also, there

were two violation types: one in which the stimulus

16
dissapeared for a short time and reappeared at 5:00, and
one in which the stimulus dissappeared for a long time and
reappeared at 5:00. Reinforcement was given for correct
responding to the 5:00 condition trials (i.e., left =
perceptual and imagery, right = violations).

After training of a single condition (4:00), response
accuracy to probes is predicted to be lower than response
accuracy to trained examples. Subjects can learn to respOnd
to the 4:00 condition accurately by simply learning a
temporal discrimination which does not apply to all of the
5:00 probes (see Figure 5). During training, subjects can
respond "left" if the stimulus never dissappears, respond
”left" if the stimulus dissappears for a brief time, and
respond "right" if the stimulus dissappears for a long
time. By applying this temporal discrimination, birds can
receive reinforcement on 100% of the trials in the 4:00
condition. Given that this type of learning occurred to the
4:00 condition, the only transfer (high accuracy of
response) predicted to the novel 5:00 condition is for
those trials types which are compatible to the 8-H
associations learned. For the (5:00) perceptual type and
for the (5:00) violation type with the long delay, the
temporal discrimination learning yields accurate responses
(see Figure 5). For the (5:00) violation type with a short
delay, the learned response is inaccurate (i.e., short
delay -/ left in that condition). For the imagery type, 50%

response accuracy is predicted since the birds have no

17

4:00 Training

perceptual imagery violation

 

 

5:00 Transfer

perceptual imagery violationI violation 2

   

D = 4:00 delaU D - 5:00 delay D a 6:00 delay

Figure 3. Trial types for 4:00 training and for 5:00 probes.

18
previous association to the location or to the new "medium"
delay. A graph of predicted results to 5:00 probes given
S-R learning to the 4:00 condition is presented in Figure
6a.

Following training at 4:00, a condition for which all
stimulus presentations and at 6:00 was introduced. A 6:00
condtion was trained to solve two problems: 1) to
counterbalance the temporal discrimination so it no longer
applied, and 2) to identify the analog process by training
boundary conditions (4:00 and 6:00) and testing an
intermediate location (5:00). Again a perceptual type
(continuous movement from 12:00 to 6:00), an imagery type
(movement from 12:00 to 3:00, dissappearance for a long
delay, and reappearance at 6:00), and a violation type
(movement from 12:00 to 3:00, dissappearance for a short
delay, and reappearance at 6:00) were trained (see Figure
4). Contingencies of reinforcement were the same as in the
4:00 condition: left responses were reinforced in
perceptual and imagery trials, and right responses were
reinforced in violation trials. The 6:00 condition was
trained until response accuracy reached 80%. Then birds
were exposed to alternate days of 4:00 and 6:00 training.

Figure 4 illustrates both 4:00 and 6:00 condition
trial types. Responses to temporal discriminations were now
counterbalanced by the 6:00 condition. Simple S-R

associations of a temporal discrimination only produced

19
Training

perceptual imagery violation

0 ”

 

 

 

 

     

 

Transfer

perceptual imagery violation ' violation 2

 

 

 

 

 

     

Figure 4, All trial types for training and transfer conditions

20
Training Learning Strategy Transfer

 

4:00 S-R associations: Very little:
no delay - go left perceptual: yes
short delay - go left imagery no
long delay - go right 10018000,: no

V' ' :
lolauOl')2 yes

 

 

4:00-6:00 Conditional rules: Very little:
If 4:00 then
no delay - go left perceptual: yes
short delay - go left imagery: no
long delay - go right violation 1: no
violation 2: no
It 6:00 then
no delay - go left
long delay - go left
short delay - go right
4:00-6:00 Cognitive Extrapolation Immediate
accurate stimulus - go left. perceptual. yes
movement
accurate stimulus - go left imagery: yes
movement
inaccurate stimulus - go right ViolationI: yes
movement violationzz yes

Figure 5. Amount of transfer produced by different
learning strategies and different training.

21
accurate responding in the perceptual trials. For all other
types, birds were forced to either use an analog process of
representation, or alternatively, to formulate
bi-conditional S-R associations contingent upon location.
(For an illustration, see Figure 5, 4:00-6:00 training.) If
they used a bi-conditional learning strategy, the following
would be learned: in the cases where stimuli ended at 4:00,
short delays require left responses and long delays require
right responses; but, in the cases where stimuli ended at
6:00, short delays require right responses and long delays
require left responses. Bi-conditional learning produces
discrete S-R associations which have little applicabilitiy
to novel probes. If an analog process is used, subjects
represent 4:00 and 6:00 in some analog fashion. If pigeons
use an analog process to preserve transformations of the
stimulus from 4:00 to 6:00, they also retain an analog form
for an intermediate location, i.e., 5:00. By this method,
they respond to novel probes accurately.

The 5:00 condition was tested after 80% proficiency
was observed to alternate days of 4:00 and 6:00 training.
This time 5:00 probes were introduced alternately within
4:00 training sessions and within 6:00 training sessions.
Immediate transfer to the perceptual type is predicted
because the S-R association still holds (e.g., no delay -
go left). However, no immediate transfer is predicted for

the imagery or the violation types unless an analog process

sees

60
50

«36883

a

.aasasaasas

«32588388388

22

Predicted response accuracy after 4:00 training,
W

A
T

L
V

L

L
V

L
T

L
r

A
fi

perceptual imagery violation perceptual imagery violation' violation2

Figure 6a.

A
fi

perceptual imagery violation perceptual imagery violation

4:00

f

Figure 60 . 4:00-6:00

L
V

Figure 60.

H——O

A
f

A
V

A
r

5:00

5:00

Predicted response accuracy after 4:00 and
6:00 training, based on conditional rule strategy.

