E , .‘ ‘ . 1--. 7'

' 4L. 53‘»?- 5-3 , ‘ 9:.th

. -- "3"”- :5 ‘3‘":373) ' - ‘ > h n; 4 3,92?”

. £52572 .:- «m: . , ,3. v- c 5 - 2”.” . ,

P" Vf-ir‘ ' ,.
‘. . _, .v

_ ‘ z: . as?“ . .

; ‘ til-.7
'I ..\ .3?: ‘7’ I:
1' v '«nrz‘. ‘“ .
a . U
_ $733331? ..

.. , .

‘ ‘ '“L . ‘

. 1 .1 u
.4
3 ~ -» - 3;: z . 3.: 2:
n . ‘1‘ ,_ a. ,. _- V n . J - 4L§Kf~MV 42-:
t. . J- 1)“; ‘ A 5», ' ' t ' 3‘ 7 . - '. ' waif-(7;, ;.., WW
1 " T V '- ,. “" ' "' '1’“ ’ ~ -‘ " » - - " ~'?‘.;::.m~ raw».
u' ~ ¢-- 4.4-.. h- - 1 .V _ ~ , 4.. y u-Quh'v
V '- A“ ,4 -‘ ‘ J-y ~-

I“ 1‘ ' ‘
..: 6 , .5“;
I. .I‘|!lfigl,; .J .y
"t'itlét
n u“ 7:

5 A
13"” 51.-

13..

”vast .T. .
‘ yaw», ‘

z

u-
‘9

 

, 41- ‘ . -
. “In” ,0, M 4:5. 3..» _ ..
' "z: . "¢‘:\;‘{g’tj1{‘?§-‘&¢L§, ' - "' , , m \1
u 7», :, 1:1 :1 ‘ 1 fl onfijma‘h‘i 1:in W ‘ _ . ‘3: a. “‘a‘fik...ii$fl-:5 7
Z ‘57“? .‘ 3:3" . w, ‘j‘§~_~“‘i 5",“, 4,- ricer-“LN ~13.» firm-535:: -
_ ‘¥ «gram A u . M > _. '
5- 5.3.; quia’E-‘E’ "‘ 4"“‘é'w :2. - w .1 32+
~‘«\ ' ..,.~.~ .‘ .‘I
‘ um: w ¢L~muegmxzrlv ‘ 3‘53" ‘ 1 WW ~s '3?“
u. u 4_ Avataklilam k v .
N ”L V -

.H'

‘ ‘ u
w 1%:- m ~ - w-
s _"~.:“~. w .u . >~ ,3- - ‘ u . . x _V_ __.~ \
1hr ‘ ”91?. ‘ ‘ “m ' .1! ' 332a? " , ‘1 §#::§:‘m7. “:4”
A! V' ' ‘: ‘ L-M— ' ‘
w l f .

b ., ,.
v. mu 4‘1?“
.\ . ~ ‘

.1
:3. J:
L .

. --' "- .:u‘.-
. ‘ . V V_ . ‘ 4:. ~ Nk.‘;..‘ m. .p .
, ‘ -nl
_. ~4- . u , ‘
and u“,

. 4331‘:

 

 

_ 1‘ u 694
‘ .3, . .‘ - Hffibk.‘
' ‘ I .4 . . , v I ‘ y
x ..
35“
in?“
V C
. ‘1‘“; 4'“
l 5'. 4.3”,
”‘71:;- .
l . . v“
‘5' )I- 'L‘N' "KY"
, I "“5337?! N
‘ P
,' .
.
L
. 1mg;- r t 4v
‘ -" {finifif
. 3:37 5
‘ "”ng
I a.“ .
l ' . _ ’. ‘ .14 ' ' - . r‘ 1
‘ ‘ws ~ ‘ ' - " - ‘ -. > , ' - ._ }.-.r.,-,:.:.'. ‘

Nutld
‘" 'NIT.” "-1 1
.1“ ' I
.1! .

.‘ ' ‘Im
~'~.<...,f.
u, 't' ‘~ ‘ 1 ‘

 

‘ _
”1» .
1;;
l

 

 

 

, .. .0

. ".430. n
w n7 -

« u

'1‘ 1"
. jaw;
\ :;.9~‘—';«"-g'

- 325
h I‘J‘V‘v
1:22,. 53.23;? a K
I

,
.,
1
‘E
5
.5
1
’§

0
.

w
a

.1 (-1
Lily-1‘ ‘
< r --l~v A1
[‘5‘ 1‘ "> ’
(:3. I1.”

    

 

n ‘
’7": v" I?
33.59%“

. . .. M ..

7:95-35}! ' ""-r v“ "=7 #3755.
:wmv'“ ' ~ ~5.
{11- ‘

 

 

 

e-
u

 

....!.'.‘.~' 1
,‘afia‘fi 3

' c .,
‘ ”1'11.-
k...
M

nap-rm, r I)!“

.a

546:.
r: j?

1

 

‘ ’
' ' ' ‘—~ ' (x.
, . . ‘ ’ ‘ h r" - fr‘k
4' _ :4 H r_ V , ‘ .. -- -—vv .3 '~’. .. .w ‘
. ', . . ,. ' ' A, a, ;: ' x ', “ht-7"“: L.” V 5"; .
' ‘ “rt -. , . . , ~— ,'" ..,, ,L .a
. ‘ . . , . . u» 4'- - "m ”D. V .3. ‘
v ~ ~ .. ' ri' 64,»! «4
'9 u I‘-
m - I

 

n fl-J ‘ - 2/ ~ 6 '
I ( . . - ' .
I '. . . ‘ ' I” ' I
{{- n‘y'mc’f: . H ‘L‘fi “ ' Q ‘ 1;. H fié'.-3:‘:_"

MIIANCHG

Illl

 

 

 

 

 

 

 

 

 

lllllllIUHHIHIJIIUIIHIllllllllllllllllllfil

881 0933

This is to certify that the

dissertation entitled

THE USE OF A “HalsTURTS IN OBJECT REC 05". rNITICN I\l
PIGTCONS
presentedby

Thomas Lucas LaClaire

has been accepted towards fulfillment
of the requirements for

Doctoral Psychology

 

degree in

 

14 (AW ("4734/ x!)

Major professor

DatCJDQa‘Z/ #7

MS U is an Affirmative Action/Equal Opportunity Institun'on 0-12771

 

 

 

 

l LIBRARY

 

University

¢Mlchtgan State

 

 

PLACE IN RETURN BOX to remove this checkout from your record.

TO AVOID FINES return on or before date due.

Max
DATE DUE DATE DUE DATE DUE

 

 

 

 

 

*l

 

 

 

-_

 

 

 

 

fill; ;

 

 

 

fl ’7“

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

fif‘

 

 

 

 

 

 

 

MSU Is An Affirmative Action/Equal Opportunity Institution

 

 

 

 

 

 

cMMprna-pd

THE USE OF FEATURES IN OBJECT RECOGNITION IN PIGEONS

BY

Thomas Lucas LaClaire

A DISSERTATION

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

DOCTOR OF PHILOSOPHY

Department of Psychology

1991

.F

‘? «3"
I 44-

’.—

{EL-k’

ABSTRACT
THE USE OF HIERARCHICAL FEATURES IN OBJECT RECOGNITION
IN PIGEONS
BY

Thomas Lucas LaClaire

Two experiments were designed and performed to examine
the application of Palmer’s (1977) model of object
recognition to animals. The hierarchical structural
organization of features was studied with stimuli
constructed on the basis of Gestalt principles of perceptual
organization. Initially, pigeons discriminated between two
different orientations of a two-dimensional line drawing for
both experiments. Following acquisition, positive and
negative generalization tests were performed using sets of
parts of the training stimuli in experiment one, and single
parts in experiment two. In experiment one, the pigeons
responded significantly more to the set of parts having high
proximity ratings than the set having low proximity ratings
during the positive generalization test. In the second
experiment, the pigeons did not respond differentially to
single parts having differing proximity ratings. This may
have been due to the large reduction in contour of the
single part probes. It may be concluded that pigeons

organize features into hierarchical structural units.

Copyright by
Thomas Lucas LaClaire

1991

This dissertation is dedicated to the memories of my
mother and my son who were both committed to the Lord’s care
this year. It is also dedicated to my wife whose continued

support, caring, faith and love have made the difference for

me.

iv

ACKNOWLEDGEMENTS

I would like to thank my advisor, Dr. Rilling, for his
patience and understanding. I would like to thank the
members of my committee for their comments and guidance. I
would also like to thank Tom DeMarse, Scott Krunchanaus, and
Dawn Vreevn for their encouragement. I am indebted and
grateful to Mark Warner for all of his assistance. I would
also like to thank Debbie, my wife, for her help in typing

portions of this manuscript.

Table of Contents
List of Tables . . . . . . . . . . . . . . .
List Of Figures 0 O O O O O O O 0 O O O O O 0

Introduction . . . . . . . . . . . . . . . .
The Comparative Approach . . . . . . . .
Object Recognition . . . . . . . . . . .

Concept Learning . . . . . . . . .
Cerella . . . . . . . . . . . . . .
Heinemann & Chase . . . . . . . . .
Blough . . . . . . . . . . . . . .
Rilling and LaClaire . . . . . . .
Palmer . . . . . . . . . . . . . .

Experiment One . . . . . . . . . . . . . . .
Method . . . . . . . . . . . . . . . . .
Subjects . . . . . . . . . . . . .

Apparatus . . . . . . . . . . . . .

Stimuli . . . . . . . . . . . . . .

Procedure . . . . . . . . . . . . .

Results . . . . . . . . . . . . . . . .

Experiment Two . . . . . . . . . . . . . . .
Method . . . .

Subjects . . . . . . . . . . . . .
Apparatus . . . . . . . . . . . . .
Procedure . . . . . . . . . . . . .
Results . . . . . . . . . . . . . . . .

Discussion . . . . . . . . . . . . . . . . .
LIST OF REFERENCES . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . .

APPENDIX A O O O O O O O O O O O O O O O O 0
Instructions for running PRII.05 .obj .
SOURCE CODE FOR PRII.OSB . . . . . . . .

APPENDIX B O O C O O O O O O O O O O O O O O
Clarification of Palmer's Model and Its
Application . . . . . . . . . . . .

APPENDIX C O I O O O O O I 0 O O O O O O O 0
Raw Data, Experiment One . . . . . . . .

Appendix D . . . . . . . . . . . . . . . . .
Raw Data, Experiment Two . . . . . . . .

vi

vii

52
53
60
60
62
82
82

87
87

97
97

Table
Table
Table
Table
Table

Table

B-1:

C-l:

C-2:

C-3:

0-1:

0-2:

List of Tables

Proximity relations ...........................83
Positive generalization response rates ........87
Negative generalization response rates ........88
Response latency and distribution data I ......89
Generalization response rates .................97

Response latency and distribution data II .....97

vii

List of Figures

Figure 1: Stimuli used in experiment one .... ..... ........31
Figure 2: Generalization gradients .. ................. ....35
Figure 3: Median response latencies . ................. ....39
Figure 4: Stimuli used in experiment two .... ......... ....42

Figure 5: Mean rates of responding to parts ..............45

viii

Introduction

The Comparative Approach

Comparative psychology has been faced with explaining
both the similarities and differences amongst the behaviors
of different species. This is similar to the paradox of
diversity and unity of life which Darwin attempted to
resolve through his theory of evolution. Indeed Darwin has
frequently been rendered as the father of comparative
psychology (Boakes, 1984; Domjan, 1987; Stebbins, 1990;
Macphail, 1990). While the historical roots of comparative
psychology grew in the fertile soil of learning theory
spreading rapidly from Rommanes to C. L. Morgan, thence to
America and E. L. Thorndike who was a forefather of
behavioral psychology, it ironically lost much of its
comparative flare soon coming to be the study of cats, rats,
and pigeons. It has more recently undergone a major
transformation as reviewed by Domjan and Galef (1983),
discussed by Shettleworth (1983), clearly influenced by
Garcia (1981), and described by Seligman (1970) whose notion
of a preparedness continuum was an attempt to reconcile some
of the major problems that faced learning theory. It is
from the context of this change that a new field has arisen,
not to remove or replace behavioral psychology, but to

extend it (see Terrace, 1984). This is the field of

2
comparative cognition (Roitblat, Bever, & Terrace, 1984).
The primary distinction between purely behavioral and
cognitive theories is in the reference to internal
representations by cognitive theories. The question of how
representations might logically be developed within an
organism leads quickly to consideration of perceptual
processes (Shepard, 1984; Rilling, in press). Perception is
the logical mediating process between objects and events in
the real world and the internal representations of these
objects and events. Thus, the most basic questions facing
animal cognition today involve how animals perceive the
world around them.

Stebbins (1990, p.23) in his review of the field of
comparative psychology has concluded that it has not yet
begun to handle internal representations. Indeed, the
acceptance of the idea of animals having internal
representations has taken considerable debate. But, methods
of behavioral control which came out of Skinner’s school of
behaviorism have slowly become tools for uncovering the
sensory and perceptual experiences of animals. Blough’s
(1958) method of finding sensory thresholds could be the
first hint of this trend. Rilling and Nieworths' (1987)
method of studying imagery is one current result of this
trend. It seems that the formal systems of behavioral
psychology which so adamantly rejected notions of internal

mental events (e.g. Skinner, and Watson) have had their own

3
methods turned upon themselves in a Godelian proof of their
incompleteness. While the cognitive revolution was in part
a startle response to the incompleteness of the behavioral
psychology of the forties and fifties, a more difficult and
impressive result of the growing recognition from within
animal research of the limitations of the behaviorist dogma
has been the growth of the field of animal perception and
cognition. Stebbins (1990) clearly points out the vast
difference in the requirements for experimentally capturing
perceptual as opposed to sensory processes. A perceptual
process involves the organism’s subjective interpretation of
the stimulus. The experimenter cannot provide a definition
of the stimulus through any procedural qualifications placed
upon the subject's response. Stebbins (1990, p. 16) states
that "It then remains for the experimenter to determine what
elements, features, complexes, or aggregates of the
stimulation introduced by the experimenter are controlling
the subject’s response." This study has been designed to
investigate what complexes of features control the subject's
responding in an object orientation discrimination task.

In order to do this, the stimulus must be adequately
described by the experimenter in order to have some idea of
what 'elements, features, complexes, or aggregates' might
exist within the stimulus. Gibson (1979) has presented a
novel approach to perception which involves describing

stimulation in terms of its information, particularly higher

4
order aspects which do not vary across circumstances which
are termed invariants. Rilling (in press) has suggested
that the notion of invariants within the stimulus array may
be useful in expanding the notion of stimulus control into
an ecologically valid theoretical concept for the study of
animal perception and representation. Ecological validity
refers to the degree to which a given phenomenon or
theoretical account of a phenomenon is applicable to an
organism living in its natural environment. Behavioral
psychology has been charged with having poor ecological
validity by ethologists as the Skinner box bears little
resemblance to most natural habitats. Animal cognition and
perception studies are concerned with developing empirical
procedures which remain ecologically valid by applying these
methods to problems which are common to a range of species
(see P. M. Blough, 1991).

To the extent that different species share the same
environment and face similar problems within that
environment, we might expect to find similar solutions
across species (see Olson's, 1991, discussion of spatial
memory). This is bolstered by similarities in the
underlying physiological systems which support the
organisms' perception of the environment. This work is
being done with pigeons, and the questions being raised
regarding object recognition come, in part, from theories

developed on the basis of studies of human object

5
recognition. So, it might be asked how one could reasonably
expect to be able to bridge the gap between human and
pigeon. Aside from the obvious broad similarities in human
and pigeon visual systems like the fact that each has a pair
of visual organs called eyes, there are more meaningful
similarities in the functioning of these systems.

Blough (1955) demonstrated that pigeons have a dark
adaptation curve similar to humans using a procedure based
on von Bekesy’s method of ascending and descending limits.
Pigeons have color vision similar to our own (Honig &
Uricioli, 1981). Further, pigeons have edge detector
ganglion cells (Maturana & Frenk,1963) which are similar to
cells found in the visual systems of many other species.
There are also differences like the existence of two
independent fields of view (lateral and frontal) with the
associated pair of foveal areas (’red field’ and fovea) in
each eye of the pigeon. Nye (1973) reviewed work on the
visual acuity of the pigeons visual fields and concluded
that the selection of the response measure is critical in
obtaining appropriate results. The pigeon appears to be
contra-prepared to associate stimuli presented in the
lateral visual field with pecking responses, and prepared to
associate laterally presented stimuli with locomotor
responses. The frontal field is well designed for guiding
the pecking response (Goodale, 1983). Further, the pigeon's

frontal binocular field of view is only about 30 degrees

6
wide (Martin & Young,1983; Bloch, Lemeignan, & Martinoya,
1987). It is this frontal binocular field which appears to
be most similar to the human visual apparatus. Most
research using visual materials with pigeons relies on the
pecking response. This study will also use pecking as the
response measure of choice.

