An Odyssey in Learning and Perception Eleanor J. Gibson
In the field of psychology, beginning in the 1950s, Eleanor J. ...
13 downloads
713 Views
30MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
An Odyssey in Learning and Perception Eleanor J. Gibson
In the field of psychology, beginning in the 1950s, Eleanor J. Gibson nearly single-handedly developed the field of perceptual learning with a series of brilliant studies that culminated in the seminal work, Perceptual Learning and Development. An Odyssey in Learning and Perception brings together Gibson's scientific papers, including difficult-to- find or previously unpublished work, along with classic studies in perception and action. Gibson introduces each paper to show why the research was undertaken and concludes each section with comments linking the findings to later developments. A personal essay touches on the questions and concerns that guided her research.
TABLE OF CONTENTS SERIES FOREWORD FOREWORD BY ELIZABETH S. SPELKE INTRODUCTION I EXPERIMENTAL PSYCHOLOGY IN THE THIRTIES (1932-1942) 1 BILATERAL TRANSFER OF THE CONDITIONED RESPONSE IN THE HUMAN SUBJECT, J. J. GIBSON AND G. RAFFEL. J. EXP. PSYCHOL., 1932, 15, 416-421. 2 RETENTION AND THE INTERPOLATED TASK, WITH JAMES J. GIBSON. AMERI. J. PSYCHOL., 1934, 46, 603-610. 3 SENSORY GENERALIZATION WITH VOLUNTARY REACTIONS. J. EXP. PSYCHOL., 1939, 24, 237-253. 4 A SYSTEMATIC APPLICATION OF THE CONCEPTS OF GENERALIZATION AND DIFFERENTIATION TO VERBAL LEARNING. PSYCHOL. REVIEW, 1940, 47, 196-299. 5 RETROACTIVE INHIBITION AS A FUNCTION OF DEGREE OF GENERALIZATION BETWEEN TASKS. J. EXP. PSYCHOL., 1941, 28, 93-115. RETROSPECT AND PROSPECT: ARE THEORIES RECYCLED? II COMPARATIVE RESEARCH ON LEARNING AND DEVELOPMENT (1952-1970) 6 THE ROLE OF SHOCK IN REINFORCEMENT. J. COMP. PHYSIOL. PSYCHOL., 1952, 45, 18-30. 7 THE EFFECT OF PROLONGED EXPOSURE TO VISUALLY PRESENTED PATTERNS ON LEARNING TO DISCRIMINATE THEM, WITH R. D. WALK. J. COMP. PHYSIOL. PSYCHOL., 1956, 49, 239-242. 8 THE EFFECTIVENESS OF PROLONGED EXPOSURE TO CUTOUTS VS. PAINTED PATTERNS FOR FACILITATION OF DISCRIMINATION. J. COMP. PHYSIOL. PSYCHOL., 1959, 52, 519-521. 9 BEHAVIOR OF LIGHT- AND DARK-REARED RATS ON A VISUAL CLIFF, WITH R. D. WALK AND T. J. TIGHE. SCIENCE, 1957, 126, 80-81. 10 DEVELOPMENT OF PERCEPTION: DISCRIMINATION OF DEPTH COMPARED WITH DISCRIMINATION OF GRAPHIC SYMBOLS. REPRINTED FROM J. C. WRIGHT AND J. KAGAN (EDS.), BASIC COGNITIVE PROCESSES IN CHILDREN, MONOGR. SOC. RES. CHILD DEVELOPMENT, 1963, 28, NO. 2 (SERIAL NO. 86), 5-24. 11 THE DEVELOPMENT OF PERCEPTION AS AN ADAPTIVE PROCESS, ELEANOR J. GIBSON. AMERICAN SCIENTIST, 1970, 58, 98-107. RETROSPECT AND PROSPECT: COMPARATIVE PSYCHOLOGY AND ANIMAL COGNITION III PERCEPTION: PSYCHOPHYSICS AND TRANSFORMATIONS (1954-1959) 12 THE EFFECT OF TRAINING ON ABSOLUTE ESTIMATION OF DISTANCE OVER THE GROUND, WITH R. BERGMAN. J. EXP. PSYCHOL., 1954, 473-482. 13 THE EFFECT OF PRIOR TRAINING WITH A SCALE OF DISTANCE ON ABSOLUTE AND RELATIVE JUDGMENTS OF DISTANCE OVER GROUND, WITH R. BERGMAN AND J. PURDY. J. EXP. PSYCHOL., 1955, 50, 97-105. 14 DISTANCE JUDGEMENT BY THE METHOD OF FRACTIONATION, WITH J. PURDY. J. EXPER. PSYCHOL., 1955, 50, 374380. 15 CONTINUOUS PERSPECTIVE TRANSFORMATIONS AND THE PERCEPTION OF RIGID MOTION, WITH J. J. GIBSON. J. EXPER. PSYCHOL., 54, 129-138. 16 MOTION PARALLAX AS A DETERMINANT OF PERCEIVED DEPTH, WITH J. J. GIBSON, O. W. SMITH, AND H. FLOCK. J. EXPER. PSYCHOL., 1959, 58, 40-51. RETROSPECT AND PROSPECT: PSYCHOPYSICS TO COMPUTATION IV PERCEPTUAL LEARNING (1955-1969) 17 PERCEPTUAL LEARNING: DIFFERENTIATION OR ENRICHMENT?, WITH J. J. GIBSON. PSYCHOL. REV., 1955, 62, 3241. REPLY BY L. POSTMAN: ASSOCIATION THEORY AND PERCEPTUAL LEARNING. PSYCHOL. REV., 1955, 438-446. WHAT IS LEARNED IN PERCEPTUAL LEARNING? A REPLY TO PROFESSOR POSTMAN. PSYCHOL. REV., 1955, 62, 447-450. 18 PERCEPTUAL LEARNING. ANNUAL REVIEW OF PSYCHOLOGY, 1963, 14, 29-56. 19 PERCEPTUAL DEVELOPMENT AND THE REDUCTION OF UNCERTAINTY. IN PROCEEDINGS OF THE 18TH INTERNATIONAL CONGRESS OF PSYCHOLOGY, 7-17, MOSCOW, 1966 20 TRENDS IN PERCEPTUAL DEVELOPMENT. EXCERPTS FROM CHAPTER 20 OF PRINCIPLES OF PERCEPTUAL LEARNING AND DEVELOPMENT (PP. 450-472). ENGLEWOOD CLIFFS, NJ: PRENTICE-HALL, INC., 1969.