A
f

1

violation

D—D—O O——O—-O——O

Predicted response accuracy after 4:00 and

6:00 training. based on cognitive extrapolation

A
T

4:00-6:00

L
T

A
Y

%
#—

5:00

A
V

perceptual imagery violation perceptual imagery violation 1 violation

2

23
of representation is used. The possible conditional rules
learned from 4:00 and 6:00 training are based on the
location of the stimulus (i.e., if at 4:00, then one set of
S-R rules; if at 6:00, then an alternate set). Since the
new probes are located at 5:00, neither of the sets of
conditional rules are applicable. Since 5:00 is between
4:00 and 6:00, simple generalization of the location rules
yields 50% response accuracy to all the imagery and
violation types. But, if an analog process is being used,
5:00 is accurately represented in the preserved
transformations of the stimulus from 4:00 to 6:00. Response
accuracy to 5:00 probes is high and immediate, if an analog
process is applied. Figure 6b illustrates response accuracy
predictions to 5:00 probes after both 4:00 and 6:00
training resulting in the learning of conditional rules.
If response accuracy to 5:00 probes is higher than in this
prediction, then predictions of an analog model are
supported. (See Figure 6c for predicted results, and see

Figure 5 for predicted learning strategies.)

METHOD
Subjects
Five White Carneaux pigeons served as subjects. They
had previously participated in some preliminary
investigations which involved pecking response keys to
stimuli presented at different locations on the screen of

the video monitor. There was no specific response bias

24
trained by the previous research to the locations used in
the present study. Additionally, this type of movement and
time delays were never used before as discriminative
stimuli.

Subjects were maintained at 80 percent +/- 20 g. of
their free-feeding weights throughout this experiment.
Subjects were individually housed in a
temperature-controlled, constantly illuminated colony room
wherein water and grit were always available. Subjects were
transported daily to an experimental room in which they
participated in the experiment. Each subject was fed in the

colony room immediately following its daily session.

Apparatus

A Life Sciences Associates secondary video monitor
(CAT #509-MD1000-190) with a 12 cm. x 9.5 cm screen was
fixed to the stimulus panel of a Lehigh Valley Electronics
pigeon chamber. It was positioned on the stimulus panel 6
cm from the t0p of the chamber and centered from the sides.
The secondary monitor was interfaced to a TRS-80 MIII
microcomputer through which computer graphics were used to
present stimuli. The microcomputer was also interfaced to
electromechanical equipment which controlled the chamber
Operands.

A three-key pigeon chamber, measuring 35 X 35 X 30
cm., was used. The 2.54 cm.-diameter clear plastic keys

were positioned in front of the video monitor: 13 cm. from

25
the tOp of the chamber, 2 cm. apart from each other, 12 cm.
from the sides of the chamber, and 22 cm. from the bottom
of the chamber. The keys were mounted in a row across the
middle of a sheet of plexiglass (dimensions: 20.5 cm. X
16.5 Cm.). The plexiglass was mounted 3 cm. from the screen
of the video monitor, and holes were cut out for the
response keys' activation areas. The keys required a force
of 15 g. (.15 N) for activation. The houselight, a 28 V.
dc GE 757 light, was positioned at the top center of the
back wall adjacent to the stimulus panel. The grain
magazine was 5 X 5 cm., and was located 9 cm. from the
bottom of the chamber and centered under the middle key.

The chamber and monitor were housed in an insulated
chamber equipped with a fan to provide ventilation and mask
noise. The microcomputer was housed in an adjacent room and
connecting wires between the computer and the chamber ran
through a conduit in the wall.

During each session, the microcomputer recorded the
type of trial presented, the type of response emitted, the
time taken to respond, and whether or not reinforcement was
provided. This trial-by-trial data was dumped to a printer
after each session. At the end of each session, the
microcomputer also calculated percent correct scores for
each trial type. Tests were given based on the accuracy of

these percent correct scores.

26
Procedure

From preliminary investigations, subjects had already
been trained to eat from the magazine hOpper and to peck
the response keys. Next, the five subjects were exposed to
both 4:00 and 6:00 conditions within single sessions of
training. The original plan of this research was to train
the birds on both conditions within the same session, and
then to probe-test at 5:00. However, months of simultaneous
training produced no better than chance performance for
both conditions in all subjects (See Appendix A for data.).
The birds could not acquire the task with all 4:00 and 6:00
trial types presented within the same session. The task was
simplified by breaking training down into two conditions. A
test after each was conducted to determine the learning
strategy used by the subjects at different points in the
experiment.

Before participating in this present study, the birds
were placed on a continuous reinforcment (CRF) schedule
contingent upon a peck to any of the three keys in the
chamber. Each bird was run on the CRF schedule until it was
pecking approximately 33% of the time to each of the three
keys. This was accomplished by decreasing the availability
of reinforcement on favored keys until all keys were
equally pecked. All five subjects reached this proficiency
within five sessions.

Training on the 4:00 condition discrimination was

outlined in Figure 2. The stimulus used throughout this

27
experiment was a bar that rotated clockwise. The stimulus
parameters followed the description in Figure 7. Typically,
the stimulus was always presented in a 12:00 position at
the beginning of each trial. After 5 seconds of this
presentation, a fixed ratio (FR) 12 on the middle key
initiated the movement of the stimulus. This response
requirement prior to movement insured that each subject was
looking at the screen when movement occurred. The bar
stimulus moved in a clockwise fashion at a rate of 90
degrees/second. Seven presentations occurred within a 90
degree change, so that the bar stimulus actually flashed on
and off, while moving at approximately 13 degrees per
change and at a rate of 1 flash every 150 milliseconds.

The stimulus either continued its movement and
stOpped, or dissappeared for some amount of time and then
reappeared at some location. In either case, the trial
ended with a static presentation of the bar stimulus at
some location. The ending locations were always either
4:00, 5:00, or 6:00. The times of dissappearance were 0.5
seconds, 0.75 seconds, or 1 second. The ending locations
and times of dissappearance were counterbalanced to provide
perceptual, imagery and violation trials for each of the
three location conditions. (For an illustration, see Figure
8.)