Comparative perception and cognition is a rapidly
growing area which relies on empirical methodology to obtain
results about internal events in animals taking the
subject's biological makeup into account. It has been
demonstrated that these events may be investigated without
invasive procedures designed to directly measure neural
and/or neuro-chemical activity. This is much like the
knowledge obtained by basic observation about color vision
in humans prior to the existence of physiological methods of
checking the optic system for three pigments in the retina,
and finding opponent-process like neural activity. While
internal events can be investigated without immediate resort
to invasive procedures, the importance of verification and
explanation of the physiological correlates of internal
experience cannot be underestimated. The current approach
is to use empirical procedures developed by behavioral
psychology to guide the formation of formal theoretical

models which can be rigorously tested.

Object Recognition

Until recently, the comparative approach has neglected
the study of object perception and recognition. In
Quinlan's (1991) review of two-dimensional shape
recognition, the discussion of the animal-learning approach
revolves around a few studies done in the 50’s and 60's.
These studies did not produce a cohesive theoretical
approach to animal object recognition. But, Heinemann and
Chase (1990) have recently developed a sophisticated
template matching model of animal object recognition.
However, there is also mounting evidence for the use of
feature analytic processes in animal object recognition
(Blough, in press; Rilling, DeMarse, & LaClaire, in
preparation). Interestingly, the return to the question of
object recognition by comparative psychology has been
prompted by work done in another area of animal cognition

involving a somewhat more cognitive act: Concept learning.

Concept Learning

Herrnstein (1984) described a variety of concept
learning or categorization studies which led him to conclude
that categorization could be conceptualized as perceptual
object constancy. Cerella's (1990) work on pattern
perception was clearly prompted in no small part by work in

concept learning. Indeed, Wasserman (1991, MPA

8
presentation) has essentially called for an increase in the
amount of work that should be devoted towards object
recognition theory and research within animal cognition
labs. Wasserman presented results from a variety of concept
learning experiments which have led him to conclude that the
pigeons in these experiments are not actually learning a
novel concept, but are uncovering some inherent aspect of
the set of stimuli which is measurable through some stimulus
similarity metric. The precise nature of this metric has
not yet been uncovered, but it should be uncovered through
studies of object and scene recognition.

An early experiment on natural concepts done by
Herrnstein, Loveland, and Cable (1976) demonstrated that
pigeons were able to correctly categorize novel instances of
several different natural classes of stimuli. The basic
procedure involves extended training in a discrimination
procedure in which the positive stimuli contain examples of
the stimulus class and the negative stimuli contain no
examples of the stimulus class. Herrnstein, Loveland, and
Cable (1976) used three classes of stimuli: Trees, water,
and an individual person. Each set of stimuli for each
class of stimuli consisted of over 1,500 unique pictures.
About forty positive and forty negative pictures were
presented on normal daily sessions. These pictures were
randomly selected from the larger stimulus set of over 1,500

pictures for each stimulus class. Upon obtaining consistent

9
discrimination of the positive and negative stimuli, the
subjects were presented with sessions of entirely novel
(previously unseen) pictures from the stimuli sets. The
performance on these novel pictures was well above chance.
The ability of the pigeon to categorize pictures which were
selected based on a human notion of category membership
leads to the question of what is it that allows the pigeon
to perform this categorization task?

In order to provide a better picture of the process of
categorization in the pigeon, Bhatt, Wasserman, Reynolds,
and Knauss (1988) developed a more sophisticated procedure
to investigate the phenomenon. This new procedure involved
the use of multiple response keys and a multiple fixed ratio
schedule of reinforcement. The pigeons are first trained to
make a response on a viewing screen, then to respond on each
of four corner keys. Using just ten slides to represent
each of four categories, each corner key is assigned to a
given category. The pigeon is reinforced for responding on
the correct corner key after viewing any slide from the
category that was assigned to that key. Incorrect responses
resulted in correction trials in which the same picture
continues to be presented until the pigeon makes the correct
choice amongst the four keys. Only the response to the
initial presentation of any given picture is counted in
terms of the percent of correct categorization. Bhatt et.

al. (1988) were able to demonstrate significant transfer to

10
novel instances of the trained categories using this new
procedure. Furthermore, Wasserman, Kiedinger, and Bhatt
(1988) used this new procedure to test pigeons abilities to
develop subcategories, and introduced a control for the
possibility of pigeons being able to develop arbitrary
categorization schemas. In the sub-categorization
experiment, Wasserman et. a1. (1988) presented 40 slides of
each of two categories per session and required the pigeons
to discriminate both the categories and subcategories by
responding on one of four response keys as previously
described. The subjects were able to learn both the
categorization and sub-categorization tasks. However, more
errors were made on the sub-categorization task than on the
categorization task. The main importance of this finding is
that the conceptual transfer found previously by Bhatt et.
al. (1988) was probably not due to an inability to
discriminate members within a category from one another.
This suggests that pigeons are able to form a category based
on some similarity amongst the members of the category while
being able to distinguish between the members of the
category which requires a flexible, yet detailed
representational format. In the second experiment done by
Wasserman et. a1. (1988), the pigeons were trained using
either human language categories or using pseudocategories
generated using equal numbers of slides from each of the

sets of slides used in the human language category training.

11
The basic procedure was similar to the procedure previously
described. The main result was that the pigeons assigned to
the pseudocategorization task did not achieve anywhere near
the same level of ability to perform the four-key choice
task as their human language categorization counterparts
were. This indicates that while the pigeon is able to form
categories which consist of distinguishable exemplars, these
exemplars need to have some degree of visual similarity in
order for categorization to achieve high levels of accuracy.
Using non-similar exemplars, pigeons are less able to
perform the categorization task which implies that the
pigeon’s memory is not as picture perfect as some have
implied (Vaughn and Greene, 1984; Heinemann and Chase,
1990). In Wasserman et. al.'s (1988) second experiment, the
pigeons were not able to memorize even a limited set of
eighty unique slides. This finding argues against the task
demands and against the photographic memory explanations of
the performance of pigeons in categorization studies. It
also points out the need for more research into the process
of object recognition in pigeons which forms at least a part

of the basis of the pigeon's conceptual capacities.

Cerella
Cerella (1990a) has developed an account of the
pigeon's frontal visual field capacities which is closely

linked to the perspective engendered by the motion of the

12
pigeon's head as it pecks at stimuli. It is clearly a
viewer-centered account. Viewer—centered approaches to
object recognition emphasize the use of the perspective of
the observer as the frame of reference for the
representation of the object. Problems with, and
alternatives to, viewer centered theories of object
recognition will be discussed at greater length in
subsequent sections. The reliance on head movement produced
variability in viewpoint to generate an aggregate
representation of the stimulus is very much like the
sampling distribution used by Heinemann and Chase (1990) in
their template matching model of object recognition which we
will discuss later. Cerella's reasoning is closely linked
to a series of studies (see Cerella, 1990a & 1990b) which he
performed on the pigeons ability to perceive perspective
transformations of two-dimensional stimuli. While I agree
that head movements during pecking would produce variability
in the perception of a stimulus on a response-to-response
basis, and that this helps to account for Cerella's
findings, there remains little reason to believe that this
variability bears a direct relationship with the development
of an aggregate representation of the stimulus. Indeed, as
will be revealed by a careful examination of Cerella's
methods and results, the degree of variability introduced by
the pigeon’s pecking movements is directly related to an

inability to readily distinguish between different stimuli

13
which indicates a failure to adequately represent these
stimuli as different from one another.

Cerella (1990a, p. 144) clearly indicates that probe
trials in his first experiment were "summarily terminated"
at the end of a ten second interval. This is readily seen
as an extinction condition by anyone with even a passing
knowledge of operant theory. Yet, Cerella (1990a, p. 142,
my emphasis) expects to find "a continued tendency to
respond to a probe" as evidence for perspective invariance.
A generalization gradient with an increasingly steep slope
over time is what would be predicted on the basis of the
schedule assuming the pigeon could discriminate between the
probe and the positive stimuli. Indeed, this is exactly
what Cerella (1990a) obtained with most of the probes. The
exceptions being right and left shifts in the location of
the positive stimuli to which the pigeons continued to
respond at a high rate. These shifts, however, were quite
small (approximately 1.6mm) and only changed the location,
not the arrangement of features, of the positive stimulus
relative to the frame of the display window which was 30mm
wide. It is quite likely that the pigeons did not attend to
stimulus location as this was not incorporated into the
original discrimination whereas stimulus configuration was.
In his second experiment (Cerella, 1990a), the procedure was
identical except that each probe was presented four times

per session over four sessions or 16 times (except in two

14
cases which received one extra session for each type of
probe). In this experiment the stimuli were simple
geometric forms. The positive stimulus was a single
prototype (either a chevron-like figure or a trapezoid-like
figure) and gross distortions generated by randomly
displacing vertices were the negatives. Moderate
distortions, either systematic (rotations,
enlargements/reductions, and translations) or random, served
as the probes. Once again, the discrimination task required
the pigeons to make a discrimination based on stimulus
configuration and not location or size. Thus, the deck is
already stacked against probe transformations which alter
the stimulus configuration as exemplified by the Z-rotation
probes which have a horizontal line on the bottom at the
lowest distortion level, but soon have a diagonal line on
the bottom at higher levels of distortion. Furthermore,
Cerella's procedure is still not a steady state procedure
and can only determine the degree of difficulty in making a
discrimination between a probe and the positive stimulus.
This, ultimately, means that Cerella’s procedure does not
provide us with a basis for indicating the degree to which
pigeons are able to recognize appropriate transformations of
a two-dimensional stimulus as being transformations of that
stimulus. The information obtained from the initial
exposure to a given probe type may provide a more accurate

representation of the pigeons abilities prior to the

15
learning effects produced by presenting the probes using an
extinction contingency, but the problem with the initial
training requiring the discrimination to be made on the
basis of stimulus configuration remains. Unfortunately,
Cerella does not present any information regarding the
initial trials of probe testing.

In conclusion, Cerella’s (1990a) account of pigeon
visual recognition abilities as being limited and based
entirely on viewer-centered distortions of the object is not
based on an appropriate set of experimental evidence. The
flaws in Cerella’s work involve his choice of stimuli for
discrimination training, and his use of prolonged exposure
to probes presented under conditions of extinction. While
the probe trials being presented under extinction conditions
can be controlled for by looking at the resulting extinction
curve across sessions, the initial discriminations which
were used did not provide a fair basis for comparing the
transformations which Cerella used. The discriminations
which Cerella used were biased against changes in the
configuration of the features of the stimulus relative to
the canonical orientation, but not location or size. The
resulting decrement in responding to the probes in which the
canonical orientation has been altered cannot be attributed
to the pigeon relying on a strict template matching approach

to object recognition which would reject such probes.

16
Heinemann & Chase

A sophisticated template theory of object recognition
for animals has been put forth by Heinemann and Chase
(1990). This theory surpasses the limitations of
traditional template models by allowing mental
transformations of templates. The templates are actually
complete descriptions of objects and scenes which utilize
all possible features. For example, a black outline of a
square one point (or pixel) in width on a white field would
be represented by a template which consisted of an
exhaustive listing of every black and every white point in
the scene. The primary distinction between a feature
analytic theory and Heinemann and Chases’ template model is
that features are not analyzed, but are used directly and
completely in the ultimate high-level representation or
template proposed by Heinemann and Chase. The only process
intervening between the stimulus and its internal
representation is a statistical sampling process which
represents the movement of the animal's focal attention
across the display.

Template theories have oft been criticized for
requiring too much memory space to be practical as a result
of needing too many templates to effectively recognize the
diverse range of objects that people are known to be able to
recognize. Chase and Heinemann (1989) and Heinemann and

Chase (1990) describe results from concept learning types of

17
experiments like Vaughn and Greene (1984), and Vaughn and
Greene (1983) as providing evidence that the pigeon’s memory
capacity is not a limiting factor for their theory of object
recognition. These experiments tested pigeon's memory
capacity by training the pigeons to discriminate between
unique sets of pictures which did not have any cohesive
visual features or distinguishing forms. The results
indicate that the rote memory capacity of pigeons was not
taxed with a set of more than 300 unique pictures.
Heinemann and Chase (1990) suggest that evidence of an
impressive capacity for remembering unique pictures gives
them a sound basis for developing their memory-intensive
template model. Unfortunately, this evidence does not speak
to the underlying process or method whereby the ability to
remember a vast number of unique images has been conferred
upon the pigeon. Indeed, an alternate hypothesis regarding
the underlying means utilized by the pigeon to perform this
sort of feat exists which rejects the use of templates.
This hypothesis is that the pigeons in concept learning
experiments are able to compress their representations of
complex scenes via analysis of the features in the scene.
This method of compressing the ultimate representation,
probably akin to 'chunking’, uses intermediate features like
segments and vertices and their structural relationships to
produce a compact ultimate representation.

Heinemann and Chase (1990) are wary of 'covert’

18
normalization processes for dealing with the ability to
recognize objects in novel orientations, and from novel
perspectives. They suggest that overt normalization
processes related to head movements are more appropriate and
less likely to be challengeable. In their focus on these
overt normalization processes, they utilized information
about pigeon head movements and ran a simulation of
Cerella’s (1990a) second experiment using their model. The
model matched Cerella’s data quite well. Unfortunately,
Cerella’s results are in question at this point as
previously reviewed. Thus, the utility of their model
cannot be judged appropriately on the basis of this
simulation.

Another major limitation of Heinemann and Chases’ model
is that it is inherently and intractably viewer-centered.
This criticism refers to the frame of reference utilized in
the representational format for visuo-spatial information.
The viewer-centered frame of reference only allows for the
representation of an object from the viewpoint of the
observer. The main criticism of the use of viewer-centered
frames of reference is that novel perspective views of a
familiar object would be unrecognizable as the stored
perspective view most likely afforded a considerably
different image from the new input. The frame of reference
which has been proposed to allow the rapid recognition of

familiar objects from novel perspectives is termed the

19
object-centered frame of reference. The representational
co-ordinates of this frame of reference are centered on the
object instead of the viewer. In Heinemann and Chases’
model, there is no provision for developing or altering
their ultimate representation such that it becomes
object-centered. The sampling process which generates the
representation is based entirely on the observer’s unique
perspective of the scene. At no point is there a good means
for the abstraction of the primary axis of elongation of a
given object (a requirement for the generation of an object
centered representation as described by Marr and Nishihara,
1978) or the generation of a single unique representation of

the object without a bias of observer perspective.

Blough

Blough has investigated pigeon object perception and
recognition for quite some time, and has used more than one
methodological approach (Blough, 1977; Blough, 1979; Blough,
1985; Blough, in press). Blough (1977) pioneered a visual
search task for pigeons. Reaction time as opposed to
response rate is the dependent measure of choice in visual
search tasks. Visual search is a task in which a target
stimulus is presented amidst an array of distractors. The
subject is required to find the target. Common variables in
visual search experiments are the array size and the

similarity of the target and distractor. Blough (1979)

20

found visual search effects similar to those documented for
humans such as increasing search time for increased numbers
of distractors and for distractors which were judged to be
more similar to the target form. Blough (1985) has also
investigated pigeons abilities to discriminate amongst
letters and amongst random dot patterns. Blough (1985) used
a three key discrimination procedure in which one letter
served as a positive stimulus and another served as a
negative stimulus during each trial. Every letter served as
the positive and negative stimulus at some point during this
experiment. Blough analyzed the error data to generate
confusion matrices amongst the letters of the alphabet. He
was able to identify clusters of letters which were similar
in physical features which the pigeons commonly confused
with each other. Blough’s (1985) discussion of this data
does not strongly favor feature analysis models over
template matching models, but does tend to lean in the
direction of feature analysis models. Blough does point out
that the features discovered using a particular set of
stimuli may not accurately represent universal features due
to hierarchical processing in feature models (cf. Palmer,
1977).