RETROSPECT AND PROSPECT: THE COMING OF AGE OF PERCEPTUAL DEVELOPMENT V YEARS OF SIGNIFICANCE: RESEARCH ON READING (1965-1977) 21 LEARNING TO READ. SCIENCE, 1965, 148, 1066-1072 22 A DEVELOPMENTAL STUDY OF VISUAL SEARCH BEHAVIOR, WITH A. YONAS. PERCEPTION AND PSYCHOPYSICS, 1966, 1, 169-171. 23 CONFUSION MATRICES FOR GRAPHIC PATTERNS OBTAINED WITH A LATENCY MEASURE, WITH F. SCHAPIRO AND A. YONAS. CORNELL UNIVERSITY, 1968. 24 THE ONTOGENY OF READING. AMERICAN PSYCHOLOGIST, 1970, 25, 136-143. 25 PERCEPTUAL LEARNING AND THE THEORY OF WORD PERCEPTION. COGNITIVE PSYCHOLOGY, 1971, 2, 351-368. 26 HOW PERCEPTION REALLY DEVELOPS: A VIEW FROM OUTSIDE THE NETWORK. IN D. LABERGE AND S. J. SAMUELS (EDS.), BASIC PROCESSES IN READING: PERCEPTION AND COMPREHENSION. HILLSDALE, N. J.: ERLBAUM, 1977, 155-173. RETROSPECT AND PROSPECT: PERCEPTION, COGNITION, OR BOTH? VI PERCEPTUAL DEVELOPMENT FROM THE ECOLOGICAL APPROACH (1972 TO THE PRESENT) 27 THE SENSES AS INFORMATION-SEEKING SYSTEMS, WITH J. J. GIBSON. (LONDON) TIMES LITERARY SUPPLEMENT, 1972, JUNE 23, 711-712. SEEING AS THINKING: AN ACTIVE THEORY OF PERCEPTION. R. GREGORY. (LONDON) TIMES LITERARY SUPPLEMENT, 1972, JUNE 23, 707-708. 28 PERCEPTION AS A FOUNDATION FOR KNOWLEDGE: THOUGHTS INSPIRED BY PAPERS OF FRAILBERG AND BELLUGI. DISCUSSION PREPARED FOR THE LENNEBERG SYMPOSIUM, CORNELL UNIVERSITY, ITHACA, N.Y., MAY 1976. 29 PERCEPTION OF INVARIANTS BY FIVE-MONTH-OLD INFANTS: DIFFERENTIATION OF TWO TYPES OF MOTION, WITH C. J. OWSLEY AND J. JOHNSTON. 30 DEVELOPMENT OF KNOWLEDGE OF VISUAL-TACTUAL AFFORDANCES OF SUBSTANCE, WITH A. WALKER. CHILD DEVELOPMENT, 1984, 55, 453-460. 31 EXCERPTS FROM THE CONCEPT OF AFFORDANCES IN DEVELOPMENT: THE RENASCENCE OF FUNCTIONALISM. IN W. A. COLLINS (ED.), THE CONCEPT OF DEVELOPMENT. THE MINNESOTA SYMPOSIUM ON CHILD PSYCHOLOGY, VOL. 15. HILLSDALE, N.J.: L. ERLBAUM ASSOC., 1982, 55-81. 32 DETECTION OF THE TRAVERSABILITY OF SURFACES BY CRAWLING AND WALKING INFANTS, WITH G. RICCIO, M. A. SCHMUCKLER, T. A. STOFFREGEN, D. ROSENBERG, AND J. TAORMINA. JOURNAL OF EXPERIMENTAL PSYCHOLOGY: HUMAN PERCEPTION AND PERFORMANCE, 1987, 13, 533-544. 33 EXPLORATORY BEHAVIOR IN THE DEVELOPMENT OF PERCEIVING, ACTING, AND THE ACQUIRING OF KNOWLEDGE. EXCERPTS FROM ANNUAL REVIEW OF PSYCHOLOGY, 1988, 39, 1-41. EPILOGUE: PROSPECTS FOR A NEW APPROACH TO PERCEPTUAL LEARNING REFERENCES AUTHOR INDEX SUBJECT INDEX
Series Foreword
Lila Gleitman SusanCarey ElissaNewport ElizabethSpelke
Forewordby ElizabethS. Spelke
:;:1. 1 This volume portrays ascientist and her science .We learn about Eleanor J. Gibson and her field ,experimental psychology ,both through the scientific papers that she has collected here and through the personal essay that weaves these papers together by highlighting the questions and concerns that guided and animated her work . There are many things togain from this book .First ,the reader will learn agood deal about the nature and development ofperception and action from Gibson 's classic papers . Some of these papers have influenced experi mental psychology so profoundly that they may seem familiar ,even com -monplace , tofirst -time readers .What psychologist inthe 1990s could doubt ,for example ,that learning can bring changes inwhat we perceive ,not just changes inhow we respond towhat we perceive ?Yet psychology owes this insight ,and the defeat of arguments to the contrary toEleanor and James Gibson 'svigorous studies of perceptual learning . ,largely Other papers describe findings that goagainst deeply ingrained beliefs about the nature and development ofpsychological capacities .For exam ple ,on consider Gibson 'sdemonstration ,as an undergraduate that learning to act an object with one hand immediately transfers to,the other hand . This experiment shows quite clearly that what one learns ,when one learns anew and appropriate action ,tone isnot aresponse contingency ("affordance such as "flex this finger when you hear this ")but what Gibson calls an ": aproperty ofthe environment with consequences forthe perceiver /actor (should such as "after the tone , this plate gives off a shock ") . Gibson 's finding give pause toof those engaged inthe long -standing effort tounder stand action interms elementary responses and their concatenation . Another finding that experimental psychology cannot quite digest is Gibson ,Walk ,and Tighe 'sdemonstration that young animals of awide range of species ,capable including humans ,avoid visually specified drop -offs as soon as they are oflocomotion .Elegant experiments with animals
.. Xll
E. S. Spelke
that locomote atbirth ,and also with animals reared inthe dark ,reveal that the capacity toperceive athree -dimensional surface layout can develop in the absence ofany visual experience ortrial and error learning .Perceptu systems appear to,be built to bring animals information about the layout and itsaffordances contrary to empiricist theories ofspace perception . The reader ofthis book also willlearn something about the flourishin ofexperimental psychology intwentieth -century America .Gibson 'spapers portray adiscipline coming into itsown , focusing onfundamenta ques tions about perception and action ,gathering insights into the nature and development ofthese functions , and raising new questions forthe next century ofpsychology topursue . Consider ,forexample ,the progress that has been made inthe study of perception action ininfancy .field With her studies ofthe the visual cliff Gibson was one ofand the founders ofthat of research in late 19505 .,Lackin her own laboratory ,she was not able tocontinue her studies ofinfants at that time .Research oninfant perception continued inother laboratorie in the 1960s .Nevertheless ,the fundamental questions about perception and perceptual development - how perceivers apprehend alayout ofsurface , objects ,and events ;how the various perceptual modes work together to produce an experience ofaunitary world ;how perceptual systems serve to guide effective and adaptive action