The possibility of obtaining reinforcement was present
in all trials. Right responses produced reinforcement in

the violation trials and left responses produced

28
Video Monitor Display

 

7

000
v

 

 

 

 

 

 

 

 

 

 

 

 

r l
O O O

L J
O e O

 

 

 

Procedural Description

5-second presentation
FR-l 2 on the middle key
to start the trial

Movement for 1 second

Delay condition
(i.e., 0.5 seconds)

Test Stimulus

Correct response is
left key. Stimulus
remains on until

a response occurs on
the left or right key.

Figure 7. Stimulus parameters and trial progression

29

End Delay Correct

 

 

 

Location Duration Response
0 seconds —— left
4:00 <05 seconds left
1.0 seconds right

 

 

0 seconds—left

5:00 0.5 seconds right
075 seconds ————left

1.0 seconds right

0 seconds left
6:00 <05 seconds right
1.0 seconds left

 

 

 

 

 

Figure 8. Delay durations and response requirements
counterbalanced within conditions.

30
reinforcement in the perceptual and imagery trials.
Reinforcement consisted of three seconds of access to mixed
grain in the magazine hOpper. Probe trials were also
reinforced according to the same contingencies.

The number of exposures to trial types was controlled
so that equal numbers of left and right responses were
required per session. During 4:00 and 6:00 training, 16
perceptual trials, 16 imagery trials and 32 violation
trials were presented each day for a total of 64 trials in
which 32 were reinforced for left responses and 32, for
right. The types were pseudo-randomly presented with the
constraint that no more than 3 of a single type were
consecutively repeated.

For sessions with 5:00 probes, only 2 of each of the
four 5:00 trial types (perceptual, imagery, violation and
violation) were presented within typical training sessions.
The frequency of exposures to probes was kept extremely low
(less than 5% of the trials per session) so that learning
of them would probably not occur. The probes were
pseudo-randomly presented within training sessions.
Approximately 24 exposures to each 5:00 probe type were
presented within 12 sessions of probe testing. Probes were
tested in this way twice: once after 4:00 training, and
again after 4:00 and 6:00 training.

Correct and incorrect responses for each trial were
recorded by the microcomputer, and percent correct scores

were calculated for each trial type. Criterion for testing

31

was 2 consecutive training sessions with averages of 80% or
better response accuracy across trials.

RESULTS
Transfer

The most basic test of these data is to determine if
response accuracy achieved in training transferred to a set
of novel probes. Positive transfer is evidenced by accurate
responding to ALL 5:00 probe types. To demonstrate maximum
or complete transfer, response accuracy to 5:00 probes must
be nonsignificantly different from response accuracy to
trained 4:00 and/or 6:00 trial types. A further test of
transfer is that choice performance is significantly
different fom 50% or chance level.

Figures 9a and 9b show average percent correct scores
to 5:00 probes after the two training phases (4:00, and
4:00-6:00). Figure 9a shows response accuracy to probes
after only 4:00 training. In Figure 9a, response accuracy
to two 5:00 probe types is obviously different from
response accuracy to the 4:00 trained examples. This
accounts for the significant condition effect found by an
analysis of variance with repeated measures (F(1,150) =
10.9074, p < .005, see Table 1). Because of this
difference, there was also a significant type effect found
by an analysis of variance (F(2,145) = 22.4792, p < .001,
see Table 1). (The subject effects and its interaction

effects were nonsignificant in both analyses).

32

 

 

 

100-
90 — H
so - 7-
t; 70 ..
8
3 60 -
u so A
3;; 4° - O perceptual
8 3° '4 B imagery
g 20 -i A violation
10 -
4:00 5:00
Figure 9a. Response accuracy to probes after 4:00
training.
100-
90 —I
a 80 —1
Q
8 7O —1
s
(.1 6° '1
5 40 q
E O perceptual
g 30 m B imagery
2° -‘ A violation
10 -

 

 

4:00 5:00 6:00

Figure 90. Response accuracy to probes after 4:00 and
6:00 training.

Table 1.
Analyses of Variance with Repeated Measures.

Two-Way ANOVA: Location Condition X Subject

SS df MS F Sig.
4-5 40592 1 40592 1009074 (.005
Subjects 0.452 4 0.113 0.2664 n.s.
Interact 0.924 4 0.231 0.5487 n.s.
Within 63.172 150 0.421
TOtal 690140 159

Two-Way ANOVA: X Subject

SS df MS F Sig.
P-I-V 15.151 2 7.5755 22.4792 (.001
Subject 0.452 4 0.113 0.3353 n.s.
Interact 4.672 8 0.584 1.7329 n.s.
Within 48.865 145 0.337
Total 69.140 159

Response accuracies to two 5:00 types,

the perceptual type

and the violation type with a 6:00 delay, did show evidence

of transfer because they were not significantly different

from response accuracy to trained counterparts when tested

by Tukey analysis (perceptual, q 8 .01501, p > .05

violation q - .05678, p > .05). But response accuracy to

the 5:00 imagery type and the 5:00 violation type with the

4:00 delay showed no evidence of transfer:

they were

significantly different from their trained counterparts

(imagery, q - .3649, p < .01; violation, q - .5136, p <

.01). Since response accuracy to two of the four 5:00 probe

types was significantly different from accuracy to trained

examples,

complete positive transfer did not occur here.

These data are discussed later in this section to determine

34

why some probes did transfer while others did not.

Figure 9b shows response accuracy to 5:00 probes after
4:00 and 6:00 training. Average response accuracies to
4:00, 5:00 and 6:00 conditions were 73.5%, 82.2%, and 73.7%
respectively. By analyses of variance with repeated
measures, no significant difference was found between
conditions (F(1,165) = 0.1229, p > .05) or between trial
type (F(2,165) = 2.43668, p > .05). (Additionally, there
were nonsignificant subject and nonsignificant interaction
effects.) These data meet the criterion of complete
transfer, i.e. there was no difference between response
accuracy to probes and response accuracy to trained

examples (see Table 2).

Table 2.
Analyses of variance with repeated measures.