Blough (in press) is more strongly favorable to the
notion of a feature model for object recognition in pigeons.
Blough presented results from carefully constructed visual

search experiments which seem to reject template matching

21
models. In one such experiment, three stimuli were
constructed such that when any pair of the stimuli was
superimposed, there was an equal amount of overlap which
would produce equal difficulty in distinguishing between all
possible pairs assuming a template matching process was
being used. The differences in relationships between the
segments of the different stimuli provides features which a
feature analytic process might be able to use to
differentiate between the stimuli. The results indicated
that search was not performed with equal facility for all
target-distractor pairings. While this supports the notion
of a hierarchical feature process being utilized, the higher
level features which were assumed to be being used were not
identified. Further work using a larger set of stimuli
composed of vertical and horizontal line segments resulted
in the identification of symmetry about the vertical axis as
a possible high level feature for pigeons. Another high
level feature which was identified in another experiment
(Blough, in press) was the ’absence of a gap’.

In conclusion, Blough has done a variety of thought
provoking experiments which seem to support a feature
analytic process as the underlying mechanism of object
recognition for pigeons. While the support for a feature
analytic process is growing, Blough (in press) still has not
presented a complete model of this process. Indeed, Blough

has not been willing to set forth any definitive set of

22
elementary features, but has restricted his theorization to
the specific sets of stimuli which he has used in his work.
This is probably wise at this time as it avoids making
claims which go beyond the base of knowledge which has been
thus far accumulated. However, the need for delineating a
more complete, specific and detailed model of pigeon feature

analysis is clear.

Rilling and LaClaire

Rilling, DeMarse, & LaClaire (in preparation) have
recently completed a pair of studies of simple object
recognition. These studies used simple line drawings of
two- and three-dimensional figures. A discrimination
procedure was employed in which a figure of a cube and a
figure of a truncated pyramid served as the 8+ and the 5-,
respectively. After the pigeons had demonstrated a
significant difference in their responding to the two
stimuli, a generalization test was run using stimuli having
different amounts of contour deleted either at the vertices
or at the midsegments of the figures. The pigeons responded
significantly more to the condition in which the midsegments
remained than the condition in which the vertices remained.
This implies that the pigeon sees a set of midsegments of an
object as being more like the object than a set of vertices.
Midsegments are more heavily weighted features for object

recognition than vertices for pigeons. Wasserman et. a1.

23
(1990) have obtained an analogous result using more complex
figures with a more complex procedure with pigeons.

Having established a fair amount of evidence supporting
the use of features by pigeons in the process of object
recognition previously (Rilling, DeMarse, & LaClaire, in
preparation; see also Blough, in press), the next question
of importance is: How are these features organized during
object identification? This question has been approached
from diametrically opposed positions: The wholistic gestalt
school and the atomistic structuralist or S-R school. It
has been recognized that neither position has a corner on
the truth in the market of object recognition processes.

The suggestion that some sort of interaction between these
approaches is closer to the truth is evident in Vitz and
Todds’ (1971) and Palmer’s (1977) models of perception of
figures. Both models have suggested that there may be a
hierarchical structural organization of elements. The
hierarchical nature of these models surpasses strict
atomistic positions by allowing for intermediate
representations comprised of more than one feature or
element. These intermediate representations are in turn
combined to form higher level representations and eventually
the figure or object is represented by its own unique
representation. Vitz and Todds’ (1971) model does not
provide intuitive intermediate representations, but rather

abstract representations of symmetry and complexity.

24
Palmer’s model utilizes more intuitive intermediate levels
of representation which could be considered to correspond to
different parts of a figure. There are many possible ways
to organize the features of a given object into parts. How
the features are organized into parts will depend on
relations between the features. Some parts are, according
to Palmer, ’better’ than other parts based on gestalt laws.
A ’better’ part in Palmer’s theory is one which has a
greater probability of being described by the perceptual
system as a unitary structural unit having its own level of
analysis. Palmer (1977, p. 443) defines structural units as
"elements of perceptual representation that can be processed
as a single entity, regardless of their internal complexity,

at a global level of analysis."

Palmer

Palmer’s (1977) research indicates that the gestalt law
of proximity accounts for most of the differences in the
’goodness’ of parts. He presented an algebraic model for
determining the goodness of a part given an object. The
model was developed for empirical and not theoretical
purposes. Unfortunately, the model is most useful with
proximity as the primary dimension according to Palmer’s
empirical results. Proximity is simply denoted as the
euclidean distance between the midpoints of a pair of

segments. Each dimension within the model is weighted. The

25
proximity dimension requires a weight of -1. The scale
value denoting the relationship between any pair of segments
in an object along the proximity dimension is then
calculated by multiplying the distance between those
features by -1. This is labeled as R(i,j) in Palmer’s
algebraic model with i and j representing two non-identical
segments of a figure.

In order to determine the goodness of a part of a given
figure, first, the scale value of the relationship between a
segment within the part and each segment not within the part
is subtracted from the scale value of the relationship
between that segment and each other segment within the part.
This function is computed for all of the segments of the
given part and the resulting values are summed. The
resulting quantity is then divided by the quantity of the
number of segments within the part times the quantity of the
number of segments within the part minus one times the
quantity of the total number of segments minus the number of
segments in the part. The equation for this calculation
which represents Palmer’s model can be expressed as
GtPlF)={(f-p)*zi=1->p2j=1->p.j<>p R(i.j) -
(p-1)*Ei=1->p2k=p+1->f R(i,k)}/p(p-1)(f-p). This equation is
for the Goodness (G) of a specified part (P) given a figure
(F) where f is the number of features in the object, p is
the number of features in the specified part, i and j are

features shared by the part, and k is a feature not present

in the part. Thus, the R(i,j) values represent the within
part scale value, and the R(i,k) values represent the
between part values. The larger the goodness rating is, the
better the part is organized.

The reason for using a weighting of negative one for
the proximity dimension should now be clear. The model
calls for the subtraction of the presumably large(r)
distances between segments not contained within the same
part and the presumably small(er) distances between segments
contained by the same part given that the part’s segments
have close proximity to one another. This would result in
small (even negative) goodness ratings for parts composed of
features which are close together. However, by weighting
the scale values of the proximity relations between features
by a negative one, the within part values are subtracted
from the between part values resulting in higher goodness
ratings for parts composed of segments having high proximity
than the goodness ratings of parts which are composed of

segments which are not in close proximity to one another.

Experiment One

Two figures differing only in orientation were

generated for the purpose of the present work. This

26

27
difference provides two important controls for this work.
Firstly, the positive and negative stimuli may be broken up
into identical sets of parts. Secondly, this discrimination
does not generate an a priori bias along the proximity
dimension. These figures were decomposed into two triplets
of parts -- one triplet having a higher average goodness
rating than the other (High proximity parts X=0.65, Low
proximity parts X=0.24). The pigeons were first trained on
a simple discrimination between the two figures. After
attaining a high discrimination ratio under steady state
conditions in which the subject was not always reinforced
for responding to the positive stimulus, unreinforced probe
trials were presented to the pigeon. These probes consisted
of either high proximity or low proximity part triplets
which were separated by a small distance form one another
(0.2 cm) or a large distance (0.5cm). Each figure was
composed of nine segments. Each part triplet contained one
part with four segments, one part with three segments, and
one part with two segments. If the proximity dimension is
important for the pigeon, then a decrement in responding
should be seen to the low proximity part triplets relative
to the high proximity triplets during the positive
generalization test. Furthermore, there should be a
decrement in responding to the large separation part
triplets relative to the small separation part triplets

during the positive generalization test. Template matching

28
theories would predict that there would be no difference in
the rates of responding along the proximity dimension, but
would also predict a reduction in responding to the large
separation part triplets during the positive generalization

test.

Method

Subjects

Seven naive White Carneaux pigeons were observed
performing discrimination and generalization tasks. These
pigeons were previously trained to respond to an
electromechanical key on a fixed ratio schedule of
reinforcement and to respond to a black dot presented on a
secondary computer monitor with a touch sensitive screen
using a fixed interval schedule of reinforcement. None of
the pigeons had prior experience with discrimination or
generalization procedures. None of the pigeons had any
prior experience with the stimuli used in this experiment or
stimuli that were similar to those used herein. The pigeons
were maintained within 20g above their 80% of free feeding

weights.

Apparatus
The apparatus used here has been described in Rilling

and LaClaire (1989). Stimuli were presented on a 30 cm

29
black and white monitor controlled by a Macintosh Plus
computer. Responses were collected through the use of a
touch sensitive frame (Carrol Touch Technologies) mounted on
the monitor and interfaced with the Macintosh Plus computer
via the parallel communications port on the computer.
Reinforcements were provided and houselights turned on and
off by a Benchtop Instrument (MetaResearch Corp.) which
maintained an interface with the Macintosh Plus via the
serial port.

The experiments were conducted using software developed
at Michigan State University by T. L. LaClaire in the Rascal
programming language which was developed at Reed College.
The program used is titled "PRII.05b" which stands for
Psychology Research -- Version II.05b. A listing of this
program and instructions for its use (instructions written
by Miss. D. Vreevn) may be found in appendix A, and
questions regarding it should be forwarded to the author at
the Department of Psychology at Eastern Illinois University
in Charleston Illinois. This program provides an easy
interface for the design and performance of simple
discrimination and generalization experiments using the
hardware previously described. It uses ’pict’ files for the
storage and presentation of stimuli which may be generated

by a number of standard drawing packages like MacDraw (tm).

3O

Stimuli

The stimuli were generated using MacDraw (tm). The
stimuli used in the training phase were two different
orientations of the same figure. The figure used was a
two-dimensional representation of a rectangular cube with
hidden lines erased. Three faces of the rectangular cube
were present in each view of the stimulus. One face was a
square, and the other two faces were parallelograms. These
figures were constructed from nine segments of varying
lengths. Probe stimuli were generated by breaking the
figure into three sets of segments called parts. These
parts were judged on the basis of the relative proximity of
the segments within and between the parts. One set of parts
had high proximity as indicated by a high goodness rating,
and the other set of parts had low proximity as indicated by
a low goodness rating. Probe stimuli consisted of a
complete set of parts (either high or low proximity) which
were separated by either 2mm (small separation) or 4mm
(large separation). All stimuli and ratings are presented
in figure 1. The definitions and equations required for the
calculation of the goodness ratings of the parts and
examples of the application of these equations to these

stimuli are provided in appendix B.

Training Stimuli

High Proximity

Mean Value . 0.65

Low Proximity

Mean Value =- 0.23

 

Figure 1:

31

Stimuli used in experiment one

32

Procedure

Discrimination training began with the presentation of
the positive and negative stimuli in randomly arranged
blocks of ten trials. Each kind of stimulus was presented
thirty times per session for a total of sixty trials per
session during the initial training phase. Each stimulus
was presented for approximately 103. Trial onset was
signaled by turning on the houselight in the chamber. The
houselight was turned off at the end of each trial. The
positive stimuli were presented on a F1 105 schedule in
which the stimulus presentation was terminated by the first
response after 10s had passed. This response produced a
reinforcement of access to mixed grain for 55. The negative
stimuli were presented for ten seconds and then the screen
was cleared and a trial onset delay was initiated dependent
on the number of responses to the negative stimulus. This
delay period was increased by 2s for every response if there
were more than three responses. Thus, a negative trial in
which there were two responses would produce no time-out
period, but a negative trial in which there were five
responses would produce a time-out period of ten seconds.
Once a pigeon’s discrimination ratio had risen above 0.80 on
two consecutive sessions, the block size was decreased.
Block sizes were decreased from ten to five to three to one.
Upon attaining a discrimination ratio of greater than 0.80

on two consecutive days with a block size of one (positive

33
and negative trials completely randomized), five
unreinforced trials of the positive stimulus were
introduced. A total of sixty-five trials per session were
given during this phase of training. This produces a
resistance to extinction of the response to trials in which
a stimulus similar to the 8+ is presented but not reinforced
which provides us with a steady state testing situation.
Despite this procedure, the responding of the subjects to
probe stimuli has been observed to decrease (see Rilling and
LaClaire, in preparation). Sessions of training with the
added unreinforced S+ trials lasted until the stability
criterion of two consecutive trials with a discrimination
ratio of greater than 0.80 was met.

At this point, generalization testing was begun. Each
session of generalization testing consisted of the
previously described trial types, plus eight novel probes.
These probes were presented under conditions of extinction
without added time-out for responding during these trials.
There were two trials of each probe type per session
resulting in a total of eighty-one trials per session during
generalization testing. The eight novel probe types were
generated using the eight combinations of the following
three pairs of variables: Positive vs. negative stimulus,
high vs. low proximity, and large vs. small separation.

Five trials of generalization testing were run, but only the
first three were analyzed due to the reduction in responding

to the probes observed over time.

34

Results

The pigeons rapidly acquired the initial discrimination
reaching criterion performance at the lowest block size with
8+ Extinction trials in an average of 28.2 sessions, ranging
from 19 sessions to 38 sessions. The most and second most
variability in the number of sessions to reach a criterion
for a given block size was found for block sizes of ten and
five, respectively. It took an average of 13.2 sessions to
reach the criterion performance of a discrimination ratio
for two successive sessions greater than 0.80 with a block
size of ten. It took an average of 8.7 sessions to reach
criterion with a block size of five. In the remaining block
sizes (three, one, and one with 8+ Extinction trials), the
average number of sessions was less than 3 which indicates
that the pigeons had learned the discrimination by the time
they had reached the criterion with a block size of five
which occurred within an average of 21.8 sessions. Figure 2
shows the stimulus generalization gradients. During the

positive stimulus generalization test, the mean rate of

Averege Rete of Reeponee

 

35

—E—- Positive Gradient

 

 

 

5° ' -—e— Negative Gradient
D\"\
40 . \
.
3o ..
i
20 .
i
10 -
-i / /\
° 1 i i i !
Training Proximity: High High Low Low
Separation: Small Large Small Large
Trlel Type

Figure 2: Generalization gradients

36
responding to the high proximity, small separation probes
was 44.8 responses per session. The mean rate of responding
to the low proximity, small separation probes was 41.9
responses per session. The mean rate of responding to the
high proximity, large separation probes was 42.7 responses
per session. The mean rate of responding to the low
proximity, large separation probes was 37.8 responses per
session. It can be seen that the rates of responding
declined across the high vs. low proximity conditions, and
across the small vs. large separation conditions in the
positive generalization test. It can be seen that the rates
of responding during the negative stimulus generalization
test remained under 6 responses per session for all negative
probes. The data from the negative generalization test will
not be considered any further as there appears to be an
obvious floor effect. The only interesting facet of this is
noting that the discrimination between the positive and
negative stimuli was not degraded under the probe conditions
in which the training stimuli were altered. The rates of
responding during testing for the individual subjects across
all conditions may be found in appendix C.

A within subjects 2 X 2 analysis of variance with three
repeated measures was performed on the two levels of
proximity, and the two amounts of separation for the first
three sessions of probe testing. No effects were
significant: Proximity (F(2,12)=2.36, p>0.1), separation

(F(1,6)=1.27, p>0.1), trials (F(2,12)<1.0), proximity by

37
separation (F(1,6)<1.0), proximity by trial (F(2,12)<1.0),
separation by trial (F(2,12)=1.79, p>0.1), and proximity by
separation by trial (F(2,12)<1.0).

A within subjects one way analysis of variance on the
rates of responding to the positive stimulus, and the four
types of positive probe trials on the first day of probe
testing was performed. No significant effect was found
(F(4,24)=2.55, p>0.05) for type of trial, although this
analysis did approach significance (p=0.065). Subsequent
linear contrasts also failed to find any significant
differences. The contrast between responding to the
positive stimulus and to all of the probes was not
significant (Fcontrast(1, 24)=4.39, a priori p<0.05, post
hoc (Scheffe’s test), p>0.05). The contrast between the two
levels of proximity within the probe conditions was not
significant (Fcontrast(1, 24)=2.09, p>0.05). The contrast
between the positive stimulus and the probes having high
proximity was not significant (F(1, 24)=1.75, p>0.05). The
contrast between the positive stimulus and the probes having
low proximity was not significant (F(1,24)=7.31, p<0.05,
post hoc (Scheffe’s test), p>0.05). As indicated, two of
the contrasts would have been significant had they been
planned a priori contrasts. Differences approaching
significance exist between the positive stimulus and all of
the probe conditions and also between the positive stimulus
and the low proximity probe conditions.