were raised consistently only by Gibson 'sunofficial students , T.G.R.Bower and Albert Yonas .Then ,in 1975 ,Eleanor Gibson was forthe first time inaposition toestablish her own infant study laboratory . The papers collected here show what has happened tothe field ofinfant perception since that time :Gibson 'sstudie ofsurface perception ,event perception ,perception ofintermodal relation , and perceptual guidance ofaction have helped bring the study ofinfant perception itsown contribution ,faculty during the four years that was both ainto member of.Her Cornell 'sactive and the director ofshe her infant laboratory ,isincalculable .Those years ofwork provide asgood an example asany ofthe progress that can bemade inexperiment psy chology when agifted scientist isgiven the freedom and the resource to pursue her science . During much ofher career , Gibson didnotenjoy this freedom . Her first goal was tobecome acomparative psychologist .With her training in perception , and with her theoretical disposition to view perception and action asinterconnected functions ,each adapted tothe environme ofa particular species , Gibson could have undertaken alifelong compara and ecological study ofperception and perceptually guided ackion .She was prevented from achieving this goal each time she tried : as a new gradua student , rejected from her psychology department 'sonly laborator of comparative psychology because of her sex ; as a technical assistan in a different laboratory ,charged with incompetence after making adiscove
Foreword
xiii
thatthedirector ofthelaboratory later published ashisOwn; asamature scientist, butwithout anacademic appointment, herattempts toconduct research atheruniversitys laboratory thwarted bythesudden removal of
theanimals she had been rearing and observing inanextended longitudinal
project.
Eleanor Gibsons career hasnotgreatly suffered from these reverses,
thanks toherindomitable spirit, butscience has. The most basic questions about surface, object, andevent perception inmost nonhuman species remain unanswered, even unasked. Thelack ofavigorous, comparative study ofperception andaction hashadserious consequences notonly for psychology but fortwoofitssurrounding disciplines: neurobiology and evolutionary biology. Eleanor Gibsons career hascoincided with a tremendous growth of research inneuroscience, most ofitconducted with laboratory animals and much ofitexploring theneural structures andprocesses subserving per-
ception andaction. Hercareer hasalsocoincided witha blossoming of zoological studies ofbehavior initsadaptive, evolutionary context. Both
these fields areinneed ofacomparative psychology ofperception and only occur ifinvestigations aregrounded onasolid understanding ofthe functions thatneural structures serve. Moreover, theright ecological and evolutionary analysis ofspecies-typical behavior requires anappreciation action. Reason andexperience suggest thatprogress inneuroscience will
notjustofananimals physical environment butofitsperceived environThecomparative, experimental psychology ofperception hasbeen so little developed, however, thatsome neuroscientists andevolutionary biologists appear tohaveconvinced themselves thatthefield isnotneeded: thatstudies ofneurobiology atoneend, andofevolution attheother, will suffice to unravel thesecrets ofmind, brain, andbehavior. Since this conclusion threatens tohamper progress inthebrain andbehavioral sciment as well.
encesforyearsto come, onecanonlyhopethatfuture scientists and
teachers with Eleanor Gibsons gifts, education, andgoals will come along and that they will beallowed tobuild theresearch discipline sheenvisaged. A thirdreason to readthisbookis to learnaboutGibson herself. Reading thispersonal account ofherlifeandwork isanexhilarating
experience. Despite theobstacles inherpath, hersisastoryofsuccess. Her
professional lifewasdevoted towork onproblems thatengaged herfrom thebeginning. Her work afforded insight into many ofthese problems and
raised newproblems tochallenge her.Welearn fromGibsons lifethat professional status andmaterial resources donotmake ascientist, however necessary theymaybetoa scientists work. Themostimportant in-
gredients ofagood scientist arethepersonal qualities Gibson exemplifies.
xiv
E. S. Spelke
Whatarethesequalities? Some aregiftsatwhich mostofuscanonly marvel: theprodigious intelligence thatenabled herto complete allher
graduate training short ofthethesis inoneyear; theextraordinary eyefor phenomena thatallowed hertorecognize ina common observation (for
example, thatnewborn goats dontfalloffstools) thegerm ofexperiments Othersare qualitiesthat we mayemulate:the of major consequence. courage toespouse anddefend original views inthefaceofopposition or indifference; thejudicious combination offlexibility, ingenuity, anddetermination thatallowedherto faceobstacles andprevail; thegenerosity
andpatience shedisplayed withstudents, giving ofhertimeandwisdom whenever it wasneededto anyonewithanycuriosity andwillingness to learn.
Forme,Gibsons moststriking quality ishernearly magical blendof firmness andopenness. Thefirmness hasbeenfeltbyeveryone whohas
ever disagreed withher,anditisevident throughout thisbook: Shehashad thestrength, conviction, andintegrity todevelop andmaintain herown enduring conception ofmind andaction. Atthesame time, Gibson is always wholly open tochange. Heropenness isexemplified atmany points
in thisbook.Note,forexample, the readinessalmost delightwith which shenowcriticizes someofherearlier theories ofreading andmany ofherearlyclaims aboutinfant perception. Thisopenness ischaracteristic
anddeep. Gibson hasalways triedto findthings out,nottovindicate previously heldopinions. Shehaschanged herideasagain andagain in response to newexperimental findings. A major partofhergenius, I believe, liesinherability to remain opento further development and
change while maintaining a conception ofhersubject thatisstrong and systematic andrichenough toserve asthesource andframework for all her work.