Two-Way Anova: Location Condition X Subject

SS df MS F Sig.
4-5-6 0.174 2 00087 001229 nos.
Subjects 0.161 4 0.041 0.0594 n.s.
Interact 0.504 8 0.063 0.0890 n.s.
Within 116.765 165 0.707

Total 117.604 179

Two-Way ANOVA: Type X Subject

SS df MS F Sig.
P-I-V 30349 2 1.674 2.4367 nos.
Subject 0.161 4 0.041 0.0597 n.s.
Interact 0.739 8 0.093 0.1354 n.s.
Within 113.355 165 0.687

Total 117.604 179

35

Evidence of transfer is typically defined as response
accuracy to probes at better than chance level ( > 50%). To
further verify the transfer result, a sign test was
conducted to test whether the probe data for each bird was
significantly better than chance level. Besides supporting
the definition of transfer, this test was necessary because
the analyses of variance results were not sufficiently
meaningful. The variances were quite heterogeneous between
probe data and training data because probe data was based
on very few exposures and training data was based on a
large number of exposures. In the analyses of variance,
these large variance differences were virtually ignored.
The resulting nonsignificant results which support the
hypothesis might be due to large variances which masked
differences. The sign test tests each bird's mean percent
correct scores across 12 sessions for perceptual, imagery,
and violation types against an hypothesized 50% score. The
variance problem does not exist because the test does not
take into account variance. In order for there to be clear
evidence of positive transfer, subjects' scores must be
significantly greater than chance level ( > 50%). In every
case, the birds' percent correct scores were significantly
greater than 50% (p-.031). The results confirm the
interpretation that there was positive transfer to 5:00

probes after 4:00 and 6:00 training.

36
Strategies

Transfer has been demonstrated and verified in the
probe data produced after 4:00 and 6:00 training. The next
question is what type of strategy had the birds used which
produced accurate responding after 4:00 and 6:00 training?
And what strategy were the birds using after 4:00 training
which produced the inconsistent results in Figure 9a? A
determination of the Strategies used requires consideration
of specific theories. A detailed review of the response
predictions and hypothesized strategies mentioned in the
introduction is necessary at this point.

There are three strategies hypothesized: 1)S-R
learning, linked with 4:00 training only, and 2)
bi-conditional learning strategy and 3)ana10g
transformation of a representation, the latter two linked
with 4:00 and 6:00 training. The S-R learning strategy is
based on a simple temporal discrimination which can be
learned from training of a single example (i.e., 4:00). In
4:00 training, maximum reinforcement can be obtained if
birds respond 1) left to trials with no delay (perceptual
type), 2)left to trials with a short delay (imagery type),
and 3) right to trials with a long delay (violation type).
If this temporal discrimination is applied to 5:00 probes,
responses to the perceptual type and the violation type
with the long delay transfer because the length of delay -
response associations still apply (i.e., "no delay - go

left, long delay - go right"). However, responding to the

37
5:00 violation type with the short delay transfers
NEGATIVELY, because short delay means "go left” in training
and short delay indicates "go right" in probe testing. And
the 5:00 imagery type carries no association, since the
length of delay here is not previously observed in
training. The response prediction associated with this
strategy is accurate responding to the 5:00 perceptual type
and to the 5:00 violation type with the long delay, very
low response accuracy to the 5:00 violation type with the
short delay, and chance performance to the 5:00 imagery
type.

Bi-conditional learning strategy is based on temporal
discriminations conditional to location which are learned
from two examples (i.e., 4:00 and 6:00). These are still
S-R associations, but they are based on two dimensions:
time and location. Different length of delay - response
associations are used, depending upon the location
condition. If a 4:00 trial is presented, birds respond left
to no delay, left to a short delay, and right to a long
delay. If a 6:00 trial is presented, birds again respond
left to no delay, but right to a short delay and left to a
long delay. For 5:00 presentations, then, response accuracy
remains high only to the 5:00 perceptual type, for which
the association "no delay - go left" is still valid. But
for the 5:00 imagery and violation types, birds respond at
chance level because their learned temporal associations do

not apply to the new location condition.

38

Analog transformation of a representation is based on
certain cognitive processes. First pigeons must retain some
memory or mental representation of the stimulus in
movement. This representation must contain information
about length of time of movement relative to stimulus
location. This information is encoded in an analog form,
for instance, in an image-like form in which the stimulus
is actually represented as moving, or by some absolute
timing mechanism through which specific "times” are
associated with stimulus movement to specific locations.
Both the example representational methods involve analog
transformations: for each there is a one-to-one
correspondence between a physical transformation and a
mental representation.

By learning how to respond to 4:00 and 6:00 examples,
pigeons mentally represent stimulus movement in some analog
form. When 5:00 probes are presented, pigeons can respond
to them accurately and immediately because a representation
of 5:00 is already retained within analog representations
of 4:00 and 6:00.

Besides the analog process of representation, a
general cognitive rule is used to determine the choice
response. This cognitive rule is as follows: If the
stimulus appears at the accurate location given the length
of time of movement (plus delay time), choose the left key;
but if the stimulus appears at a location which violates

continuous stimulus movement, choose the right key. By

39
representing the moving stimulus in some analog form and by
applying this cognitive decision rule, pigeons are able to
respond to all 5:00 probes as accurately as they respond to
4:00 and 6:00 training examples.

Which of these strategies did pigeons use in different
phases of this experiment? Certainly the strategy supported
strongly by the probe data after 4:00 and 6:00 training is
the analog transformation strategy. Figures 10a-e show
birds' percent correct scores to the various trial types
after 4:00 and 6:00 training. Figure 10f shows the various
predictions made from the two strategies which apply to
this type of training. The bi-conditional learning strategy
does not explain the accurate 5:00 probe data because the
data are significantly higher than 50% (as was demonstrated
by the sign test presented earlier). The results confirm
the prediction that birds followed the cognitive rule and
represented stimulus movement by some analog process in
order to solve the problems posed by the probes.