A closer examinatin of the data indicated that one

38
subject was responding approximately between two-thirds and
three-quarters as much as the rest of the subjects to the
positive stimulus. This subject was not run under my direct
supervision, however its lower rate of responding to the
positive stimulus is cause for some concern about its
overall performance. Analysis of the data without this
subject does result in significant differences being
detected.

The main effect for level of proximity was significant
(F(1,5) = 7.42, p<=0.05). No other effect or interaction
was significant: Trials (F(2,10)=1.01, p>0.1), Separation
(F(1,5)<1.0), Trial X Proximity (F(2,10)<1.0), Trial X
Separation F(2,10)=2.24, p>0.1), Proximity X Separation
(F(1,5)<1.0), and Trial X Proximity X Separation
(F(2,10)<1.0).

Response latencies to the different probe types were
also recorded and analyzed. Figure 3 presents the median
response latency to each type of probe trial. As can be
seen, there are only slight differences in median response
latencies. A Friedman analysis of variance by ranks was
used as latency data frequently does not conform to
assumptions of normality required for performing standard
analysis of variance statistical methods. This analysis was
performed on the latency to respond on the first probe trial
for each type of probe across all subjects. The analysis
indicated that there were no significant differences in

response latencies to the different stimuli (X3 = 5.25,

Latency In mlllleeconde

39

 

1500
wool
1300‘
1200‘
1100"1

1000‘

500-4
400«
3004
200..

100*

 

 

 

Figure 3:

Condition

Median response latencies

 

p>0.05). The response latency data, and the response time
distribution data are provided in appendix C for the

interested reader.

Experiment Two

The results of the first experiment conformed fairly
well to the expectation that pigeons use features to
organize features of objects into hierarchical parts. The
pigeons responded more to the group of parts which had high
proximity values. There was no effect of the amount of
separation on the amount of responding which may just
indicate that the difference between the amounts of
separation were not sufficient to produce any generalization
decrement that would denote a loss of the gestalt of the
complete form. Unfortunately, this experiment was not able
to conclusively implicate proximity as the organizational
principle being used.

The second experiment was designed to account for
different possible organizational principles which may be
controlling the subject’s responding. In the initial
experiment, the results indicate that some organizational
principle appears to be being used by the pigeons to
hierarchically group features into parts of an object. The
principle of organization which held our interest was

proximity. However, the principle of closure could have

40

41
accounted for the results as well since the high proximity
stimulus had the only part which was closed. So, for this
experiment, a new high proximity stimulus was constructed
which did not have any closed parts. The probe stimuli used
in this experiment are presented in figure 4.

It should be noted that in this experiment the stimuli
were single parts of the object which represent a major
reduction in the amount of total contour of the training
objects presented during any given probe trial. The
rationale for using single parts was to refine and restrict
the comparison to the within part proximity rating without
additional higher level between part relations having any
part in the results. The calculation of individual part
proximity ratings proceeded via the same methods used in

experiment one.

Method

Subjects
The same pigeons were used in this experiment as in the
previous experiment. They were maintained in the same

manner as in the initial experiment.

Apparatus
The same equipment was used in this experiment as in
the previous experiment. The computer program PRII.05b was

also used to run this experiment.

42

High Proximity~ Low Proximity
Part B Part D
0.94 -0.17

Part G Part E
1.04 0.04

Figure 4: Stimuli used in experiment two

43

Procedure

The procedure was identical to the initial experiment,
except that the subjects were given ten sessions of
re-training on the initial discrimination task during which
the number of 8+ Extinction trials were increased from five
to ten per session. The subjects were then probe tested for
five sessions using the probe stimuli presented in figure 4
with their associated proximity rating values. Trial
durations and contingencies were identical to those in
experiment one for 8+, S-, S+ Ext, and Probe trial types.
Trial order was also randomly determined for each session as
it was in the previous experiment. However, only a positive
stimulus generalization gradient was sought. No individual
parts which corresponded to the negative orientation were

used in probe testing.

Results

The initial discrimination ratios did not fall below
0.80 despite the change in the relative frequency of
unreinforced positive trials. The response rate for the
positive stimulus remained high throughout, averaging
between 21 pecks per trial for the least responsive subject
to 31 pecks per trial for the most responsive bird. The
rates of responding to probe trials represented very large
decrements in responding relative to the rates of responding

to the positive stimulus with two pigeons being practically

44
totally non-responsive. The effects of this near total lack
of responding to probe trials is evident in figure 5 which
presents the mean rates of responding to the positive
stimulus, and each part probe. The complete response rate
data may be found in appendix D.

A within subjects 4 way analysis of variance with three
repeated measures was performed on the four parts for the
first three sessions of probe testing. No significant main
effects or interaction effects were found. The main effect
of part was not significant (F(15,3)=1.96, p>.1). There
were no significant effects across trials (F(10,2)=1.52,
p>.1). The interaction between parts and trials was not
significant (F(30,6)=2.03, p>.05).

The response time data was not analyzed as this data
was not helpful in experiment one where there was a
significant amount of responding to the probe trials, and
significant differences in rates of responding to different
conditions. To say that the response time data is not
likely to provide any insights in this instance is supported
by the size of the decrement of responding across all
conditions. However, for the interested reader, the

response time data is provided in appendix D.

Averege reeponeee per eeeelon

45

 

10

 

 

1.04 0.94 Pert 0.04 -0.17

Figure 5: Mean rates ot responding to parts

 

Discussion

The primary conclusion which can be drawn from this
experiment is that pigeons’ perceptual systems may use some
principle of perceptual organization, either proximity or
closure, to group features of objects into parts. This
conclusion remains tentative as there were no significant
differences in response rate to probe stimuli which
consisted of parts with high proximity values compared with
the rate of responding to parts having lower proximity
values during the positive generalization test. The data
appear to be in the correct direction, however the methods
used may not have been sensitive enough to detect a
difference in a higher perceptual process. Furthermore, the
power of the test remained relatively low -- a larger sample
size would have been more desirable.

A template matching theory would not predict this
result, and cannot readily account for it without
incorporating feature analytic processes in the service of
constructing the ultimate high level representation to be
used as a template. There were no a priori biases created
by the selection of the initial stimuli which were used in
the discrimination training as these stimuli were not
distinguishable on the basis of the proximity of the

features of which they were composed. The testing was

46

47
performed under steady state conditions as there was no
significant effect across trials despite the probe stimuli
being presented under conditions of extinction. Thus, one
means of grouping features which has been found to be
significant in human perception of line drawings (Palmer,
1977) may also be used by pigeons.

This experiment is somewhat similar to Palmer’s (1977)
third experiment which was a part verification task. This
task involved asking a subject to make a yes/no judgement
about a part being present in a given figure. Parts which
had high, medium, and low goodness based on the proximity
dimension were tested. Palmer (1977) found that the
reaction times were significantly faster for parts with
higher proximity ratings. This led Palmer to reject
template matching theories. Palmer also indicated that this
result could be dealt with by the use of a hierarchical
perceptual processing model in which features were grouped
into structural units which could be combined to generate
higher level structural units. Recognition in this process
would occur in a top down fashion from the highest level
structural unit to lower level structural units. Parts with
high proximity will more likely be coded as higher level
structural units than parts with lower proximity ratings.

At this point in time, there is a growing body of
evidence that the visual perceptual systems of pigeons use
features as a means of developing high level representations

of objects. This evidence comes from studies by Rilling,

48
DeMarse, and LaClaire (in press), Blough (in press), and by
Wasserman, DeVolder, van Hamme, and Biederman (1990). The
Rilling, DeMarse, and LaClaire (in preparation) work led to
the conclusion that midsegments are weighted more heavily as
features than vertices. Wasserman, DeVolder, van Hamme, and
Biederman (1990) studied vertex and midsegments using more
complex figures than Rilling, DeMarse, and LaClaire, but
came to the same conclusion. Blough’s (in press) conclusion
is that higher level relations of features, like symmetry,
are critical to pigeons performance in target detection
tasks. Indeed, the present investigation suggests that the
higher level relation between features which is defined as
the relative degree of proximity of a part given the figure
containing that part may be important for pigeons in the
process of discriminating between visual forms.

Feature theories of animal object recognition have not
yet been formalized as the initial data base for generating
such a model has not been completely formed. It is only in
very recent work that low level features have been
tentatively identified using two-dimensional line drawings.
There may be a different set of elementary features
available in actual three dimensional objects as suggested
by Biederman’s (1987) recognition-by-components theory of
human object recognition in which he argues for the use of a
three-dimensional primitive called a geon. Geons are
defined as a particular set of generalized cones. In order

to provide an adequate test of this notion either

49
three-dimensional objects need to be used, or strong
evidence for animals being able to detect the
three-dimensional structure of two-dimensional drawings
needs to be provided.

Furthermore, it would be premature to make any
conclusions about the precise nature and dimensions used in
generating higher level representations based on these low
level features. There are a number of promising means by
which pigeons may organize features like symmetry,
proximity, and connectedness. But, other means of
organizing features need to be investigated like complexity,
axes of elongation, non-accidental properties, and
redundancy. Ultimately, whatever organizing principles are
identified as being best able to account for the results of
further empirical work will need to be integrated into an
account of the stages of the feature analytic process which
is assumed to result in a unitary representation of the
object or scene.

Furthermore, the issue of the frame of reference used
in the process of object recognition has not been directly
investigated. Template theories like Heinemann and Chases’
(1990) clearly prefer the use of viewer-centered
coordinates. Feature models tend to be more amenable to the
use of object-centered coordinates. The main advantage of
the object-centered coordinate system is in terms of an
economy of storage as viewer-centered coordinates require

multiple representations for an object as the viewer’s

50
spatial relationship with the object changes. While there
is some evidence (Vaughn & Greene, 1984) that pigeons may be
capable of impressive feats of storage, there is no reason
to assume that either the process underlying those feats is
memory intensive or that this is ecologically valid.
Furthermore, evidence from sub-categorization experiments,
and pseudo-categorization control groups (Wasserman,
Kiedinger, & Bhatt, 1988) clearly implicates the existence
of some metric of similarity influencing the pigeon’s memory
for category membership. It is highly likely that both
object and viewer centered coordinate systems are available
for use in the perceptual systems of animals as both types
of information are important for functioning in the real
world. However, template matching theories have difficulty
in dealing with the development of object-centered
representations. Whereas feature analytic theories are
capable of providing both types of coordinate systems.

In conclusion, the present research is an important
step forward in the pursuit of a feature analysis model for
object recognition in animals. The main result of the
experimental work is only suggestive of the pigeons use of
high level organizing principles in object recognition
processes. This work fits well with Blough’s (in press)
evaluation of the process of object recognition in pigeons.
It extends our knowledge of the processes whereby higher
level representations are generated which Heinemann and

Chases’ (1990) template matching theory leaves largely

51
undescribed. Furthermore, theoretical arguments and other
indirect evidence against the use of a strict template
matching theory have been presented. But, a formal model of
object recognition in pigeons based on feature analysis is
not presented as more research is needed in order to make a
strong proposal regarding the exact nature of this process.
Suffice it to say that feature analysis has been implicated
as being the process underlying pigeons visual object

recognition capabilities.

LIST OF REFERENCES

References

Bhatt, R. S., Wasserman, E. A., Reynolds, W. F., & Knauss,
K. S. (1988). Conceptual behavior in pigeons:
Categorization of both familiar and novel examples from
four classes of natural and artificial stimuli.

Jggrnai of Experimental Psychology: Animal Behavior
Processes, i3, 219-234.

Biederman, I. (1987). Recognition-by-components: A theory
of human image understanding. Ps cholo ical Review,
94, 115-147.

Biederman, I. (1990). Higher-level vision. In Osherson, D.
N., Kosslyn, S. M., & Hollerbach, J. M. (Eds). Ap

Invitation to Cognitive Science: Visual Cognition And
Action. Vol. 2, Cambridge, MA: The MIT Press.

Bloch, S., Lemeigan, M., & Martinoya, C. (1987). Coordinated
vergence for frontal fixation, but independent eye
movements for lateral viewing, in the pigeon. In
O’Regan, J. K., & Levy-Schoen, A. (Eds). Eye Movements:

From Physiology to Cognition. Holland: Elsevier

Science Publishers.

Blough, D. S. (1955). Method for tracing dark adaptation in
the pigeon. Science, 121, 703-704.

Blough, D. S. (1958) A method for obtaining psychophysical
thresholds from the pigeon. Journal of the

Experimental Analysis of Behavior, i, 31-43.

Blough, D. S. (1979). Effects of the number and form of
stimuli on visual search in the pigeon. Journal of

Experimeptal Psychology: Animgi Behavior Processes, 5,
211-223.

Blough, D. S. (1985). Discrimination of letters and random
dot patterns by pigeons and humans. Journal of

Experimental Psychology: Animal Behavior Processes,
ii, 261-280.

Blough, D. S. (In Press). Features of forms in pigeon
perception,

53

54

Blough, P. M. (1991). Selective attention and search images

in pigeons. Journal of Experimentai Psychology:
Animal Behavior Precesses, i1, 292-298.

Boakes, R. (1984) From Qarwin To Behaviourism. London:

Cambridge University Press.

Cerella, J. (1990a). Shape constancy in the pigeon: The
perspective transformations decomposed. In Commons, M.
L., Herrnstein, R. J., Kosslyn, S. M., & Mumford, D. B.

(Eds). Quantitative Analyses of Behavior:
Compuperional and Clinical Approaches to Pattern
Reeggnition and Concept Formation. Volume IX. New

Jersey: Lawrence Erlbaum Associates.

 

 

Cerella, J. (1990b). Pigeon pattern perception: Limits on
perspective invariance. Perceprion, i2, 145-159.

Chase, 8., & Heinemann, E. G. (1989). Effects of stimulus
complexity on identification and categorization. The

Internarigna; Journal of Comparative Psychology, A,

165-181.

Domjan, M. (1987). Animal learning comes of age. American

Psychoiogisr, 5;, 556-564.

Domjan, M. & Galef, B. G. (1983). Biological constraints on
instrumental and classical conditioning: Retrospect

and prospect. Animel Learning & Beneyigr, ii, 151-161.

Garcia, J. (1981). Tilting at the paper mills of academe.
Anericen Psychologist, er, 149-158.

Gibson, J. J. (1979). The Ecologieal Approach To Visual
Perception. Boston: Houghton-Mifflin.

Goodale, M. (1983). Visually guided pecking in the pigeon

(Columba livia). Brain Behavior end Evolution, 2;,
22-41.

Heinemann, E. G., & Chase, S. (1990). A quantitative model
for pattern recognition. In Commons, M. L.,
Herrnstein, R. J., Kosslyn, S. M., & Mumford, D. B.

(Eds). Qnantitative Anelyses of Behavior:
ngputational and Clinical Approaches to Pattern
Reeognition and Concept Formation, Volume I . New

Jersey: Lawrence Erlbaum Associates.

 

55

Herrnstein, R. J. (1984). Objects, categories, and
discriminative stimuli. In Roitblat, H. L., Bever, T.
G., & Terrace, H. S. (Eds). Aninei Cognition. New

Jersey: Lawrence Erlbaum Associates.

Herrnstein, R. J., Loveland, D. H., Cable, C. (1976).

Natural concepts in pigeons. Jgurnal of Experimental
Eeyehoiogy: Animai Behavior Processes, ;, 285-302.

Honig, W. K., & Urcuioli, P. J. (1981). The legacy of
Guttman & Kalish (1956): Twenty-five years of research
on stimulus generalization. Journal of the

Experimental Analysis of Behavior, gg, 405-445.

Macphail, E. C. (1987) The comparative psychology of
intelligence. Behavioral and Brain Sciences, i9,
645-695.

Martin, G. R., a Young, S. R. (1983). The retinal binocular
field of the pigeon (Columba livia: English racing
homer). Vision Research, x3, 911-915.

Marr, D., & Nishihara, H. K. (1978). Representation and
recognition of the spatial organization of
three-dimensional shapes. Proceedings of the Roval

Soeiety of London (Series B), 200, 269—294.

Maturana, H. R., & Frenk, S. (1963). Directional movement
and horizontal edge detectors in the pigeon retina.

Seience, i42, 977-979.

Nye, P. W. (1973). On the functional differences between
frontal and lateral visual fields of the pigeon.

yisign Research, re, 559-574.

Palmer, 8. E. (1977). Hierarchical structure in perceptual

representation. Qggnitive Psycnoiogy, 2, 441-474.