Whatisthisconception? Thisbookisanextended answer tothatquestion.Gibsons conception canbefeltintheearliest papers inthiscollection; anddeepened, butnotfundamentally changed, inthe it emerges enriched
papers thatfollow. Herconception cannot beidentified with anyparticula period inthehistory ofpsychology: itwasnever fashionable inthepast, anditisnotantiquated now.I tendtodoubtthatGibsons approach will solve allthemysteries thatpsychology seeks tounravel. When weareface toface, however, Icannot helpwondering whether shemight notberight
abouteverything. Eleanor Gibson seestheproblems ofperceiving, acting
organisms soclearly, andshehasalways hadsuch afirm sense ofhowto practicepsychology as a science.
In tr ad ucti an
xvi
Introduction
force ordynamism wasresponsible forthechanges? Wearereaching that
point justnowinperceptual development, andtheory should takea new
spurt. There wasa strong developmental approach inpsychology before
1950(forexample, Gesell andPiaget). Intheupsurge ofinfant psychology, itwaslostasweadmired thefeatsofthecompetent infantdisclosed with ournewtechnology; andnowthetimehascomefora renewed andstrong developmental approach leading toanunderstanding ofwhatmoves perceptual development.
Inputting thisbooktogether, I chose papers thatseemed significant in theirtime,andinaccompanying textI haveexplained theoccasion for
writing them andconsidered what significance they have now. Aretheold questions stillwithusinnewclothing, orhave weformulated newones? Is
thereprogress andcontinuity inascientific history, oreveninonepersons living of it?
I havedivided thepapersintosixsections andarranged themin a
roughly chronological order. Aconsistent chronological order wontdo, because accidents intervene ina lifetime andupsettheconsistency ofones advance toward a goal.Yet,thereisanunderlying focus. Inmycase,it
began withthese concernsan interest inanimal behavior anddevelopment,inlearning, andgradually inperceptionand thesewovethemselves together overtime.Myaimforthisbook,sinceI believe thata developmental approach isthemostfruitful one,istoshowhowchanges takeplace, where thetransition points lie,andwhere thecurrent isheading. Id liketo thinkId shown,too,howmuchfunit canallbe.
Towhomis thisbookaddressed? Principally to todaysgraduate stu-
dents and young
researchers. I had themin mind,becauseI thoughtthey
might findencouragement intheearlyreverses thatbefell mewhich, however discouraging at thetime,leftchannels thatledonto interesting work. Myowngraduate students havereadit,theysaidwithprofit. Naturally, I hopeanyone insearch ofatheory ofperceptual learning, ora wayofthinking about cognitive development andtheorigins ofknow!edge, willreadit.AndI hope itfills ina fewdetails ofthehistory of psychology.
I ExperimentalPsychologyin the Thirties (1932- 1942)
Introduction to Part I
- In the 1930spsychology was a lusty young science. The American tradition of functionalism, madememorableby Jamesand Dewey, flourished as a strong conviction of evolutionary theory and a lot of researchon animals; the shadeof Titchener survived in the thriving enterpriseof psychophysics; behaviorismwas not just a legacy but a fact. Psychology laboratories(all universities had one and so did any self-respecting college) were full of activity and an air of excitement. Here was a new science, bursting with discoveriesto be made, fields to be mined, theories to be destroyed or constructed. The brightest young academicswere attracted to the scene, so the company was good and the competition challenging. Learningwas the big topic of the day, as befit the functional tradition, but there were plenty of other tempting ones, such as motivation and psychoanalysisfor the daring, and psychophysicalresearchpursuedunder the most stringent rules for the less imaginative. University faculties acceptedpsychology as a biological scienceand even allowed sciencecredit to introductory courses with laboratory sections, which nearly every department of psychology
offered .
This was where I camein. The hardheadedatmospherepromising new findings from laboratories focused on behavior was incredibly appealing, and I was won over before I was twenty years old. Clinical psychology and testing were already looming large, but it was the laboratory scenethat dazzled. Nearly sixty years and more than sixty experiments later I can
easily capturethe stirring moments. It seemsworth describingthe psychology of that era as it struck a student of the times, illustrating it with papers of the era. They are my own (often with a collaborator to make things more interesting). I explain what motivated me at the time, and how times changedas psychology (and I) grew up through the rest of the twentieth century. I begin with a sketchof the psychology of the 1930s, as it existed in the laboratory, and follow with a brief sketchof my introduction to it.
PartI
4 The
Functional
The
Tradition
functional
tradition
universities
,
was
a
the
of
call
influenced
by
malized
by
extends
.
colonies
had
a
were
1925
)
fanned
country
the
,
and
,
,
and
The
-
but
finding
the
on
how
line
he
major
of
an
descent
Yerkes
and
,
Yale
research
never
.
.
had
interest
in
other
psychologically
min
and
course
for
but
a
Hobhouse
Park
topics
flames
with
from
of
Orange
favored
for
resurfaced
,
.
,
and
emphasis
,
and
it
in
rewarded
expanded
action
stronger
would
couched
have
descended
Romanes
in
of
were
be
of
too
was
at
to
niche
animals
hand
learning
we
were
problems
a
-
what
they
soon
Chicago
maze
.
other
chimpanzees
to
how
environmental
research
Lashley
related
theories
and
these
however
of
investigations
learning
putting
its
like
(
control
,
Darwin
communication
(
years
to
"
. )
perceptual
homing
primate
of
-
etc
and
"
numerous
cues
,
recent
with
of
were
Maze
Columbia
control
,
functionalisffil
The
.
at
Sensory
connectionism
behavior
to
"
floor
right
for
research
followers
oriented
In
and
was
.
.
,
s
"
on
left
.
There
,
so
,
Hull
back
primates
and
'
its
There
,
left
information
adapts
ded
(
foraging
animal
.
cues
Thorndike
of
in
research
)
turns
Clark
speak
too
olfactory
echolocation
of
way
,
strong
colleges
for
vision
particularly
many
topic
sequences
we
was
at
popular
role
now
and
who
superintended
solving
'
-
cognitively
Problem
Kohler
an
was
s
and
Mentality
animal
of
Apes
laboratory
in
this
.
Functional
ural
psychology
outcome
infant
Darwin
biography
Man
(
)
,
thrived
than
theory
were
put
the
The
his
,
.