The strategy which best explains the 5:00 probe
results after 4:00 training is the S-R learning strategy.
Figures 11a-e show birds' percent correct scores to trial
types after 4:00 training. Clearly, the 5:00 perceptual and
violation type with the 6:00 delay transferred, as was

predicted by the S-R learning strategy. And response

Itorrect

IPercent

100-
90 —
80 —

60-1

40 d
30 —
20 _
10

 

 

 

 

100-
90 ..

70-1

L

50

88
L]

l

 

 

perceptual

 

 

 

perceptual

 

perceptual

1+0

   

 

 

 

imaoeru

 

 

 

   

 

Bird ‘25

 

4 :ooD
5:00-
6:00-

 

 

 

Bird '2

 

4:001:
5 :00.

6 col

 

 

 

violation

Bird '9

 

 

 

 

 

 

violation
Figures loa-e. Percent correct scores for each bird after
4:00 and 6:00 training.

Correct

lPercent

1001

40-
so-

10—

 

41

 

 

 

 

 

.,
S .i

,

....

 

 

 

 

8
14L]

20—1
10.4

 

 

perceptual

 

 

 

 

 

100-
90-1
90.
70-
60-
50-l
40-

10-

 

 

 

perceptual imagery violation

 

 

 

 

perceptual imagery violation

Figure 101. Response accuracy predictions to 5:00 probes

42

 

4.
I). ((.k (ICC,

 

 

 

 

 

 

 

 

 

 

 

violation

imagery

perceptual

 

 

 

 

 

 

 

 

 

 

9
a
d
.w
B

n

m

h

h

V

m.

m

.m

m

w

c

- w

 

 

 

100-

70-

50-

 

 

 

 

 

 

 

 

 

 

 

 

 

10—

violation

”I

perceptual

1 la-e.Percent correct scores per bird after 4

1111110.

00 lf'd

igure

F

Correct

Percent

;-. .
rte-x-

m‘i‘l
\‘vfi'
\' . . -

. a

h

(A
-.
.

 

 

 

 

 

 

 

52252;:
c \

a

 

5 v
5....
5 1':

 

 

43

 

 

 

 

 

 

 

 

 

perceptual

 

 

imagery
1 1f. Response accuracy predictions to 5:00 probes
after 4:00 training.

   

utytkfl. ~.<-.
. .' ‘C' .

 

 

 

 

     

  
  
   

xfiiiidflx-

 

violation

44
accuracy to the 5:00 imagery and the other 5:00 violation
type was obviously at or below chance level in every case.
Moreover, the S-R learning strategy predicted that
responding to the 5:00 violation type with the 4:00 delay
should be near 0%, since the learned association transfers
negatively in this case. Indeed, the range of scores for
this 5:00 type was between 4 and 13% for all subjects.
Figure 11f describes the predictions. The data clearly
support the prediction made from S-R learning strategy. The
strategy first used after 4:00 training was an S-R
strategy. This strategy did not work for both 4:00 and
6:00. Therefore, the birds had to learn a new strategy. The
interpretation which best fits the data is that the birds
represented stimulus movement in an analog fashion and
applied a cognitive decision rule after 4:00 and 6:00

training.

Practice Effect

Another question of interest was whether birds
performed so much better on 5:00 probes after 4:00 and 6:00
training becuase they had previously been exposed to these
probes and had learned how to respond to them by the second
test. Figures 12a-e show birds' percent correct scores for
each session throughout the experiment. The point of
interest is that in the first 5:00 probe phase, the birds
consistently responded around chance level without any

pattern of increased response accuracy across sessions. In

 

 

r M
Key 0 4 00 trawmg
For 0 6.00 tramng
L Graphs: I 500 probe testing
1

 

 

Figures l2a-e. Percent correct scores for each bird for each
session throughout the experiment.

46

O—

ON

OM

Ov

om

O0

Oh

OD

OO

OO—

OV- OM—

:DD_bbbbhbhbb—bbbhbbhbb_bbbhbbbbb—bbbhbhbbDI—bbbbbDbbb—bbbbbbbbb—hbbhbbbbb—bbbbhbhbb—b bbbbbbbb—bbbhbbbz—bbbbbbbbb—bbbbhbbbb_bbbbbbbbb_bhbbbbbbb

ON. O:

OO—

OO

o:e.umem

OO

Os- O0

on O?

O» ON

O.

O

 

 

menace
oonm

1“) 1‘ it‘ii‘ii‘

mc_c_oc~

oouoioouv

‘N‘iiiiii A!t“i“““‘i‘l‘““1“‘ii

l“““ itid‘NR 1“ idi‘i44‘1

9.23: con

w h OLE

menace
0on

-:>

 

s2

oc_c_oca
oonv

 

 

it 11“iti“““‘i‘i‘i“‘i‘i“i N‘tti“ii {411‘

A

 

O—

ON

on

c> <3 <3
10 v1 iv
1331.103 iuazuad

O
p

O
Q

OO

OO—

47

«Baammfim
31 OM— ON— O: OO— on on Oh OO on O? OM ON O— O

 

     

 

 

 

      

     

)38JJ03 iuezued

 

 

   

 

 

O ::Faun”faunarumntrufimtru.:”.i._uu.n:_.i.uuaLttufir”mummuumum—uant”—.u:auruxmummuutm O
u -_ ........ n .
. u . u
u u u . 9:58
0— l. mmnoia . 9:50.: . . manor—a “ . . all 0—
. . . . . .
comm moocooem mess: cos n 8m " 8o
ON 1. u u m m i ON
. . _ .
_
u u _ u
on i. m n u u .1 On
.
. . _
L n n u . n
8. . u u u n i av
_ .
u u u u
. . E u
a .. u " 3N .. i .. a
.
. . A
u u m m .
0° L! n u . . - AT! OO
" u u u
. .
.. u u u
. n . . u .
n u . .
o. . . .
OD 1r . A u .1 O0
. . .
. .
u u m m
09 L. m u _ . .l 00
l u n u
. . . _
. . mm 35 .
8. L. n u a m u t 8.