Quinlan, P. T. (1991). Differing approaches to
two-dimensional shape recognition. Psychological
Bulietin, igg, 224-241.

Rilling, M. E. (In press). An ecological approach to
stimulus control and tracking. In Honig, W. K., &

Fetterman, G. (Eds.) Cognitive Aspects of Stimulus
gonrrgi. New Jersey: Lawrence Erlbaum.

Rilling, M. E., DeMarse, T. & LaClaire, T. L. (In

press) Accepted by The British Journal of
Experimenral Beychology.

56

Rilling, M. E., & LaClaire, T. L. (1989). Visually guided
catching and tracking skills in pigeons: A preliminary

analysis. Journal of rhe Experimental Analysis of
Benavior, 52, 377-385.

Rilling, M. E., & Neiworth, J. J. (1987). A method for

studying imagery in animals. Journal of Experimental
Esyenglogy; Animal Behavior Processes, ll, 203-214.

Roitblat, H. L., Bever, T. G., & Terrace, H. S. (1984).
Animal Cognition. New Jersey: Lawrence Erlbaum
Associates.

Seligman, M. E. P. (1970). On the generality of the laws
of learning. Psychological Review, 11, 406-418.

Shepard, R. N. (1984). Ecological constraints on internal
representation: Resonant kinematics of perceiving,
imagining, thinking, and dreaming. Psychological
Reyiew, gl, 417-447.

Shettleworth, S. J. (1983). Function and mechanism in
learning. In Zeiler, M. D., & Harzem, P. (Eds.)
s 'n a sis o Beh v'o vol. 3, (pp. 1-39).
New York: John Wiley & Sons Ltd.

Stebbins, W. C. (1990). Perception in animal behavior. In
Stebbins, W. C., & Berkley, M. A. (Eds.) Comparative

Perceprion: Complex Signals. vol. 1, (pp. 1-26). New
York: John Wiley & Sons Ltd.

Terrace, H. S. (1984). Animal cognition. In Roitblat, H.
L., Bever, T. G., & Terrace, H. S. (Eds). Animal
Cognition. New Jersey: Lawrence Erlbaum Associates.

Vaughn, W. Jr., & Greene, S. L. (1984). Pigeon visual

memory capacity. Journal of Experimental Psychology,
lg, 256-271.

Vitz, P. C., & Todd, T. C. (1971). A model of the
perception of simple geometric figures. Psychological
Review, 18, 207-228.

Wasserman, E. A. (1991, April). Paper presented at the
meeting of the Midwestern Psychological Association,
Chicago, IL.

57

Wasserman, E. A., DeVolder, C. L., Van Hamme, L. J., &
Biederman, I. C. (1990, November). Recognition by
components: Comparative evaluations of visual
discriminations by pigeons. Paper presented at the
meeting of the Psychonomic Society, New Orleans, LA.

Wasserman, E. A., Kiedinger, R. E., & Bhatt, R. S. (1988).
Conceptual behavior in pigeons: Categories,
subcategories, and pseudocategories. Journal of

Experimental Psyehology: Animal Behavior Processes,
15, 235-246.

APPENDICES

APPENDIX A

APPENDIX A
Instructions for running PRII.05b.obj

1. The first screen -- Subject/Time/Place data. This screen
is fairly self-explanatory. Note: Never enter the same
session number twice, and keep session numbers under 100.

2. The second screen -- Saving your data. This screen shows
you which disk it will save your data on and asks you for a
title. Since you’ll probably want to save your data on a
data disk in the external drive, you may need to click on
the "drive" button. Always include the date and your bird’s
number when giving the data a title. For example: "A1137
4/11/91 test" is a proper data file title which indicates
the subject’s number (A1137), the date of the session
(4/11/91), and that this was a test session.

3. The third screen -- Activating the touch screen. What the
computer needs at this point are the coordinates that define
the active touchscreen. These can vary depending on your
purpose, but default coordinates are 0, 70, 0, and 60. If
you do not know how much of the touchscreen to activate,
enter these coordinates in the order given and press Return
after each number. Coordinates for each experiment being run
should be available on the ’experiment info sheet’ which can
be found in the green binder.

4. The fourth screen -- Trial types and block size. Next,
the computer will ask you how many "trial types" you want to
run. What this means is that the computer needs to know how
many different contingencies you want to run. Each picture-
contingency pairing is one trial type. For example, during
discrimination training, you are running two different
contingencies: An S+ that is reinforced and an S- that
isn’t (trial types = 2). If you add a probe to the previous
example, then you have three contingencies: an S+ that is
reinforced, an S- that isn’t, and a probe that isn’t (trial
types = 3). The picture which is being used as the S+
stimulus may also be used for the non-reinforced probe trial
in this example (trial type still = 3) -- this is an example
of an S+/extinction trial introduced to stabilize
performance during probe testing. Also, if more stimuli
(pictures) are to be introduced which all have the same
contingency (8+, S-, or Probe), each one is a unique trial
type. So, if we had two different probe pictures (let’s call
them picture A and picture B) which we wanted to run during
a testing session which also had one S+ picture, one S-
picture, and the S+ picture as a probe (an S+/extinction
trial), then this session would have five trial types: 1.
the 8+, 2. the S-, 3. the S+/extinction (using the S+
picture declared as a probe type), 4. picture A declared as
a probe type, and 5. picture 8 declared as a probe type.

60

61

After you enter the number of trial types, the computer
asks for "block size." Block size determines how many times
in a row the same stimulus will be presented. The larger the
block, the greater the ease of learning. Remember that the
block size and the total number of trials are related -- for
example you cannot have a block of size 8 with 50 total
trials because 50 is not evenly divisible by 8. The total
number of trials must be evenly divisible by the block size.

5. The fifth screen -- Setting up the pictures. The
following screens repeat over and over until the number of
trial types that you requested in step four have been
specified. What happens in this loop is that you actually
select the contingency and the picture to go along with it.

a. Type of stimulus: Read the screen; it’s self-
explanatory. It’s a good idea to always enter your stimuli
in the same order so you won’t make mistakes.

b. Milliseconds: 1,000 milliseconds = 1 second. Enter
the amount of time you want the stimulus to stay on the
screen in milliseconds.

c. Choosing the picture(s) -- this screen is similar to
the "saving data" screen. It lets you view the contents of
both disks to find where your pictures are. Again, you can
change the active disk drive by clicking on the "drive"
button. When you find the picture(s) you are going to use,
double click on their titles to activate them. The computer
only acknowledges one picture at a time, so if you are using
moving stimuli, choose the pictures in the correct order.
After you are done choosing the picture(s) for the trial
type you are working on, click on the "cancel" button.

d. The last piece of information needed in this loop is
how many trials of the stimulus that you just chose are
needed to appear throughout the entire session. After you
enter this number, the computer will return to step (a) and
ask for all of the same information on the next trial type.
When all of the trial types have been specified, the loop
ends.

6. The sixth screen -- Randomized trials. The computer now
presents you with the randomly generated trials. They whiz
by pretty fast, but don’t worry. Glance at them to make sure
they look random. Enter any number to continue.

7. The seventh screen -- Final check. The computer finally
presents you with a summary of how the program will run.
Look carefully at what is listed: make sure the
contingencies, pictures, and number of trials are correct.
If everything looks okay, enter any number to begin the
experiment.

SOURCE CODE FOR PRII.05B
Program cycling;

(* This is a program designed to utilize the graphics
capabilities of the MacIntosh, to present stimuli, touch
sensitive technologies for recording responses, and the
BenchtopTM instrument for providing a real world interface
for doing discrimination and generalization research. *)

(* Compiler Instructions *)
Uses __Quickdraw, __Graphutils, __SaneIO, __Tooltraps,
(*$U+*) uQuickdraw, uPackIntf:

(* Linker Instructions *)

Link __Benchlib, __Touchlib.51290, __Quickdraw,
__Graphutils, __IO, __OSTraps, __Scrollwind, __Sane,
__Uniform, __ExtendIO, __Extras, __Stringlib,
__SFNames, __NoSysCall, __PackTraps : PRII.05b.obj:

(* Declaration of Globally Defined Constants *)
Const

(* Trial Information Constants *)

sPlus = 0:
sMinus = 1;
probe = 2;

moving = 18;
stationary = OB:

(* Touch Screen Constants *)
Enterpoint = 37:

tracking = 39:

exitpoint = 40;

addexit = 1;

subexit = 5;

reportranson = 1:
reportransoff = 5;
coordmode = 35;

scanmode = 34;

(* Graphics Constants *)
patC = 8: (* Black *)
patB = 11; (* White *)

(* Declaration of Variable Type for this program’s *)

(* specific use. *)
Type
(* Variable type "trialinfo" is used to store *)
(* information about each unique type of trial *)

62

63

(* entered by the user for any given session. *)

trialinfo = record
number:Integer;
type:Integer:
numpicts:Integer;
duration:Longint;
motion:bytei
nameptr:stringptr;
name:Str255;
stimulus:pichandle[101]
wrongstim:pichandle[101]
End:

(* Global Variable List *)

Var rft,ttypes,lasterr,tottrials,blocks,blocktrials:Integer;
tct1,tct2,tct3,hix,lox,hiy,loy:Integer;
rndnoa,brndno:Integer:
current:trialinfo[301]
vref,good,ID,cue,count:Integer;
neg,pos,prob:Integer[500]:
typetot:Integer[25];
order:Integer[500];

Amt:Longint;
numbit,timebit:Longint[400];
destined:byte[400];
Destination:integer[400];
nameptr:ptrL:
rectframe,pfr,rde:rect;
hiddenport,execport:grafptr;
blacker,whiter:Pattern:
LReply:SFReply;

(* List of Externally Defined Functions *)

Function Random():Integer;
EXTERNAL;

Function Black():APattern;
EXTERNAL;

Function White()APattern;
EXTERNAL:

Function CheckIn(first,in,last:Integer):Integer;
EXTERNAL:

Function Evenodd(val:Integer):Integer;
EXTERNAL;

Function FSread(refnum:Integer;counter:Ptr1:buffptr:Ptr):
integer;

64

EXTERNAL:

Function Openpicture(Picframe:rect): Pichandle:
EXTERNAL:

Function Newhandle((size:longint):Handle;

EXTERNAL:

(* A procedure to obtain the amount *)

(* of information in the A-port input *)

(* buffer from the touchscreen. *)

Procedure Aserwaiting(num:*Longint):

{
setserport(0):
serwainting(num);
l:

(* A procedure to remove a block *)
(* of data from the A-port input *)
(* buffer to variable space. *)

Procedure Agetblock (des:PtrB;mun:PtrL):

{
lasterr:=FSRead(-6,mun,des):

l;
(* Initialization Procedure *)
procedure _Init();

var beta,numrand,sessnum,peat:Integer;
bird,dday,weight,sesstime,chamber,repeat,halfnum,
testnum,typing:Integer;
mark:Integer[200]:

{

(* Initialize BenchtopTM Instrument. *)

Wakebench(cue):

(* Initialize the Touch Screen. *)

Wakescreen(reportranson,coordmode,enterpoint,subexit);

clear():

(* Initializes Trigonometric Functions *)

InitTrig()i

Moveto(10,10);

(* Collect information about the active area for the touch

screen. *)

Writestring("Enter the leftmost (lowest) active touch screen
coordinate:"):

readint(@lox):

Writestring("Enter the rightmost (highest) active touch
screen coordinate:")i

readint(@hix):

Writestring("Enter the topmost (lowest) active touch screen

65

coordinate:");

readint(@loy)i

clear()i

moveto (10,10);

(*

Collect information about the number of kinds of trials.

*)

Writestring("Enter the number of trail types you want:
(1-20) ")r

readint(@ttypes):

writeln();

(* Collect Information about the block structure of the

session. *)

Writestring("Enter the size of trialtype blocks desired:
(1 is random) ");

readint(@blocks);

writeln():

(* Begin to initialize variables. *)

tct1:=0;tct2:=0;tct3.=0;

loop(,repeat.=0,++repeat,repeat>=25)

typetot[repeat]:=
loopi:repeat:=0,++repeat,repeat>=200)

mark[repeat]:0 = 0;
loopi:repeat:=0,++repeat,repeat>=70)

{

pos[repeat]:=0:
neg[repeat]:=0;
prob[repeat]:=0:

l:
loop(,repeat:=0,++repeat,repeat>=100)
{
order[repeat]:=
l:
clear()i
tottrials:=0
halfnum:=0:
testnum:=0i
(* End initialization of variables. *)

(**********************************************************)
(* Begin loop to collect information about each kind *)

(* or type of trial to be run during the session. The
(* following information is collected and stored in

*)
*l

(* variable space: The type of trial (positive, negative,*)

(* or probe), the duration of the trial, the stimulus to
(* be used in terms of ’PICT’ file(s) indicating the use
(* of static or moving forms, and the number of times

(* that the trial should be repeated within the session.

*)
*)
*)
*)

66

(**********************************************************)
loop(,repeat:=1,beta:=1,++repeat,repeat>ttypes)
{
clear():
moveto(10,10):
writestring("What type of stimulus do you want this to
be?"):
writeln();
writesting(”-- Enter a 1 for an S+, A 2 for a S-, or a
three for a probe:"):
readint(@typing):
case typing of
1:current[repeat].type:=sPlusi
2:current[repeat].type:=sMinus:
3:current[repeat].type:=probe:
otherwise { writeln():
writestring("O.K. wiseguy, this is
going to be a probe,
like it or not!"):
current[repeat].type:=probe;
i:
END;
clear():
moveto(10,10):
writestring("Enter the desired trial duration in
milliseconds:");
writeln():
readlong(@current[repeat].duration)i
clear():
setpictures(repeat); (* Procedure to obtain
PICT file(s) *)
if current[repeat].numpicts=1 then
{current[repeat].motion:=stationary;
sConCat(current[repeat].name,".Static",
current[repeat].name);
}

else
{current[repeat].motion:=moving;
sConCat(current[repeat].name,".Moving",
current[repeat].name);
}:

case current[repeat].type of
sPlus:sConCat(current[repeat].name,".S+",
current[repeat].name):
sMinus:sConCat(current[repeat].name,".S-",
current[repeat].name);
probe:sConCat(current[repeat.name,".Probe",
current[repeat].name);
END:
clear():
moveto(10,10)i
writestring("How many trials of this stimulus would you
like? (max/type=70)"):

67

readint(@current[repeat]/number):
tottrials+=current[repeat].number;
if tottrials>200 then
writesting(“May not work properly with over 200
trials!");
resettimer():
attime 1000 do:
if current[repeat].number<>o then
halfnum:=current[repeat].number/Z:
loop(current[repeat].number>0,testnum:=1,(++testnum:
++beta),beta>tottrials)

order[beta]:=repeat:
i:
i:

(* End trial type information collection loop. *)

clear();

(* Begin to collect subject, and date information for
session. *)

moveto(10,10):

writestring("Enter Bird’s no."):

readint(@bird):

writeln():

writestring("Enter Bird’s Weightz):

readint(@weight):

writeln():

writestring("Enter Date:(i.e., 327 for Mar 27)"):

readint(@dday):

writeln():

writestring("Enter time:(i.e., 1200 for 12:00)"):

readint(@sesstime):

writeln():

writestring("Enter Session Number: (1 for the first session,

etc.)");

readint(@sessnum):

writeln():

Writestring("Enter the box designation: (1 for box 1a, 2

for box 1c)"):

readint(@chamber):

writeln():

(* Finish collecting subject and date information for the
session. *)

(* Create data file for the session. *)
putfile (@nameptr, @Vref, @good):
if good then
{
fcreate(nameptr," RCMP"+2, "TEXT"+2,vref):
fopen(@ID,nameptr,3,vref):
i:
Amt:=0:

68

(* Store subject and date information in the data file. *)
Basicdata(ID,bird,weight,dday,sesstime,sessnum):
clear():

(* Instruct the BenchtopTM to turn on panel light.*)
turnon(0,56,cue):

blocktrials:=tottrials/blocks:
numrand:=sessnum*(blocktrials*2);

(* Call procedure to obtain random ordering of trials. *)

loop(,repeat:=0,++repeat,repeat>=numrand(),
randomization(););

clear():

numrand:=0:

(* Loop to display the ordering of the trial types. *)
loop(,repeat:=1,++repeat,repeat>tottrials)
{
writeint(order[repeat]):
numrand+=1:
if numrand>=blocks then(numrand.=0;writeln(););
typetot[order[repeat]]+=1:
if order[repeat]<0 then mark[repeat]:=repeat:
l:
(* End loop. *)

writeln():
writestring("Enter any number to continue:"):
readint(@repeat);
clear():
loop(,repeat:=1,++repeat,repeat>ttypes)
{
loop(typetot[repeat]<>current[repeat].number,peat:=1,
++peat, peat>tottrials)

{
If (mark[peat]<>0) then

{
order[mark[peat]]:=repeat;
typetot[repeat]+=1;
i:

If typetot[repeat]=current[repeat].number then
BREAK:

(* Loop to list the session information out to the
screen. *)

loop(,repeat:=1,++repeat,repeat>ttypes)

Writestring("Trial type"):
writeint(repeat):

69

writestring(", called"):
writestring(current[repeat].name);
writestring(", has");
writeint(typetot[repeat]):
writestring("trials scheduled for this session.");
writeln():
l:
(* End Loop. *)

writeln():
writestring("Enter any number to continue:"):
readint(@repeat);
clear():
loop(,repeat:=0,++repeat,repeat>=25)
{
typetot[repeat]:=0:
l:

repeat:=10:

(* Initialize Graphics Ports for use. *)
getport(@execport):

getrect(@pfr);

getrect(@rectframe):
loop(,repeat:=0,,repeat<>0)

{
hiddenport:=newoffport(pfr):
if hiddenport then
{ setport(execport):
moveto(10,10):
Writestring("ERROR, Enter a 0 to try again, 1 to
Break:"):
readint(@repeat):
l

else repeat:=5:

i:
if repeat=1 then reqhalt():
clear():
setport(execport):

(* End Initialization Procedure. *)
i:

(* The following procedure ’shuffles’ the order of the
trials so that the order of the trial types is completely
random. *)

procedure randomization();

var hold,spacer,rndcter:Integer;
{

hold:=0:

spacer:=0:

rndcter:=0:
loop(,,,rndnoa<>brndno)

70

(

loop(,rndnoa:=-1,rndnoa:=random() mod
blocktrials,rndnoa>0):

loop(,brndno:=-1,brndno:=random() mod
blocktrials,brndno>0):

i:

rndnoa:=rndnoa*blocks:
rndnoa:=rndnoa-blocks:

rndnoa:=rndnoa+1;

brndno:=brndno*blocks:
brndno.=brndno-blocks;

brndno:=brndno+1:
loop(,rndcter:=1,++rndcter,rndcter>blocks)

hold.=order[rndnoa]:
spacer:=order[brndno]:
order[brndno]:=hold;
order[rndnoa]:=spacer:
rndnoa+=1:

brndno+=1:

i:

(* The following procedure allows the user to select *)
(* ’PICT’ files from the available files stored on disk *)
(* to use for each trial type. The information stored in *)
(* these files is then moved to active variable space in *)

(* the program for use at any point as required by the *)
(* user. Multiple ’PICT’ files may be selected to *)
(* produce trials which display apparent motion. *)

Procedure Setpictures(homeboy:Integer):
var pvref,nrefn,gd,cter:Integer:
len:Longint;
where:Point:
exts:str255;
pictname,extension:stringptr:

{

moveto(10,10):

writestring("Enter number of pictures for this stimulus:"):
readint(@current[homeboy].numpicts);

clear():

moveto(10,10):

nrefn:=1:

getrect(@rde);

where.v:=70:

where.h:=100;

pictname:=@Lreply.fname;
Current[homeboy].nameptr:=@Current[homeboy].name;
loop(,cter:=1,++cter,cter>current[homeboy].numpicts)

SFgetfile(where.vh"",Nil,2," PICT"+2,nil,@Lreply):

l

71
if lLreply.good then
{

gd:=-1:

return:

i:
pvref:=Lreply.vrefnum;
fopen(@nrefn,pictname,1,pvref):
ferr(@gd):
flength(nrefn,@Len):
fmoveto(nrefn,512L):
Len-=512:
current{homeboy].stimulus[ctr[:=NewHandle(Len);
Hlock(current[homeboy].stimulus[ctr]);
fread(nrefn,current[homeboy].stimulus[ctr]‘,@Len):
Hunlock(current[homeboy].stimulus[ctr]):
fclose(nrefn):
writesting("Picture "):
writeint(cter);
writestring(" of Stimulus ");
writeint(homeboy):
writestring(" is set."):
writeln():

e
I

Copystr(pictname,Current[homeboy].nameptr):

l:
(*

The following procedure is the procedure that

(* produces the display for each trial, collects data,

(* and defines the appropriate consequence for that trial

(* type.

Procedure Picturethat():

var sickrect:Rect;
account,ohno,xres,yres,quo,ct,side,trialstim,
swlzInteger:
waiter,statwait:Longint;

{

account:=0;ohno:=0:xres:=-1:yres:=-1:quo.=1;
trialstim:=order[count]:
setrect(sickrect,100,30,130,60):
AserWaiting(@waiter):
loop(,,--waiter,waiter<=0)

{

Anodwellchar(@ct):
If ct=-1 then BREAK:}:

Resettimer():

ct:=0:
setport(hiddenport):
ct:=1:

swl:=2:
loop(current[trialstim].motion=moving,accoun :=0,++account,

{

rft=1)

clear():

72

drawpicture(current[trialstim].stimulus[ct],rde);
loop(,waiter:=TickCount(),,TickCount()>=waiter+3);
copybits(hiddenportc.portbits,execport*.portbits,
pfr,pfr,srccopy,nil):
Aserwaiting(@numbit[account]):
Timerval(@timebit[account]):
If swl=1 then --ct
else ++ct;
If ct=current[trialstim].numpicts then swl:=1;
If ct=1 then swl:=2:
If timebit[account]>=current[trialstim].duration then

{
if ohno=0 then

{
amt:=numbit[account]:
if (amt<=400) then
lasterr:=FSRead(-6,@amt,@destined)
else
{
fputc(ID,l3):
fputs(ID,"Response Overload, only part of
responses recorded."):
fputc(ID,lB):
amt:=360:
lasterr.=FSRead(-6,@amt,@destined):
amt.=numbit[account]-amt:
amt/=4:
fputi(ID,amt);
fputs(ID"Responses were not recorded,
approximately."):
fputc(ID,lB):
i:
serflush():
xres:=-1:
yres:=-1:
l:
ohno:=1:
Case current[trialstim].type of
splus:(
recording(1,@xres,@yres);
loop(,waiter:=TickCount(),,TickCount()>
waiter+1):
if (xres>=lox) and (xres<=hix) and
(yres>=1ay) and (yres<hiy) then
{
side:=evenodd(count):
if side<=0 then side:=2;
clear():
copybits(hiddenporte.portbits,
execport*.portbits,
pfr,pfr,srccopy,nil):
reinforce(quo,300,side):
rft:=1:

73

}:
i:
sminus,probe:(
clear():
copybits(hiddenport*.portbits,
execport“.portbits,
pfr,pfr,srccopy,nil):
rft:=1;
BREAK:
i:
END:
i:
if current[trialstim].motion=stationary then

drawpicture(current[trialstim].stimulus[1],rde):
copybits(hiddenport*.portbits,execport*.portbits,
pfr,pfr,srccopy,nil):
loop(,waiter:=TickCount(),,TickCount()>+waiter+3);
i:
Resettimer():
loop(current[trialstim].motion=stationary,account:=0:
statwait:=50,++account,)
{
Aserwaiting(@numbit[account]);
Timerval(@timebit[account]):
if timebit[account]>=current[trialstim].duration
then BREAK:
Attime statwait do statwait+=50:
i:
if current[trialstim].motion=stationary the
{
amt:=numbit[account]:
if (amt<400) then lasterr:=
FSRead(-6,@amt<@destined)
else
{
fputc(ID,l3):
fputs(ID,"Response Overload, only
part of responses recorded."):
fputc(ID,l3):
amt:=360;
lasterr:=FSRead(-6,@amt,@destined):
amt:=numbit[account]-amt;
amt/=4:
fputi(ID,amt):
fputs(ID,"Responses were not
recorded, approximately."):
fputc(ID,l3):
i:
serflush();
xres:=-1;
yres:=-1;

l:

l:

(*

74

loop(current[trialstim].motion=stationary,,,rft=1);

{
If current[trialstim].type=splus the

{
recording(1,@xres,@yres):
100p(,waiter.=TickCount(),,TickCount()>
waiter+1):
if (xres>=lox) and(xres<=hix) and (yres>=loy)
and yres<hiy then
{
side:=evenodd(count):
if side<=0 then side.=2;
clear():
copybits(hiddenportc.portbits,execportc.
portbits,pfr,pfr,srccopy,nil):
reinforce(quo,300,side):
rft:=1;
BREAK:

}:
}

else
{
clear():
copybits(hiddenportc.portbits,execportc.
.portbits,pfr,pfr,srccopy,nil);
rft:=l:
BREAK:
i:
i:
Timerval(@waiter);
fputs(ID,"Stimulus duration was");
fputl(ID,waiter):
fputs(ID,"milliseconds.");

The following procedure examines the data *)

(* collected during each trial and filters it and *)
(* stores it in the data file. *)

Procedure figure();
var intr,bone,bold,smooth,lamt,actuary,cuteout,ts:Integer:

{

skip,out:Longint:

cuteout:=0:out:=3000:ts:=order[count]:
Resettimer():

Attime 500 do:

lamt:=amt;

turnon(48,48,cue);
smooth.=5:bone:=1;actuary:=1:
Resettimer():

Attime 500 do hold:=0:
loop(lamt>0,intr:=0++intr,intr>=lamt)

{ destination[intr]:=destined[intr]: }:

75

intr:=0:
loop(lamt>0,actuary:=1,++actuary,intr>lamt)

if numbit[actuary]>0 then
{
loop(,,++intr,intr>numbit[actuary])
{
cuteout:=cuteout+1:
arrivaltime[intr]:=timebit[actuary]:
if cuteout>=8 then
{
cuteout:=0:
writeln():
};

l: .
loop(lamt>0,intr:=0,++intr,intr>lamt)
{

smooth:=0:
loop(,bone:=1,++bone,smooth=1)

If (destination[intr]>=245) or
(destination[intr]<3) then
{
bold:=intr+bone:
destination[intr]:=destination[bold]:
arrivaltime[intr]:=arrivaltime[bold];
destination[bold]:=255:):
If (destination[intr]:=<245) and
(destination[intr]>3) then smooth:=1:
If destination[intr]=300 then smooth:=1:
If bold>amt then smooth:=1:
}:
l:
l:
Attime 500 do cuteout:=0:
fputs(ID,"Trialnumber");
fputi(ID,count);
case current[ts].type of
sminus:(fputs(ID,", S-.");tct1+=1;
splus:(fputs(ID,", S+,");tct2+=1;)
probe:(fputs(ID,", Probe,");tct3+=
end:
Fputs(ID,current[ts]].name):
fputc(ID,l3):
loop(destination[0]<>300,intr:=0,,destination[intr])>=245)
{
if destination[intr]<=0 then BREAK;
if intr>lamt then BREAK:
if destination[intr]=255 then BREAK:

fputi(ID,destination[intr]):
fputc(ID,09):

):
1:)

76

fputL(ID,arrivaltime[intr+1]):

fputc(ID,09):

fputL(ID,arrivaltime[intr]):

fputc(ID,09):

fputL(ID,arrivaltime{intr+1]):

fputc(ID,l3):

int :=intr+2:

)3

Case current[ts].type of

sminus:{if intr<>0 then neg[tct1]:=(intr/2):
fputs(ID,"There were a total of");
fputi(ID,neg[tct1]):
fputs(ID,"responses throughout this trial."):
timerval(@out):
if neg[tct1]>5 then out:=out+(neg[tct1]*1000):

Resettimer():
loop(,,timerval(@skip),skip>=out)
typetot[ts]+=neg[tct1];
l:
splus:{if intr<>0 then pos[tct2]:=(intr/2)+rft
else pos[tct2]:=rft;
fputs(ID,"There were a total of"):
fputi(ID,pos[tct2]);
fputs(ID,"responses throughout this trial."):
typetot[ts]+=pos[tct2];
}:
probe:{if intr<>0 then prob[tct3]:=intr/2:

fputs(ID,"There were a total of"):
fputi(ID,prob[tct3]):
fputs(ID,"responses throughout this trial."):
typetot[ts]+=prob[tct3]:

i:
END:
fputc(ID,l3):
fputs(ID,"Total to this type =");
fputi(ID,typetot[ts]);
fputc(ID,lB):
i:

(* The following procedure resets variable array space *)
(* to allow that space to be reused anew on each new *)
(* trial. *)

Procedure Resetarrays():
var cutting,orct:Integer:
creased:Longint;

{

orct:=order[count]:
loop(,cutting:=0,++cutting,cutting>=400)
{ destination[cutting]:=300:
arrivaltime[cutting]:=-1: }:
loop(,cutting:=0,++cutting,cutting>=500)

77

{ numbit[cutting]:=-1:
timebit[cutting]:=-1: }:
Aserwaiting(@creamed):
loop(,,--creamed,creamed<=0)
{ Anodwellchar(@cutting):
if cutting<0 then BREAK: }:

i:

(* The following procedure is the MAIN body of the *)
(* program. It structures the flow of processing occurring
*) (* when the program is run following the initialization

*) (* sequence.
*)

Procedure Main():
var cutout,slop:Integer:
drect;Rect:

{
cutout:=0:

count:=1:

(* Insure that the array space is ready for use. *)
Resetarrays():

(* Wait 3 seconds to proceed. *)
resettimer():
attime 3000 do;

(* Turn on the houselight. *)
turnon(56,48,cue):

(* Main loop. *)
loop(,count:=1,++count,count>tottrials)

{

(* Reset array space each time through the loop. *)
Resetarrays():

clear():

(* Present a trial. *)

Picturethat():

(* Store the trial’s data. *)
figure():

(* Turn on the houselight. *)
turnon(56,48,cue):

l:

(* Call the Halt() procedure to finish the session. *)
Reqhalt():
i:

(* The following procedure terminates the session. It *)

78

(* has two primary functions: To take care of programming,*)
(* equipment and computer operating needs, and to provide *)
(* session summary data both as stored in a data file and *)

(* as screen output.

Procedure Halt():

var cutter,totpos,totneg,totpro,outs,ocut:Integer;

{

Setport(execport):

Diposeoffport(hiddenport):

whiter:= White()‘:

backpat(whiter):

clear(): outs:=0:totpro:=0;totpos:=0:totneg:=0:
tct1:=0;tct2:=0;tct3:=0;

Fputc(ID,13):

prtS(ID,"*************** SUMARY ***************");

fputc(ID,l3):
fputs(ID,"Trial:"):
fputc(ID,09):
fputs(ID,"variety")
fputc(ID,09)
fputs(ID,"Responses"):
fputc(ID,l3):
loop(,cutter:=1,++cutter,cutter>tottrials)
{
fputs(ID," "):
fputc(ID,09):
fputi(ID,cutter):
fputs(ID,","):
fputc(ID,09):
ocut:=order[cutter]:
fputs(ID,current[ocut].name):
fputc(ID,09):
Case current[ocut].type of
sminus:{ tct1+=1:
fputs(ID," S-, "):
fputc(ID,09):
fputi(ID,neg[tct1]):
fputc(ID,l3):
totneg:=totneg+neg[tct1]:
}:
splus:( tct2+=1:
fputs(ID," S+ "):
fputc(ID,09):
fputi(ID,pos[tct2]):
fputc(ID,l3):
totpos:=totpos+pos[tct2]:
l:
probe:( tct3+=1:
fputs(ID," Probe, ")

*)

79

fputc(ID,09):
fputi(ID,prob[tct3]):
fputc(ID,l3):
totpro.=totpro+prob[tct3]:

END:

}:

fputc(ID,l3):

fputs(ID,"There were"):

fputi(ID,totpos):

fputs(ID,totpos):

fputsI(ID," responses during S+.")

fputc(ID,l3):

fputs(ID,"There were"):

fputi(ID,totneg):

fputs(ID,"responses during S-,"):

fputc(ID,lB):

fputs(ID,"There were"):

fputi(ID,totpro):

fputs(ID,"responses during Probes.");

fputc(ID,l3):

loop(,cutter:=1,++cutter,cutter>ttypes)
{
fputs(ID,"There were"):
fputi(ID,typetot[cutter]):
fputs(ID,"responses during"):
fputs(ID,current[cutter].name):
fputc(ID,lB):
}:

fputc(ID,l3):

Moveto(10,10):

writesting("S+ Responses:")

writeint(totpos):

writeln():

writestring("s- Responses:"):

writeint(totneg):

writeln():

writestring(:P Responses:"):

writeint(totpro):

writeln():

loop(,cutter:=1,++cutter,cutter,cutter>ttypes)
{
Writestring("There were"):
writeint(typetot[cutter]):
writestring("responses during"):
Writestring(current[cutter].name):
writeln():

l:

(* Call to terminate the BenchtopTM operations. *)
Bigturnoff(cue):

(* Call to terminate the Touch Screen operations. *)

80
Aputchar(67):

(* Disconnect serial port communications. *)
closeserport(1);

(* Close the data file. *)
fclose(ID):

l:

(* End of Program. *)

APPENDIX B

APPENDIX B

Clarification of Palmer’s Model and Its Application

Proximity was defined as the euclidean distance between

the midpoints of the segments as generated by the equation:
=/(xl-x2)2-+ (yl-y2)2. Where D stands for distance, and
(x1, yl) and (x2,y2) represent the cartesian coordinates of
the midpoints of any two different segments in the figure.
Thus, the distance between the leftmost vertical line (line
1) and the uppermost horizontal line (line 2) may be
calculated from the knowledge of the locations of their
midpoints. Line 1’s midpoint is at x=0, y=1.9 or (0, 1.9).
Line 2’s midpoint is at x=0.9, y=2.5 or (0.9, 2.5).
Inserting these values into the equation we obtain
D=J(0-0.9)2+(1.9-2.5)2. This reduces to D=./(-0.9)2+(-0.6)2 =
J0.81+0.36 = J1.17 = 1.08. In order to obtain the scale
value for the proximity dimension in Palmer’s model, we need
only multiply this distance by a —1.00, obtaining
R(1,2)=-1.08. This process is carried out for all pairs of
line segments in the figure. You may refer to Table B-1 for
a listing of all of the scale values, R(i,j), of all of the
pairs of segments, (i,j), in the figure used in this

research.