It
sixties
,
'
a
tale
Psychophysical
hold
Developmental
was
.
to
be
.
but
it
Boring
and
' s
was
psychology
phenomenological
his
remains
but
it
book
later
,
in
,
.
-
an
in
"
Child
Psycho
with
laboratory
-
norms
where
really
nat
ontogenetic
concerned
a
a
wrote
Evolution
and
more
of
,
fact
Mental
fascinating
was
excitement
told
well
theories
came
into
its
own
.
I
what
admired
for
of
the
'
t
being
call
the
to
on
in
1933
do
.
)
G
but
America
was
a
to
was
.
As
the
bible
for
percep
-
perception
days
the
considered
term
at
in
those
opposed
-
Boring
research
in
which
psycho
.
did
as
,
measurement
,
E
psychophysicists
scientific
psychologists
weak
1930
by
(
wanted
allowed
Gestalt
and
by
alight
Consciousness
who
what
rigorously
the
kept
of
else
Establishment
,
absent
flame
Dimensions
anyone
wouldn
the
conspiculously
,
Physical
and
Nowadays
was
in
,
phylogenetic
,
,
was
alive
students
tion
,
Darwin
psychology
introspection
physics
graduate
Romanes
Yale
the
.
approach
.
Enterprise
Titchenerian
Harvard
.
fanciful
at
t
test
developmental
comparing
however
didn
a
theory
chart
Gesell
the
supported
evolutionary
son
a
with
to
s
own
provided
that
logyll
'
of
1889
development
in
also
of
,
,
and
perception
II
psychophysics
it
soft
, "
Experimental Psychology(19321942)
5
indicates, it wasdualistic, aiming to matchup physical dimensions of frequency, intensity, andsoonwithmental scales ofjudgment. It was nonfunctional, dry,eveninhuman, buta strong technology wasspawned andit haditsuses,making possible applications to industry andother sciences. Graduate students hadto learnallthepsychophysical methods
andthemathematics thatevolved tosupport them.Ina latersection I turn
topsychophysics withoutdoor studies ofdistance perception. Themethodswerealsouseful instudies ofperceptual learning, laterona major
interestof mine.To become a psychophysicist, onewentto Harvard,
whereBoringreigned. Learning in the 1930s
Learning wasthe greattopicof the day.It mightbe studiedas verbal
learning orasmemory,wherethebignames wereE.S.Robinson, author
ofAssociation Theory Today (1932), andfollowers ofhissuchasJohn
McGeoch andArthurMelton; or it mightbecentered aroundthecondi-
tioned response. Clark Hull, atYale, epitomized thehardheaded learning tionalist, however. I wasinspired asa young student byhisPsychological Review papers onconditioning, especially A Functional Interpretation of theConditioned Reflex(1929), inwhich hestrove to interpret various manifestations ofconditioning asadaptive behavior. Thefourpapers inthis theorist, basinghistheoryonconditioning. Heconsidered himself a func-
series areclassics, clear andinsightful. Theypredated hispapers composed stress wasplaced onmechanism anddeductive rigor. Whether Hullreally asminiature logical systems andthebooksonlearning theory,wherethe
achieved rigorornot,hehadvigorandhislaboratory wasanexciting
place,seethingwithactivity.
Hullwasnottheonlyfunctionalist concerned withlearning theory.
Therewashisarch-rival, Tolman, at Berkeley. Howa rat learneda maze wasasinteresting to Tolman asit wasto Hull,butalthough heeschewed
thewords andonlydescribed behavior, Tolman wasathearta cognitive
psychologist. Therat learnedthemazebecause it discovered whatledto whatandformeda cognitive mapofthelayoutandwhatit afforded. One
of thejoysof a meeting of theAmerican Psychological Association in thosedayswasseeingHullandTolman withtheirrespective followers
holding forthinthesameroom, everyone arguing excitedly (andgood
naturedly).
TheImpactof the Thirtieson a Student
I attended Smith College asanundergraduate, a greatplace fora young
womanto be introduced to science. Asa womenscollege, Smithhada
6
PartI
strongfemale faculty inscience andthearts,because thatwaswhere a
womanscholarcouldfindhonorand promotion.Thereweretruly great womenat Smith,for exampleMarjorieNicholson, the worldsexperton
scienceand imagination. I firstencountered psychology in a year-long introductory coursethatincluded a weeklylaboratory (I latertaughtit).
MysecondcoursewascalledAnimal Psychology, taughtby Margaret Curti,a Chicago-trained functionalist. It wasentrancing, becausewe studentswerepermittedto runan experiment ourselves. Theanimalswere rats, of course, and they learned a maze.
The realexcitementcamemy senioryearwhenI had a year-longcourse
in advancedexperimental psychology withJamesGibson,a youngassis-
tantprofessor freshfromPrinceton. Therewereeightstudents in the
class(fourof themlaterobtainedPh.D.s). Weworkedin pairsandperformedan experiment per month,settingup apparatus, obtaining and
running ourownsubjects, researching thebackground, andwriting it all up.Every experiment wasa novelproblem andwecovered a widefield, withexperiments onreaction time,learning, memory, perception, adaptation to prismsthe gamut of the fieldat the time.It was one way
thata youngprofessor whowanted to doresearch couldkeephishandin, despite teaching twoorthreecourses aterm.Manypapers came outofthe workbeguninthatcourse. 1wasproudthatanexperiment ofminewith GertrudeRaffelSchmeidler (latera ColumbiaPh.D.)waspublishedin the
Journal ofExperimental Psychology. Theexperiment wasonbilateral transfer ofa conditioned reflex.Wehada functional interpretation oftheCRofour own (see followingpaper),ahead of its time.
Thatcoursetaughtme the thrillof doingexperiments andthe great satisfaction of discussing problems dailywithcolleagues. Smithhadan excellent courseon historyandsystemsofpsychology, taughtby a Ph.D. ofBorings,andby thetimeI wasa senior,KurtKoffka wasgivinga course on GestaltPsychology, whichI attended.Later,whenI wasa graduate student, I regularly attended Koffkas seminar asmostoftheSmithpsycho-
logyfaculty did.Itwasthelocalforum fordebate, despite Koffkas rather authoritarian style.I wasnevermuchattractedto Gestaltpsychology, as I haveeverbeenuneasywitha phenomenological approach,a factthat undoubtedly influenced mychoiceofYaleforgraduateschool. SmithCollegewasverysupportive andencouraged its studentsto go
onto graduate work.I stayedonat Smith fortwoyearsaftergraduation asa teaching assistant andobtained a masters degree there.Mythesis (donewithJames Gibson) wasonretroactive inhibition inverballearning,
verymuch inafunctional tradition. Itwasdeepdepression times andthere werefewfellowships for women.WhenI wentto graduateschoolat Yale,Smithgavemea smallfellowship, butYalecontributed nothing.