 

 

48

bbhh _-hbbbhth—bbbbhbbDb—bbhbhbbt_bbhbhhbhb_hhbhbDbDh—bbbbbthb—bbbbbbbbb—bbbtbhbh—b :bbbbti—hbbbbbbbh_bbbbhbbbb—bbbbhthb—bbbhbIbbh—bhbpbbbbb

«5.30m

Oh OD OW O?

 

T

 

1111 11111111111111 111111111111111111111111111111111111111111111111

9:50:

006.00%

4

—‘

1
--O----------------- -----------------------------------—--

 

 

 

 

 

11 11111111111111111111111111111111111111141 1111 1111 1111 1111 1 1111

@559; 8c

 

m. a. Ohm

9:59: .

 

 

 

--- -------d.------------------------

3
on

OM

O
V

o
11':
33:1qu waned

O O
N 10

O
O

OO

OO—

49

«5.3.5
on: OM. ON— O: OO— 00 00 Oh O0 on O? On ON O— O

 

     

 

   

 

     

 

O : ........... . .:Lu.$33”mexflrwimmTumunruuxumrut.Jarmafia?:5..._.::::_:::.:_.: _
u n -1111- - 11w .............. “flanntux.” O
.
o. i. 38.5 “ 9:59: m u m
. u . " $8.5 _ 9:50.: i o.
m com 83.81 m . u . u .
ON . u “ 550.5 00.0 . ooh " 09v
. _ . .
u u u " .. as
on i. u u _ u
_ . .
. u n n :1 OM
" . u .
s. 1. u m u m
m n m u T 8.
. —
cm i. m u u u
. u u n i on
u . . . :
O0 1. . . u u
u . . . I
u m u .. oc
.
2. l. n m m m
u u u r 2.
on ... . m I m
. . . il.
m a a m 8
cm i. n u u
u m _ i on
. .
oo. 1. _ u M n
. . O u. Ohm“
. .1. OO—

 

 

 

1331103 iuanad

50

Ov— On—

acopmmom
on. o: oo— oa on 2. on em ov

can .oN

O—

AV

b- —bbbbbbbbb—hbbbhbhbb—bhhbbhbbb_hbbbbbhbb—hbbbbbbbb—Db.bbbhbb_bbhbbhbbb—thbbbhbb—b bbhbbbbb_bbbbb:hb—Pbbbbbbbb—bbhbhbhbb—bhbbbbhbb—Dbhhbhbbb

 

O— 1

T

.gw 1'

om Ii.

.20 1w

2. 1,

on lr

 

OO—ir

mmooga
OOAm

 

mc_c_oga

oobuoonv mcEEb ooh

fi------------------

 

 

N... 35

 

“““A““‘4“ 4“ {“““““A““‘1“1“““““ ““1‘1“‘ “““““4‘ “‘44‘ ‘1“‘4‘4‘ “‘{‘ 4111 1 1“{““““11““1

moaogn
nxvm

 

 

 

     

mc_c_ogu an

ooaV

 

“ “

 

O—

Om“

O O O O
O ID V M
1331.103 luazuad

O
N

O
O

Oar

azu—

51
the second 5:00 probe phase, birds consistently responded
around 80% from the first session of probe testing without
any systematic increase in response accuracy across
sessions. Two sign tests were run to verify this finding
and thereby reject the hypothesis that the birds had
”learned" how to respond accurately to probes from
practice. The first compared the mean of each subject's
first five scores with the mean of its last five scores
within the first 5:00 probe test phase. The rationale was
that if subjects were improving with practice, there should
be a tendency for the last few scores to be consistently
higher than the first few scores within a particular probe
test phase. The result of this sign test, depicted in Table
3, was that there was no consistent increase or decrease in
scores between the first and last five probe sessions. Thus
the results of probe testing do not appear attributable to
practice.

Table 3.
Sign test of mean percent correct scores per bird to first
and last five probe sessions in first probe phase.

Subject First 5 Last 5 Difference
25 57.5 50.0 +

2 52.5 57.5 -

9 45.0 60.0 -

7 50.0 55.0 -

8 60.0 52.5 t

Nonsignificant Difference (N=S, x=2; p=0.50)

52

The second sign test compared the first and last

scores per subject
again to determine
to practice. Table

this test.

during the second probe test phase

if there was an increase in scores

4 shows the nonsignificant results

five

due

of

Sign test of mean

Subject First 5
25 80.4
2 80.2
9 72.2
7 86.0
8 83.5

Table 4.

percent correct scores per subject from
first and last five sessions of second probe phase.

Last 5

79.4
78.0
77.0
83.0
84.5

Difference

+
+

+

Nonsignificant Difference (N=5, 1’3; P=0.80)

The result of the sign tests was that the increase in

response accuracy in the second 5:00 probe phase was not

simply due to practice,

increase across sessions in either of the probe test

phases.

because scores did not consistently

A final sign test was conducted to demonstrate that

scores in the first 5:00 probe phase were significantly

different from scores in the second 5:00 probe phase.

Response accuracy was significantly different between the

two testing phases (p = 0.031).

DISCUSSION

Significance

This study tested pigeons' performance to novel
instances after specific types of training. The purpose of
the test was to determine whether pigeons can apply analog
transformations to represent movement. This purpose was
accomplished by a behavioral demonstration of positive
transfer to novel probes (5:00) after extensive training of
two examples (4:00 and 6:00). The interpretations from the
two test phases of this experiment were as follows: 1)after
training of a single example, chance level performance to
some of the novel probes indicated the use of a simple
temporal discrimination, and 2)after training of two
examples for which the temporal discrimination no longer
applied, highly accurate performance to all novel probes
demonstrated the use of some generally applicable cognitive
process. No S-R learning strategy accounted for the highly
accurate performance to probes; a cognitive process was at
work.

The following processes emerged from this study.
First, pigeons demonstrated the use of a cognitive process.
The results of the final phase of the experiment support
the interpretation that pigeons employed a cognitive
process in order to respond so accurately. The
counterargument, that pigeons used a bi-conditional

53

54

learning strategy, was not supported by the data since the
data were significantly better than 50% for all probe
types. The only explanation predictive of the data is that
the pigeons acquired some cognitive process by which they
could predict stimulus location based on remembrances of
speed of motion, time, and location.