82

Figure

83

Part A

Scale values for the proximity dimension between the above tabled

segments:
1

1

2 -1 .08
3 -1 .80
4 -1 .08
5 ~2.70
6 -2.98
7 -1 .54
8 -3.83
9 -3.30

2

-1 .08

-1 .08

-1 .20

-1.91

-2.58

-‘| .85

-3.30

~3.08

3 4
-1 .80 -1 .08
-1 .08 -1 .20
-1 .08
-1 .08
-O.90 -1 .88
-1 .54 -1.91
—1 .54 ~0.65
-2.22 -2.79
~2.10 -2.22
Table B-1:

5 6
-2.70 -2.98
-1 .91 -2.58
-0.90 -1 .54
-1 .88 -1 .91

-1 .20
-1.20
-2.16 -1 .80
-1.54 0.90
-1 .85 -0.65

Proximity relations

7
-1 .54

-1 .85

-1 .54

-0.65

~2.1 6

-1 .80

#270

-1.91

8
-3.83

-3.30

-2.22

-2.79

-1 .54

-0.90

-2.70

9
-3.30

-3.08

-2.10

~2.22

-1 .85

-O.65

-1.91

-1 .08

84

The next step is fairly complicated. It involves
finding the summed scale value for pairs of segments within
a part, R(i,j), and the summed scale value for pairs of
segments between the part and the remainder of the figure,
R(i,k). Where i and j represent different pairs of segments
in the part and k represents segments not in the part. Let
us figure out these values for Part A. We already know the
scale value of R(1,2)=-1.08. By looking in Table B-1 we
find that the other within part scale values for proximity
are: R(l,3)=-l.80, R(l,4)=-1.08, R(2,l)=-l.08,
R(2,3)=-l.08, R(2,4)=-l.20, R(3,1)=-l.80, R(3,2)=-1.08,
R(3,4)=-1.08, R(4,l)=-l.08, R(4,2)=-1.20, and R(4,3)=-l.08.
The sum of all of the within part scale values is R(i,j)t =
-14.64. Where R(i,j)t stands for the grand sum of all
within part scale values. We can also find all of the
between part scale values by reference to Table B-1:
R(1,5)=-2.7, R(l,6)=-2.98, R(1,7)=-l.54, R(l,8)=-3.83,
R(l,9)=-3.30, R(2,5)=-1.9l, R(2,6)=-2.58, R(2,7)=-l.85,
R(2,8)=-3.30, R(2,9)=-3.08, R(3,5)=-0.90, R(3,6)=-l.54,
R(3,7)=-1.54, R(3,8)=-2.22, R(3,9)=-2.10, R(4,5)=-l.88,
R(4,6)=-1.9l, R(4,7)=-0.65, R(4,8)=-2.79, and R(4,9)=-2.22.
Thus the grand sum of the between part scale values is
R(i,k), = -44.82.

Then R(i,j)t is multiplied by the number of segments in
the figure, f, minus the number of segments in the part, p
(or f-p). Giving the equation (f-p) * R(i,j)t. This gives

the number of between part comparisons which will be made.

85
There are a total of 9 segments in the figure and 4 segments
in part A, or f=9 and p=4 which gives f-p = 9-4 = 5. Thus,
there would be 5 between part comparisons for each within
part scale value. Giving the result of (f-p) * R(i,j)t =
5*-14.64 = -73.2. R(i,k)t is then multiplied by p minus one
which represents the number of times a within part scale
value would be compared with each between part scale value.
Giving the equation (p-l) * R(i,k)t. There are 4 segments
in part A. So, we would obtain (p-l) * R(i,k)t = 3 * -44.82
= -134.46. We then subtract the between part value from the
within part value. Giving us ((f-p) * R(i,j)t} - {(p-l) *
R(i,k)t) = -73.2 - (-134.46), or 134.46 - 73.2 = 61.26.

This result must then be divided by what amounts to the
degrees of freedom remaining when a part having a given
number of segments is used. This can be calculated by the
equation p * (p-l) * (f-p). For a part with 4 segments taken
from a figure with 9 segments, this equation gives 4 * (4-1)
* (9-4) = 4 * 3 * 5 = 12 * 6 = 60. And the final equation
and calculation for part A are represented by [((f-p) *
R(irj)t} ‘ {(P‘ll * R(ipkfltll/{P * (P'll * (f’Pll = 61-25/60
= 1.02. It will be left up to the reader to apply this
equation to any other parts which were used in these

experiments.

APPENDIX C

Raw Data,

87

APPENDIX C

Experiment One

45.
46.
33.
37.
18.
.0+/-10.7
36.

47

Low
Large

41
48
41
61
27
53
46

53
55
24
39

56
30

43
37
35
13
24
32
32

7+/-5.3
7+/—7.4
3+/—7.0
7+/—19.6
O+/-10.7

0+/-7.1

Table C-l: Positive Generalization Response Rates
Proximity: High High Low
Separation: Small Large Small

Session 1
Pigeon Number
1 51 52 41
2 41 56 29
3 47 53 42
4 55 49 55
5 26 35 17
6 58 58 35
7 32 29 47
Session 2
l 44 53 49
2 43 25 42
3 52 50 49
4 55 49 38
5 48 28 51
6 36 43 26
7 57 31 61
Session 3
l 50 41 49
2 32 38 29
3 44 51 32
4 54 61 61
5 51 38 51
6 45 1 26
7 20 55 50
Averages and standard deviations
1 48.3+/-3.1 48.7+/—5.4 46.3+/-3.8
2 38.7+/-4.8 39.7+/-12.7 33.3+/-6.l
3 47.7+/-3.3 51.3+/-l.3 41.0+/-7.0
4 54.7+/—O.5 53.0+/—5.7 51.3+/-9.7
5 41.7+/-11.2 33.7+/—4.2 39.7+/—16.0
6 46.3+/-9.0 34.0+/-24.l 29.0+/-4.2
7 36.3+/-15.4 38.3+/-11.8 52.7+/-6.0
Group 44.8+/—10.2 42.7+/—14.0 41.9+/-11.8

37.

8+/-14.3

88

Table C-2: Negative Generalization Response Rates

Proximity: High High Low Low
Separation: Small Large Small Large

Session 1
Pigeon Number

1 2 5 0 0
2 0 2 2 0
3 15 4 6 9
4 2 1 3 7
5 l l 12 10
6 7 0 3 0
7 8 ll 10 8
Session 2
1 5 7 15 2
2 15 4 0 5
3 4 9 6 10
4 12 5 33 30
5 0 2 0 0
6 O 0 0 2
7 7 l8 7 4
Session 3
1 10 7 1 16
2 3 3 2 1
3 4 1 1 3
4 1 7 7 0
5 3 l 1 1
6 1 0 0 0
7 10 6 25 3

89
Table C-3: Response Latency and Distribution Data I

Latency is first value listed, all times are in
milliseconds. Only responses to the first trial are listed.

Pigeon number 1:
Session one

High proximity, small separation: 1100, 1200, 1400,
1600, 1950, 2150, 2350, 2550, 2750, 2950, 3150, 3350, 3550,
3750, 4100, 4300, 4500, 4700, 4900, 5100, 5300, 5500, 5700,
5900, 6100

High proximity, large separation: 800, 900, 1100, 1300,
1500, 1700, 1900, 2100, 2300, 2500, 2700, 2900, 3100, 3450,
3650, 3850, 4050, 4250, 4450, 4650, 4850

Low proximity, small separation: 700, 750, 950, 1150,
1350, 1700, 1900, 2100, 2300, 2500, 2700, 3050, 3250, 3450,
3650, 3850, 4050, 4250, 4450, 4650, 4850, 5200, 5550, 5750,
5950, 6150, 6350

Low proximity, large separation: 500, 600, 800, 1000,
1200, 1400, 1600, 1800, 2000, 2200, 2400, 2600, 2800, 3000,
3200, 3400, 3600, 3800, 4000, 4200, 4400, 4600, 4950

Session two

High proximity, small separation: 3650, 3700, 3900,
4100, 4300, 4500, 4700, 4900, 5100, 5300, 5500, 5700, 5900,
6100, 6300, 6500, 6700, 7050

High proximity, large separation: 1400, 1500, 1700,
1900, 2100, 2300, 2500, 2700, 2900, 3100, 3450, 3650, 3850,
4050, 4250, 4450, 4650, 4850, 5050, 5250, 5450, 5650, 5850,
6050, 6250, 6450, 6650

Low proximity, small separation: 700, 800, 1000, 1500,
1700, 2050, 2550, 2750, 2950, 3150, 3500, 3700, 4050, 4250,
4600, 4950, 5300, 5500, 5700, 5900, 6250, 6450

Low proximity, large separation: 1000, 1100, 1300,
1500, 1700, 1900, 2100, 2450, 2650, 2850, 3050, 3250, 3600,
3800, 4000, 4350, 4550, 4750, 4950, 5300, 5500, 5700, 5900,
6100

Session three

High proximity, small separation: 550, 650, 850, 1050,
1250, 1450, 1650, 1850, 2050, 2400, 2750, 3100, 3450, 3650,
4000, 4350, 4550, 4750, 4950, 5150, 5500, 5700, 5900

High proximity, large separation: 2700, 2800, 3000,
3200, 3400, 3600, 3800, 4000, 4350, 4550, 4750, 4950, 5300,
5500, 5700, 5900, 6100, 6300, 6650

Low proximity, small separation: 1300, 1400, 1600,
1800, 2000, 2200, 2400, 2600, 2800, 3000, 3200, 3400, 3600,
3800, 4000, 4200, 4400, 4600, 4950, 5150, 5350, 5550, 5750,
5950, 6150

90

Table C-3, (Cont’d).

1250,
3400,

Low proximity, large separation: 450, 500, 850, 1050,
1450, 1650, 1850, 2050, 2400, 2600, 2800, 3000, 3200,
3600, 3800, 4000, 4200, 4400, 4600, 4800

Pigeon number 2

2450,
4450,

1350,
3350,
5350,

1450,
3450,

1450,
3450,

1250,
4150,
6150,

2450,
4600,

2750,
4750,

850,
2850,
4850,

1200,
3200,
5200

1250,
3250,
5250,

6500,

Session one

High proximity, small separation: 1950, 2050, 2250,
2650, 2850, 3050, 3250, 3450, 3650, 3850, 4050, 4250,
4650, 4850, 5050, 5250, 5450, 5650, 5850

High proximity, large separation: 700, 750, 950, 1150,
1550, 1750, 1950, 2150, 2350, 2550, 2750, 2950, 3150,
3550, 3750, 3950, 4150, 4350, 4550, 4750, 4950, 5150,
5550, 5750

Low proximity, small separation: 800, 850, 1050, 1250,
1650, 1850, 2050, 2250, 2450, 2650, 2850, 3050, 3250,
3650, 3850, 4050

Low proximity, large separation: 800, 850, 1050, 1250,
1650, 1850, 2050, 2250, 2450, 2650, 2850, 3050, 3250,
3650, 3850, 4050

Session two

High proximity, small separation: 550, 650, 850, 1050,
1450, 1650, 1850, 2050, 2550, 3050, 3400, 3600, 3800,
4350, 4550, 4750, 4950, 5150, 5350, 5550, 5750, 5950,
6350, 6550, 6750, 6950

High proximity, large separation: 2000, 2050, 2250,
2650, 2850, 3050, 3250, 3600, 3800, 4000, 4200, 4400,
4800, 5000, 5200, 5400, 5600, 5800, 6000

Low proximity, small separation: 2250, 2350, 2550,
2950, 3150, 3350, 3550, 3750, 3950, 4150, 4350, 4550,
4950, 5150, 5350, 5550, 5750, 5950, 6150, 6350

Low proximity, large separation: 150, 250, 450, 650,
1050, 1250, 1450, 1650, 1850, 2050, 2250, 2450, 2650,
3050, 3250, 3450, 3650, 3850, 4050, 4250, 4450, 4650,
5050, 5250, 5450

Session three
High proximity, small separation: 500, 600, 800, 1000,
1400, 1600, 1800, 2000, 2200, 2400, 2600, 2800, 3000,
3400, 3600, 3800, 4000, 4200, 4400, 4600, 4800, 5000,

High proximity, large separation: 550, 650, 850, 1050,
1450, 1650, 1850, 2050, 2250, 2450, 2650, 2850, 3050,
3450, 3650, 3850, 4050, 4250, 4450, 4650, 4850, 5050,
5450, 5650

Low proximity, small separation: 6050, 6100, 6300,
6700, 6900, 7100, 7300, 7500, 7700, 7900, 8100

91

Table C-3 (Cont’d).