ExperimentalPsychology (1932- 1942)
7
At Yale, I intended to specializein comparativepsychology and work in Yerkes's laboratory, not so much becauseof Yerkes's ideas but for the glamor of working with chimpanzees(a poor reason, I fear). That expectation was not fulfilled . Yerkes informed me, none too graciously , that he allowed no women in his lab- a real shock after the nurturing atmosphere
of Smith. Clark Hull acceptedme as a graduate student, however. He was deep in his enterpriseof building logicodeductive systems. He allowed me to choosemy own topic, but I was required to back it up by a theoretical presentationwith formally stated premises, deductions, etc. I managedto
do it, but with a bit of tonguein cheek . I wantedto work on verballeaming andforgetting, andto showthat remembering andforgettingdepended on differentiatingcontentinto meaningfulunitsthat did not confusewith one another in an incoherent melange. I could borrow my main concepts, differentiation and generalization , from the conditioned response literature , which made them very respectable at the time . But the terms held com-
monsensepsychological meanings for me, perceptual meanings I would have said later . Noone at Yale worked on perception or was in the least interested
in it .
I performed my thesis experiments , except the one on generalization in a reaction time experiment , at Smith, where I returned as an instructor
after a year at Yale. I had a very busy year at Yale getting through all the many requirements, including a proseminar for all the new students. It was a showcasefor the professors, each of whom had a few weeks to display his ideas (all were men, of course). Yale was generous in letting me take a degreewith only one year's residencecredit. I took Clark Hull's enormous
memory
drum back to Smith and worked
there on my disserta -
tion. Smith retainedme on its faculty and kept me so busy that, except for directing a few master's theses, I got little further researchdone. It was after World War II before the scenereally changed for me, as it did in many ways for psychology in America.
Bilateral Transfer oftheConditioned Response
in the Human Subject
JamesJ. Gibson, Eleanor G.Jack,Gertrude Raffel
This paper isafunctional treatment oftheconditioned response, emphasizing its
adaptiveness asa response ofthewhole organism rather thananisolated reflex as Paviovians would haveit.Thediscussion emphasizes thispoint: Ourhypothesis would haveit thatintheadult,organized habits ofwith-
drawing orpulling awayhavebeen developed forallpartsofthebodywhich havecome incontact withpainful stimuli. Wecansuppose thenthatinour experiment, theconditioned stimulus hasbecome effective notonlyforthe
particular withdrawal response inconnection withwhichit waslearned, but alsoforthisentirerepertory ofavoiding reactions.
Thispointwassupported inlaterexperiments byWickens (Wickens 1939a andb)in whichfingerwithdrawal wasconditioned withthehandin normal position andthenreversed sothata flexorratherthanextensor movement hadto
beperformed iftransfer occurred. Itdid.InpartII,anexperiment thatIperformed much later with goats assubjects leads toasimilar conclusion, soitwasnotonly human adultsubjects forwhom avoidance toshock isnota simple reflex. When
wecometopartVI,I showhowthemuchlaterconcept ofaffordance fitsthese
cases, which arebasically similar totheresponse tolooming presented there.
It is interesting to findHilgardandMarquisin theirclassic work,Con-
ditioning andLearning, referring toourresults andWickenss as insightful behavior infypical conditioning experiments (Hilgard andMarquis 1940, 243). Theyalsosayin thesecases theequivalence isobviously learned. I wouldnt be
so sureaboutthat.I thinkwe havediscovered sincethenthat behavior is
organized originally inlarger synergies thatinvolve patterns ofpostural activities
thatcanbe broken downor individuated intosmaller ones,ratherthanthe
oppositenot a learned equivalence ofsmaller pieces asHilgard andMarquis supposed in1940. Theelementarism ofearlybehaviorism diedhard(ifit has). JournalofExperimental Psychology, 1932,41O421.
10
J. J. Gibson , E. G. Jack , & G. Raffel
However , thesetwo importantcriticsof theirtimedid raisemanydoubtsabout buildinga theoryof behavior on reflexes that couldonly widenthedomainof actingby being"conditioned " to whatever stimulichance andopportunity happened t-""" .""... ..., toJjuxtapose rr~rr-r --- with -- -- them. During the courseof an experimenton the conditionedwithdrawalresponseof the handto an electricshock,the opportunitypresented itselfto testfor transferof this conditionedresponse from the trainedright handto the untrainedleft hand. The subjectsat at a table with the palm of the handrestingon onelargebrasselectrodeandthe middlefingeron another smallerone. The handwasnot strappeddown or constrained in any way. Theapparatus andelectricalcircuitwereessentially the sameasin Watson's research ,l exceptthat the arrangemen ~ for graphicallyrecordingthe responseson a kymographdrumwas not employed . One of the Es simply satat the sideof S andrecordedthefingermovements asthey occurred . In order to keepthe attentionof S off the experiment , and to lessenany anticipatoryfearof the electricshock,S wasrequiredto readaloudduring an experimentalsitting. The conditionedstimuluswas the soundof an electricbuzzerlastingfor one secondand startingapproximately~ sec. beforethe shock.This timing wasaccomplished by usinga doublecontact key which, whendepressed , first completedthebuzzercircuitandthenthe shockcircuit. Theprocedurewasasfollows. After Swasseatedwith the right handon theelectrodes , andhadbegunto readaloud, thebuzzerwassoundedalone. Forseveralsubjectswho weresuspicious or fearful, the soundby itselfwas adequateto producea withdrawalmovement . Although this response alwaysdisappeared afteroneor two repetitionsof the buzzerstimulus , the recordsof suchsubjectshavebeenomittedfrom the dataon transfer . After the abovetest, combinedstimulationsof buzzerand shockweregiven at irregularintervalsvaryingfrom 3 to 20 seconds . When 10 repetitionshad beencompletedoneof two procedures wasused.(1) In thefirst, thebuzzer wassoundedaloneandany conditionedwithdrawalmovementwasnoted. If noneappeared , anotherseriesof reinforcements wasgivenanda second test for the conditionedresponsemade.Whenthe latter madeits appear ance, a final seriesof reinforcements was given and then S was told to shift over and placethe left hand on the electrodes , with the middle fingeron the far electrodeasbefore. The buzzerwasnow soundedalone andany withdrawalmovementnoted. (2) In the secondprocedure , instead of waitingfor theconditionedresponse to appearin theright hand, E made a testfor thetransferred response in theleft handimmediatelyafterthefirst 10 combinedstimulations . If noneappeared , Swasrequiredto shiftbackto the right handand more reinforcements were given, after which another