Second, pigeons used an analog process in order to
respond accurately. By this statement, I adhere to COOper
and Shepard's (1982) definition of an analog process:

We classify an internal process as the analog of a
particular external transformation if and only if the
intermediate stages of the internal process have a
demonstrable one-to-one relation to intermediate stages of
the corresponding external process -- if that process were
to take place. (1982, p 13)

The 5:00 example was an intermediate stage of the rotation
of the stimulus from 4:00 to 6:00. In training, this
intermediate stage never took place in physical form. But,
accurate performance to 4:00 and to 6:00 indicated a unique
correspondence between choice response and external stage.
Consequently, accurate responding to 5:00 probes after 4:00
and 6:00 training indicated a one-to-one correspondence
between this physcial intermediate stage and its
represented internal intermediate stage.

Note that this does not necessarily mean that pigeons
use "imagery", i.e., that they rotated a representation of
the stimulus in their heads to represent 4:00, 5:00, or

6:00. To speculate, they could just as easily have started

counting when the stimulus moved from 12:00 to its end

55
location. By this method, they retain a true "number"
system to represent 4:00 and 6:00 from which they could
generate or generalize the representative ”number” for
5:00. For example, they might count to ”10” in a perceptual
4:00 trial and they might count to ”15" in a perceptual
6:00 trial. From this they know that if the stimulus ends
at 4:00 and they count to any number other than "10", they
respond on the right key, but if they count to "10" and the
stimulus staps at 4:00, they respond on the left key.
Similar rules apply to 6:00 trials: if the stimulus stOps
at 6:00 and they count to "15", they respond to the left;
if the count to any other number while the stimulus stOps
at 6:00, they respond to the right. To a novel perceptual
or imagery 5:00 probe, they may count to "13”. Now they
must decide if this "number" is apprOpriate, given the 5:00
location, and then respond accordingly. This "number”
system is not an erroneous suggestion, for there is some
evidence that pigeons can count to time objects (Davis &
Memmott, 1982).

By using "rotating pictures in the head", or by
applying the true ”number" system, accurate performance to
5:00 probes results. And, in both cases an analog process
is being used. For each, there is a one-to-one
correspondence between each physical stage (4:00, 5:00, or
6:00) and its internal stage (i.e., a pictorial
representation, or a "number").

Third, pigeons were demonstrated to have behavioral

56
plasticity. That is pigeons behaved apprOpriately when
stimuli of the environment changed. Moreover, pigeons'
strategies of response were demonstrated as somewhat
plastic in this study. Pigeons used an S-R strategy when
training was based on a single simple example and when that
strategy produced high levels of reinforcement. Pigeons
switched to a cognitive strategy after the S-R strategy no
longer produced reinforcement, and after they had been
exposed to sufficiently complex training examples. This
study demonstrated behavioral plasticity in the pigeon, and
more importantly, it tracked the deve10pment of the

acquisition and use of a cognitive process.

Limitations of Study and Future Directions

This study supported a general cognitive process
interpretation, it indicated that an analog process was
being used, and it demonstrated some behavioral plasticity.
But these general findings lack specific description. What
components of the stimulus controlled or cued the cognitive
process? What form did the analog process take -- an
imagerial form, a unique representative form? How plastic
were the demonstrated behaviors? These questions cannot be
answered by the present study. Certain experimental

manipulations are suggested which test these issues.

Constraints of the Cognitive Process. There are two

stimulus dimensions about which information must be encoded

57
in order for the process to work: time and location.
Decisions about choice response must be made from
information about the duration of the delay as well as
about the stimulus location at the beginning and at the end
of the trial. Information processed about either of these
dimensions may provide the context for responding to probes
and for applying the process. For example, it may be that
the location of the trained examples must be physically
near the location of the probes in order for there to be
accurate application of the cognitive process. Or, it may
be that, which increasingly longer delays between stimulus
movement and final stimulus presentation, application of
the analog process deteriorates. In order to determine the
context specificity of the process and the limited
application of the process, I need to find location and
time constraints. Location constraints could be tested by
introducing a new probe example, i.e., 7:00, within
different training session. Accuracy of response to 7:00
perceptual, imagery and violation proves could be collected
within 4:00 training sessions, and within 6:00 training
sessions. I hypothesize that within 6:00 training, positive
transfer would occur to 7:00 probes, but within 4:00
training, positive transfer would probably not occur. This
hypothesis suggests that the cognitive process uses some
generalization of time-location which is only accurate if
applied in a limited fashion and only to examples

physically near those in training.

58

Another variable which may prove important in the
cognitive process is initial placement of the stimulus. In
order to test this, the start location of the stimulus can
be altered, while the other dimensions of the task remain
constant but altered by the same amount. For instance, I
could start the stimulus at 1:00, and st0p it at 5:00, 6:00
and 7:00. This experiment tests the variability in the
process for overall general physical changes (i.e.,
complete transformation by a few degrees).

Time constraints could be investigated by lengthening
the duration of delay within a few probe trials and within
a few training trials. This would test the hypothesis that
the process is inaccurate if taxed by long delays. In
addition, the accuracy of the time component of the
cognitive process can be tested by systematically altering
the delays already used in training. For example, how much
longer must I extend the 0.75-second delay before pigeons
treat it as a 1.0-second delay? This test would indicate
how generalized or how specific time discriminations are

within the cognitive process.