1250,
3250,
5250,

Low proximity, large separation: 550, 650, 850, 1050,
1450, 1650, 1850, 2050, 2250, 2450, 2650, 2850, 3050,
3450, 3650, 3850, 4050, 4250, 4450, 4650, 4850, 5050,
5450, 5650

Pigeon number 3

2000,
4000,
6000,

1250,
3250,
5400,

1250,
3250,
5250,

1450,

1200,
3200,
5200,

1350,
3650,

1650,

3800,
5950

2900,

Session one

High proximity, small separation: 1400, 1450, 1650,
2200, 2400, 2600, 2800, 3000, 3200, 3400, 3600, 3800,
4200, 4400, 4600, 4800, 5000, 5200, 5400, 5600, 5800,
6200

High proximity, large separation: 550, 650, 850,
1450, 1650, 1850, 2050, 2250, 2450, 2650, 2850,
3450, 3650, 4000, 4200, 4400, 4600, 4800, 5000,
5600, 5800, 6000

Low proximity, small separation: 550, 650,
1450, 1650, 1850, 2050, 2250, 2450, 2650,
3450, 3650, 3850, 4050, 4250, 4450, 4650,
5450

Low proximity, large separation: 600, 850,
1650, 1850, 2050, 2250, 2450, 2650, 2850,

1050,
3050,
5200,

850,
2850,
4850,

1050,
3050,
5050,

1050,
3050,

1250,
3250

Session two

High proximity, small separation: 350, 600, 800,
1400, 1600, 1800, 2000, 2200, 2400, 2600, 2800,
3400, 3600, 3800, 4000, 4200, 4400, 4600, 4800,
5400

High proximity, large separation: 500, 750, 950,
1550, 1750, 1950, 2150, 2350, 2550, 2900, 3100, 3450,
3850, 4050, 4250, 4450, 4650, 4850, 5050, 5250, 5450

Low proximity, small separation: 850, 1050, 1250, 1450,
1850, 2050, 2250, 2450, 2650, 2850, 3200, 3400, 3600,
4150, 4350, 4550, 4750, 4950, 5150, 5350, 5550, 5750,

1000,
3000,
5000,

1150,

Low proximity, large separation: >10,000 (no responses)

Session three
High proximity, small separation: 2450, 2500, 2700,
3100, 3300, 3650, 3850, 4050, 4250, 4450, 4650, 4850,

5050, 5400, 5600, 5800, 6000, 6200, 6550

High proximity, large separation: 550, 650, 850,
1250, 1450, 1650, 1850, 2050, 2250, 2450, 2650, 2850,
3250, 3450, 3650, 3850, 4050, 4250, 4600, 4800, 5000,
5400, 5600, 5800

Low proximity, small separation: 400, 500,
1250, 1450, 1650, 1850, 2050, 2250, 2450, 2650,
3250, 3450, 3650, 3850, 4050, 4250, 4600, 4800,
5400, 5600, 5950, 6150

2850,
5000,

1050,
3050,
5200,

700, 900,

3050,
5200,

92
Table C-3 (Cont’d).

Low proximity, large separation: 900, 1000, 1200, 1400,
1600, 1800, 2000, 2200, 2550, 2750

Pigeon number 4
Session one

High proximity, small separation: 850, 900, 1100, 1300,
1500, 1700, 1900, 2100, 2300, 2500, 2700, 2900, 3100, 3300,
3500, 3700, 3900, 4100, 4300, 4500, 4700, 4900

High proximity, large separation: 1850, 1950, 2150,
2350, 2550, 2750, 2950, 3150, 3350, 3550, 3750, 3950, 4150,
4350, 4550, 4750, 4950, 5150, 5350, 5550

Low proximity, small separation: 600, 650, 850, 1050,
1250, 1450, 1650, 1850, 2050, 2250, 2600, 2800, 3300, 4050,
4250, 4450, 4650, 4850, 5050, 5250, 5450, 5650, 5850, 6050,
6250, 6450, 6650, 7000

Low proximity, large separation: 300, 350, 550, 750,
1100, 1300, 1500, 1700, 1900, 2250, 2600, 2950, 3150, 3350,
3700, 4050, 4250, 4450, 4800, 5150, 5350, 5550, 5750, 5950,
6150, 6350, 6550, 6750, 6950, 7150

Session two

High proximity, small separation: 300, 350, 550, 750,
1100, 1750, 1950, 2300, 3100, 3300, 3800, 4150, 4500, 4700,
4900, 5100, 5300, 5650, 6000, 6350, 6700, 7050, 7250, 7450,
8000, 8200, 8400, 8600

High proximity, large separation: 700, 900, 1100, 1300,
1500, 1700, 1900, 2100, 2300, 2500, 2700, 2900, 3100, 3300,
3500, 3850, 4050, 4250, 4450, 4800, 5000

Low proximity, small separation: 800, 850, 1050, 1250,
1450, 1650, 1850, 2050, 2550, 3650, 4150, 4500, 5600, 5950,
6150, 7250, 7750

Low proximity, large separation: 3850, 5000, 5350,
5550, 5750, 5950, 6150, 6350

Session three

High proximity, small separation: 700, 800, 1000, 1200,
1400, 1600, 1800, 2000, 2350, 2550, 2750, 2950, 3150, 3350,
3550, 3750, 3950, 4150, 4350, 4550, 4750, 4950, 5150, 5350,
5550, 5750, 5950

High proximity, large separation: 150, 250, 450, 650,
850, 1050, 1250, 1450, 1650, 1850, 2050, 2250, 2450, 2650,
2850, 3050, 3250, 3450, 3650, 3850, 4050, 4250, 4450, 4650,
4850, 5050, 5250, 5450, 5650, 5850, 6050, 6250

Low proximity, small separation: 700, 800, 1000, 1200,
1400, 1600, 1800, 2000, 2200, 2400, 2600, 2800, 3000, 3200,
3550, 3900, 4250, 4600, 4800, 5000, 5200, 5400, 5750, 5950,
6150, 6500, 6700, 6900, 7100, 7300, 7500, 7700

93
Table C-3 (Cont’d).

Low proximity, large separation: 550, 900, 1250, 1600,
2100, 2900, 3550, 4200, 4400, 4750

Pigeon number 5
Session one

High proximity, small separation: 7450, 7500, 7700,
7900, 8100, 8300, 8500, 8700, 8900, 9100

High proximity, large separation: 1500, 1600, 1800,
2000, 2200, 2400, 2600, 2800, 3000, 3200, 3400, 3600, 3800,
4000, 4200, 4400, 4600, 4800, 5000, 5200, 5400, 5600, 5800,
6000, 6200, 6400, 6600, 6800, 7000

Low proximity, small separation: 2300, 2400, 2600

Low proximity, large separation: 2400, 2500, 2850,
3050, 3250, 3450, 3650, 3850, 4050, 4250, 4450, 4800, 5150,
5500, 5700, 5900, 6100, 6300, 6500, 6700, 6900, 7100, 7300,
7500

Session two

High proximity, small separation: 3550, 3650, 3850,
4050, 4250, 4450, 4650, 4850, 5050, 5250, 5450, 5650, 5850,
6050, 6250, 6450, 6650, 6850, 7050, 7250

High proximity, large separation: 2150, 2200, 2400,
2600, 2800, 3000, 3350, 3550, 3750, 3950, 4150, 4350

Low proximity, small separation: 1800, 1850, 2050,
2250, 2450, 2650, 2850, 3050, 3250, 3450, 3650, 3850, 4050,
4250, 4450, 4650, 4850, 5050, 5250, 5450, 5650, 5850, 6050,
6250, 6450, 6650

Low proximity, large separation: 1750, 1850

Session three

High proximity, small separation: 2500, 2550, 2750,
2950, 3150, 3350, 3550, 3750, 3950, 4150, 4350, 4550, 4750,
4950, 5300, 5500, 5700, 5900, 6100, 6300, 6500, 6700, 6900,
7100, 7300, 7500, 7700

High proximity, large separation: 5600, 5700, 5900,
6100, 6300, 6650, 6850, 7050, 7250, 7450, 7650, 7850, 8050,
8250

Low proximity, small separation: 1500, 1600, 1800,
2000, 2200, 2400, 2750, 3250, 3750, 4100, 4450, 4950, 5150,
5350, 5550, 5750, 5950, 6150, 6350, 6550, 6750, 6950, 7150,
7350, 7550

Low proximity, large separation: 6200, 6300, 6500,
6700, 6900, 7100, 7300, 7500, 7700, 7900, 8100, 8300

94
Table C-3 (Cont’d).

Pigeon number 6
Session one

High proximity, small separation: 900, 950, 1150, 1350,
1550, 1750, 1950, 2150, 2350, 2550, 2750, 2950, 3150, 3350,
3550, 3750, 3950, 4150, 4350, 4550, 4750, 4950, 5150, 5350,
5550, 5750, 5950, 6150, 6350

High proximity, large separation: 550, 650, 850, 1050,
1250, 1450, 1650, 1850, 2050, 2250, 2450, 2650, 2850, 3050,
3250, 3450, 3650, 3850, 4050, 4250, 4450, 4650, 4850, 5050,
5250, 5450, 5650, 5850, 6050, 6250

Low proximity, small separation: 1200, 1300, 1500,
1700, 1900, 2100, 2300, 2500, 2700, 2900, 3100, 3300, 3500,
3700, 3900, 4100, 4300, 4500, 4700, 4900, 5100, 5300, 5500,
5700, 5900, 6100, 6300

Low proximity, large separation: 1300, 1400, 1600,
1800, 2000, 2200, 2400, 2600, 2800, 3000, 3200, 3400, 3600,
3800, 4000, 4200, 4400, 4600, 4800, 5000, 5200, 5400, 5600,
5800, 6000, 6200, 6400

Session two

High proximity, small separation: 1250, 1350, 1550,
1750, 1950, 2150, 2350, 2550, 2750, 2950, 3150, 3350, 3550,
3750, 3950, 4150, 4350, 4550, 4750, 4950, 5150, 5350, 5550,
5750, 5950, 6150

High proximity, large separation: 3800, 3900, 4100,
4300, 4500, 4700, 4900, 5100, 5300, 5500, 5700, 5900, 6100,
6300, 6500, 6700, 6900, 7100

Low proximity, small separation: 1550, 1650, 1850

Low proximity, large separation: 1200, 1250, 1450,
1650, 1850, 2050, 2250, 2450, 2650, 2850, 3050, 3250, 3450,
3650, 3850, 4050, 4250, 4450, 4650, 4850, 5050, 5250, 5450,
5650, 5850, 6050, 6250

Session three

High proximity, small separation: 2300, 2400, 2600,
2800, 3000, 3200, 3400, 3600, 3800, 4000, 4200, 4400, 4600,
4800, 5000, 5200, 5400, 5600, 5800, 6000, 6200, 6400, 6600

High proximity, large separation: 1350

Low proximity, small separation: 1450, 1500

Low proximity, large separation: 1000, 1050, 1250,
1450, 1650, 1850, 2050, 2250

95
Table C-3 (Cont’d).

Pigeon number 7
Session one

High proximity, small separation: 850, 900,
1200, 1400, 2350, 3000, 3650, 4150, 4350, 4550, 4900,
5600, 5800, 6000, 6200, 6400, 6600, 6800, 7000

High proximity, large separation: 500, 600, 800,
2550, 2750, 2950, 3150, 3350, 3550, 3750, 3950, 4150,
4550, 4750, 4950, 5150, 5350, 5550, 5750, 5950, 6150,
6550

Low proximity, small separation: 650, 700, 900,
1450, 1650, 1850, 2051, 2850, 3200, 3700, 3900, 4250,
4650, 5000, 5200, 5400, 5600, 5800, 6000, 6200, 6400,
6800

Low proximity, large separation: 600, 700, 900,
1300, 1500, 1700, 1900, 2250, 2450, 2800, 3000, 3200,
3900, 4250, 4600, 4950, 5150, 5500, 5700, 5900, 6100,
6500, 6700, 6900, 7100, 7300, 7500

Session two

High proximity, small separation:
1950, 2300, 2500, 2850, 3200, 3550,
5000, 5350, 5850, 6200, 6450, 6800,
8200, 8400, 8600, 8800, 9150

High proximity, large separation: 750, 1150,
1750, 2100, 2450, 2650, 2850, 3050, 3250, 3450, 3650,
4050, 4250, 4450, 4650, 4850, 5050, 5250, 5450, 5650,
6050, 6250

Low proximity, small separation: 600, 700, 1300,
2150, 2300, 2500, 2700, 2900, 3400, 3600, 4400, 4600,
5150, 5350, 5550, 5900, 6100, 6300, 6650, 6850, 7050,
7600, 7800, 8150, 8350, 8550, 8750

Low proximity, large separation:
1550, 1750,
3700,
5850

800, 900,
3750, 4250,
7000, 7250,

4450,
7650,

500, 750, 1150,
1950, 2150, 2500, 2700, 2900, 3100, 3300,
3900, 4100, 4300, 4500, 4700, 4900, 5100, 5450,

Session three
High proximity, small separation:
High proximity, large separation:
1150, 1350, 1550, 1750, 1950, 2150, 2350, 2550, 2750,
3150, 3350, 3550, 3750, 3950, 4150, 4350, 4550, 4750,
5150, 5350, 5550, 5750, 5950, 6150
Low proximity, small separation: 750, 850,
1450, 1650, 1850, 2050, 2250, 2450, 2650, 2850, 3050,
3450, 3650, 3850, 4050, 4250, 4450, 4650, 4850, 5050,
5450, 5650, 5850, 6050, 6250, 6450, 6650
Low proximity, large separation: 550, 650, 800,
1200, 1400, 1600, 1800, 2150

750
450, 550, 750,

1050,

1100,

5250,

2350,
4350,
6350,

1250,
4450,
6600,

1100,
3400,
6300,

1300,

4650,
7950,

1550,

3850,
5850,

1800,
4950,
7400,

1350,
3500,
5650,

950,
2950,
4950,

1250,
3250,
5250,

1000,

Appendix D

Appendix D
Raw Data, Experiment Two

Table D-1: Generalization Response Rates

Subject Part D Part B Part G Part B
Session one
1 6 1 2 1
2 27 1 8 1
3 35 4 15 1
4 4 2 1 0
5 0 0 0 0
6 0 0 0 2
Session two
1 1 0 2 0
2 37 2 1 1
3 9 3 8 5
4 0 6 1 3
5 0 0 0 0
6 0 0 0 0
Session three
1 1 0 1 0
2 0 15 0 0
3 9 3 5 5
4 1 0 2 2
5 0 0 0 0
6 0 0 0 0

Table D-2: Response Latency and Distribution Data II

Latency is first value listed, all times are in
milliseconds. Only responses to the first trial are listed.

Pigeon number 1
Session one
Part D: 400, 500, 700, 900, 1100
Part E: 1300
Part G: 6900, 7150
Part B: 8650

Session two
Part D: No responses (>10,000)
Part E: No responses (>10,000)
Part G: No responses (>10,000)
Part B: No responses (>10,000)

97

98

Table D-2 (Cont’d).

Part D:
Part E:
Part G:
Part 8:

Session three
responses (>10,000)
responses (>10,000)
responses (>10,000)
responses (>10,000)

No
No
No
No

Pigeon number 2

Part D:
Part E:
Part G:
Part B:

Part D:
3200, 3400,
5200
Part E:
Part G:

Part B:

Part D:

2500, 2700, 2900,

Part E:
Part G:
Part B:

3600,

Session one
750, 800, 1000, 1200, 1550
No responses (>10,000)
1100, 1200, 1300, 1500,
No responses (>10,000)

1600, 1800, 2000, 2200

Session two
2000, 2200, 2400,
4000, 4200, 4400,

1700, 1800,

3800,

2600, 2800, 3000,
4600, 4800, 5000,

No responses (>10,000)
No responses (>10,000)
2950

Session three
1100, 1100, 1300, 1500,
3100
No responses (>10,000)
No responses (>10,000)
No responses (>10,000)

1700, 1900, 2100, 2300,

Pigeon number 3

Session one

Part D: 400, 500, 700, 900, 1100, 1300, 1500, 1700,
1900, 2100, 2300, 2500, 2700, 2900, 3100, 3300, 3500, 3700,
3900, 4100, 4300, 4500, 4700, 4900, 5100, 5300, 5500

Part E: 2200, 2450, 2650

Part G: 650, 900, 1000, 1200, 1400, 1600, 1800, 2150,
2350, 2700, 3050, 3250, 3450

Part B: No responses (>10,000)

Session two

Part D: 350, 450, 650, 850, 1100

Part E: 2050

Part : 1000, 1050, 1250, 1450, 1800

Part B: 2350, 2400, 2600, 2800, 3000

Session three

Part D: 3350, 3400, 3600

Part E: 1100, 1300

Part G: 500, 750, 950

Part B: 6750

99
Table D-2 (Cont’d)

Pigeon number 4

Session one

Part D: 7650, 7800

Part E: No responses (>10,000)

Part : 1700

Part : No responses (>10,000)
Session two

Part D: No responses (>10,000)

Part E: 2150, 2350, 2700, 2900, 3100, 3300

Part : No responses (>10,000)

Part : 1550, 1650, 1850
Session three

Part D: No responses (>10,000)

Part : No responses (>10,000)

Part : 4400

Part : 7750

Pigeon number 5

Session one

Part D: No responses (>10,000)
Part E: No responses (>10,000)
Part G: No responses (>10,000)
Part B: No responses (>10,000)
Session two
Part D: No responses (>10,000)
Part E: No responses (>10,000)
Part G: No responses (>10,000)
Part B: No responses (>10,000)
Session three
Part D: No responses (>10,000)
Part : No responses (>10,000)
Part : No responses (>10,000)
Part : No responses (>10,000)

Pigeon number 6

Session one

Part D: No responses (>10,000)
Part E: No responses (>10,000)
Part G: No responses (>10,000)
Part B: No responses (>10,000)

Table D-2

Part
Part
Part
Part

Part
Part
Part
Part

100

(Cont’d).

D: No
: No
: No

B: No

D: No
E: No

: No
B: No

Session two
responses (>10,000)
responses (>10,000)
responses (>10,000)
responses (>10,000)

Session three
responses (>10,000)
responses (>10,000)
responses (>10,000)
responses (>10,000)