Bilateral Transfer of Conditioned Response test for transfer was made . As a check on the transferred
11
response when it
occurred, a final test for the conditioned movement in the right hand was
subsequentlymade. Results
Twenty subjects were used in the experiment . Of these, 19, or 95 percent,
were successfullyconditioned to withdraw the finger of the right hand when the buzzer was sounded alone . From 10 to 52 repetitions of the combined stimuli were necessary to establish the response . (The one sub -
ject who failed to becomeconditioned had undergoneelectricaltreatments for a year , and showed a very weak withdrawal
response even when the
shockadministeredwas much stronger than that usually employed. At the end of 200 repetitions no conditioned response had occurred and the
experimentwas stopped.) Of the 19, 6 showed someindication of responding to the buzzeralone before the experimentwas begun, and their results have accordingly not been considered. Thirteen subjectsremain for whom the data on transfer of the conditioned responseseemvalid. Although at no time did any subject in this group receive a shock with the left hand, 8,
or 62 percent made definite withdrawal responsesof the finger when the left hand was placed on the electrodesand the buzzer alone was sounded. Three of this group had given evidence of a conditioned response of the
right hand before the left hand was tested, whereas5 showed the response in the left or transferhand first and subsequentlyin the right . It is possible that a larger number of subjects might have manifested transfer if the experiment had been continued longer. No more than 100 combined presentations of buzzer and shock were given to anyone subject.
If by that time no indication of transfer had shown itself the attempt was abandoned. This was done in part out of consideration for the subject for whom the experiment was naturally somewhat uncomfortable , and in part
out of a desire to investigate other problems in addition to the transfer phenomenonduring the experimentalsitting.2 The appearanceof the conditioned withdrawal responsein the right hand itself was found to be somewhat irregular and unstable, this finding being in accord with the results of previous experiments.3 In 2 subjects, only one definite conditioned response could be elicited , all attempts there-
after being negative. These subjectswere apparently surprisedand somewhat chagrined to find that they had reacted to the sound of the buzzer alone , and thereafter
set themselves
" not to be fooled
more than once ."
They constitute two of the group for whom transfer could not be demon strated. In general the reports of subjects indicate that various attitudes such as timidity towards shocks, a desire to appear indifferent , a desire not
to be tricked into responding, and the like, played a large part in the
12 J. J. Gibson , E. G. Jack , &: G. Raffel experiment , andwerethe probableexplanationof muchof the irregularity with
which
the conditioned
response
showed
itself . In view
of this in -
stability of the responsein the right hand it is not surprising that the transferred response in the left hand could not always be demonstrated . Discussion
In a significant number of cases(62 percent) the establishing of a conditioned withdrawal responseof the right middle finger is accompaniedby the formation of a similar conditioned responseof the correspondingfinger of the other hand . From this fact alone little can be concluded
with certain -
ty . Chiefly it suggests the need for further research on such questions as the
extent of transfer to other (unsymmetrical) fingers of the same and the opposite hand, the latent periods of transferredresponsesascomparedwith the latency of the primary conditioned and unconditioned response, the
possible occurrenceof implicit movements in the transfer hand, and the like. But it also suggests a possible explanation towards which further experiments may be oriented .
It is clear in the first place that the conditioned withdrawal movement can scarcelybe considereda single isolated reflex, or even a conditioned reflex, if by the term we mean a responselimited to a specificmuscle or muscle group. This latter criterion is frequently regarded as definitive for the true reflex. The hypothesis is suggestedthat the conditioned with drawal to shock involves, or perhapsis itself, a generalizedhabit of avoiding or withdrawing when the buzzer is heard, which may be evoked from another part of the body than that in which the response was learned. Unfortunately no systematic data were obtained as to transfer of the
responseto other fingers. It was, however, informally determined with several subjects that the conditioned response did readily transfer to the
index finger of the same hand. The extent to which a general avoidance response to the buzzer was set up can only be inferred . But from the above
evidence and from what is known of transfer of learning to symmetrical and unsymmetrical parts of the body4 it is possible to supposethat the withdrawal response could have been elicited in the other fingers and perhapseven in other parts of the body. It should be noted that the conditioned response which is concurrently
establishedin the untrained hand, is really a latent or potential response. It becomes actual and overt only when the finger of the untrained hand is for
the first time resting on the electrode. Our hypothesis would have it that in the adult, organized habits of withdrawing or pulling away have been developed for all the parts of the body which have frequently come in contact with painful stimuli. We can then supposethat in our experiment , the conditioned stimulus has become effective not only for the
Bilateral Transfer ofConditioned Response 13
particular withdrawal response inconnection withwhichifwaslearned, but
alsoforthisentire repertory ofavoiding reactions. Furthermore when any
part,saya finger, is ina potentially painful situation (e.g.,restingonan
electrode) wemayconceive thattheappropriate withdrawal response for
thatfingeris in a stateof readinessperhaps of sub-activation. It is this
particular finger, therefore, which iswithdrawn undersuchcircumstances,
ratherthananyother.Consequently thedifference between thelatentor
potential conditioned response andtheactual onedepends onthepresence ofa specific preparatory setarousedby thesituation. Ontheconscious side,insofarastheseactivities arerepresented in
consciousness, the aboveformulation wouldrun as follows.The condi-
tionedstimulus hasbecome during training a general signal to withdraw.
Together withthishabit, there ispresent (normally) a definite preparatory
set to withdraw the particular fingerwhichis restingon the electrode. Hence whenthefingeroftheuntrained handisplaced inthatsituation, the soundofthebuzzer setsofftheproperresponse.