Form of Analog Process. In order to determine if
subjects are using visual images to solve a task, cognitive
psychologists typically introduce a visual interference
task and observe whether performance deteriorates. Such a
task is invoked while the remembered stimulus of the

primary task is absent, so that the visual nature of the

59
secondary task interferes with the visual-like mental
process required to solve the primary task. In the present
experiment, an hypothetical imagery process would be used
during the delay in which the rotating stimulus is absent.
In the delay, subjects using imagery would have to mentally
rotate a representation of a stimulus in their heads in
order to have an accurate picture of stimulus location at
the end of the delay. To interfere with the imagery
process, a simple secondary task which requires a visual
discrimination could be employed. In theory, performance in
the primary task should deteriorate if imagery is being
used since the visual system and visual memory are being
taxed by a secondary task. In practice, performance of the
primary task could deteriorate due to any type of
interference, for it is well-established that pigeons'
performance changes drastically with small environmental
changes (i.e., turning the houselight off in the middle of
a task can severely impair performance in the task). The
form of the analog process is difficult to discover,
because of the critical tie between environmental change
and pigeons' performance. A better idea might be to pursue
a set of converging Operations which point to an analog
process, i.e. like representational momentum (Fryde &
Finke, 1984) and directional scanning (Finke & Pinker,

1983).

Behavioral Plasticity. Another variable of interest is

60
the changes in response and the strategy shifts caused by
changes in the environment. The plasticity, or the general
applicability in the face of change, of the cognitive
process and the resulting behaviors can be investigated by
altering crucial components of the stimulus, like its speed
of motion, its direction of motion, the duration of its
motion, etc. Variable manipulations mentioned in the
constraints section also test the plasticity of behavior in

this task.

LIST OF REFERENCES

List of References

Cerella, J. Visual classes and natural categories in the
pigeon. Journal of Experimental Psychology: Human
Perception and Performance, 1980, Vol. 5, No.1, 68-77.

COOper, L.A. a Shepard, R.N. Chronometric studies of the
rotation of mental images. In W.G. Chase, ed., Visual
Information Processing, New York: Academic Press,

1973.

Davis, H. & Memmott, J. Counting behavior in animals: A
critical evaluation. Psychological Bulletin, 1982,
v01.92, No.3, 547-571.

Finke,-R.A., & Pinker, S. Directional scanning of
remembered visual patterns. Journal of Experimental
Psychology: Learning, Memory and Cognition, 1983,
Vol. 9, No.3, 398-410.

Freyd, J.J. 8 Finke, R.A. Representational momentum.
‘ Journal of Experimental Psychology: Learning, Memory
and Cognition, 1984, Vol. 10, No. 1, 126-132.

Granda, A.M. & Maxwell, J.H. Neural Mechanisms of Behavior
in the Pigeon. New York: Plenum Press, 1979.

Griffin, D.R3 The Question of Animal Awareness. New York:
Rockefeller University Press, 1981.

Herrnstein, H.J. & deVilliers, P. Fish as a natural
category for peOple and pigeons. In. G.H. Bower (ed.)
The Psychology of Learning and Motivation, Vol 14.
New York: Academic Press, 1980.

Herrnstein, R.J., Loveland, D.H., & Cable, C. Natural
concepts in pigeons. Journal of Experimental
Psychology: Animal Behavior Processes, 1976, 2,
285-311.

Hollard, V.D. & Delius, J.D. Rotational invariance in
visual pattern recognition by pigeons and humans.
Science, 1982, Vol.218, November 19, 802-804.

Jagacinski, R.J., Johnson, W.W. & Miller, R.A. Quantifying
the cognitive trajectories of extrapolated movements.
Journal of Experimental Psychology: Human Perception
and Performance, 1983, Vol.9, No.1, 43-57.

Jerison, H.J. Evolution of Brain and Intelligence. New
York: Academic Press, 1973.

61

62

Kosslyn, S.M. Image and Mind. Cambridge, MA: Harvard
University Press, 1980.

Kubovy, M. & Pomerantz, J.R. Perceptual Organization.
Hillsdale, NJ: Lawrence Erlbaum Associates, 1981.

01ton, D.S. Characteristics of spatial memory. In S.H.
Hulse, H. Fowler, & v.K. Honig (Eds), Cognitive
Processes in Animal Behavior, Hillsdale, NJ:

Lawrence Erlbaum Associates, 1978.

01ton, D.S. & Collison, C. Intramaze cues and "odor trails"
fail to direct choice behavior on an elevated maze.
Animal Learning and Behavior, 1979, 7, 221-223.

Premack, D. The codes of man and beasts. The Behavioral
and Brain Sciences, 1983, Vol. 6, No. 1, 125-168.

Pylyshyn, z.w. Computation and Cognition. Cambridge, MA:
MIT Press, 1984.

Rilling, M.E. Stimulus control and inhibitory processes.
In v.x. Honig and J.E.R. Staddon, (Eds) Handbook of
Operant Behavior. Englewood Cliffs, NJ: Prentice-Hall
1977.

Roitblat, H. L. Codes and coding processes in pigeon
short-term memory. Animal Learning and Behavior, 1980,
8’ 341-3510

Roitblat, H.L. The meaning of representation in animal
memory. The Behavioral and Brain Sciences, 1982, 5,
353-406.

Schwartz, B. Psychology of Learning. New York: W.V. Norton
& Co., 1984.

Shepard, R.N. & COOper, L.A. Mental Images and Their
Transformations. Cambridge, MA: MIT Press, 1982.

Shepard, R.N. Ecological constraints on internal
representation: Resonant kinematics of perceiving,
imagining, thinking, and dreaming. Psychological
Review, 1984, Vol. 91, 4, 417-447.

Straub, R.0., & Terrace, H.S. Generalization of serial
learning in the pigeon. Animal Learning and Behavior,
1981, 9, 454-4680

Wright, A.A., Santiago, H.C., Urcuioli, P.J., & Sands, S.F.
Monkey and pigeon acquisition of same/different
concept using pictorial stimuli. In M.L. Commons,

R.J. Herrnstein, & A.R. Wagner (Eds), Quantitative
Analyses of Behavior: Vol III, Acquisition.
Cambridge, MA: Ballinger, 1984.

63

Zentall, T.R., Edwards, C.A., Moore, B.S. & Hogan, D.E.
Identity: The basis for both matching and oddity
learning in the pigeon. Journal of Experimental
Psychology: Animal Behavior Processes, 1981, 7, 70-86.

IHHIIIH

45 7652

3111111.

11111111