Theintrospective reports givenbythesubjects indicated notonlythat thesetorattitude wasanimportant factor intheexperiment butalsothat
theresponses, bothconditioned andunconditioned, couldnotbeclassed as
definitely either voluntary orinvoluntary. Theywerenotcompletely vol-
untaryin thesenseofbeingintended, as,e.g.,a fingermovement in the
reaction timesituation isintended. Theconditioned responses wereusually accompanied bya feeling ofsurprise. Ontheotherhand,theywerenot wholly involuntary inthesense ofbeing automatic asisthekneejerk.The conditioned responses couldhavebeeninhibited hadthesubject wished
to inhibitthem.Thecasesof thetwosubjects whoset themselves not to
befooled morethanonceareinpoint. Thattheunconditioned response itselfcouldbeinhibited inlargepartwasshown byonesubject whotook
the instructions to forgetaboutthe fingeron the electrode so far as
possible, asmeaning thatheshould notrespond voluntarily. Thefinger
wassosuccessfully forgottenthattheonlyresponse to theshock wasa slighttrembling or twitching movement, evenwhenthe stimulus was painfully strong. Fromtheaboveevidence ifappears evenmoreclearthat
avoidance toshock isnotareflex. Itisrather aresponse highenough inthe scaleofcomplexity tobeinpartdependent ona voluntary set.
Sincea conditioned responsecanbe formedbetweena stimulus anda
response which wasnever overtly made during theconditioning process, theinadequacy ofthesimple diagram orphysiological schema usedfrequently inexplaining conditioning isoncemoredemonstrated. Anyreal
physiological explanation of the presentfactswouldhaveto be based onanadequate theoryofthephysiological basisofbilateral transfer. Such a theoryhasnotyetbeenformulated. Itispossible thatfurther research on
14
J. J. Gibson , E. G. Jack , &: G. Raffel
transferof the simple type of learning here describedwill contribute to this problem. Notes
1. For the original researchon this subject, seeJ. B. Watson, The Place of the Conditioned Reflex in Psychology, Psychol . Rev., 1916, 23, 89- 116. 2. These additional problems had to do with the fact of so-called 'sensory irradiation' (Pavlov) and the establishing of differential responsesto buzzers of different pitch. 3. J. B. Watson, The Placeof the Conditioned Reflex in Psychology, Psycho 1. Rev., 1916, 23, 89- 116. G. Humphrey, The Effect of Sequencesof Indifferent Stimuli on a Reaction of the Conditioned Reflex Type, J. Abn. and 50c. Psychol ., 1927, 22, 194- 212. 4. Seefor example C. W . Bray, Transfer of Learning, J. Exp. Psychol ., 1928, II, 443- 467.
2
Retentionand the InterpolatedTask Eleanor J. Gibson, James J. Gibson
-
.
This paper is a condensationof my master's thesisat Smith College. JamesGibson
wasmy thesissponsor(hebecame my husbandwhile I wasworkingon thethesis ). Verballearningand memory , in the Chicagotradition, was muchin vogue(note thereference to E. S. Robinson , oneof the Chicagogroup). Thematerialemployed in theseexperimentswas typically letters, digits, or nonsensesyllables. Retroactive inhibition was a popular notion for explaining forgetting, and similarity of
interpolatedmaterial to whateverwas originally presentedfor learning was considered critical. But thesimilarity wasgenerallycontentoriented , and referred to likeness between individual items, consistent with the elementarism of the
thirties. My functional-mindedness ledme to the hypothesis that the taskengaged in, which is oriented toward process rather than content, must be of equal importance. It makes me think now of the later information processingtermi-
nology, emphasizing "processing " of "alphanumeric characters ." But of coursethat would be twenty-five years later. The experiment varied both content and task with respect to original and interpolated task relations. " The similarity of two tasks, then, will be expressedin
termsof the significant featureswhich the two taskshave in common." The features chosenwere the "material" and the "operation" performed on it . As the
experiment demonstrated , bothso-calledfeatureshad the effectof reducingretention when they were presentedseparately in an intervening task, but detrimental effectswhen they were combinedweregreater than the sum of the injury done by the two separately. I supposethat one could say that "processing" numbers is not the same task as "processing" letters, whatever the operation. This experimentled me to think further about retroactive inhibition and consequentlywas the fore-
runner to my Ph .D . thesis.
American Journalof Psychology , 1934, 46, 603- 610.
16
E. J. Gibson & J. J. Gibson
Thesimilarity problem inretroactive inhibition arises fromthefactthatthe integrity ofretention ofmemorized material isa function ofthesimilarity between thelearning andtheinterpolated activity. According totheSkaggsRobinson theory,thefunction is of thefollowing nature.Thecurveof retentionstartsat a highlevelwithmaximum similarity, whichRobinson took to meanthat the interpolatedtaskis identicalwith,or a continuation
of,the original learning. Thecurvefallsas thesimilarity of interpolated materialis reducedfromidentityuntilretentionreachesa minimum at the
stagewhich, presumably, otherinvestigators havecalled a similarinterpolation. Thecurvethenrisesuntil,at the stageof dissimilar or rest interpolation, retention isat thelevelusually considered normal. Thelatterportionof thiscurveembodies thetypeof resultsusually obtained in experiments on thesimilarity problem. Thefirstpart,showingdecreasing retention withdecreasing similarity, hasbeenverified by Robinson 2 and Dreis 3 with two experimentswhereinthe interpolations
consisted of varying amounts ofrehearsal to whichwereaddedcomple-
mentary amounts ofnewmemorization. Thegreater theproportion of rehearsal, thegreater wasthesimilarity oftheinterpolated activity to the original learning. These experiments, along withsomeothers, 4 arosefrom Robinsons tentativeassumption thatimprovement through practice andthe disintegrating effect ofinterpolation arerelated phenomena; inotherwords,
thatpractice-effect passesoverintointerpolation-effect without break. It shouldbe notedthat,in thisattackon the similarity problem, the interpolated activity isnota singletask,asit canbeonlyifit is separate anddistinctfromtheoriginal learning task.In otherwords,theinterpolationsof Robinson andDreiswerenot activities havingan effect(either facilitatory or inhibitory) on the integrityof retention of the primary learning. Theywereinstead continuations oftheprimary learning invary-
ingamounts, combined withnewlearning. Likewise, theretention which wasmeasured intheseexperiments wasnotretention ofa primarylearning, but of that plussomecontinuationof it.
Theposition canbetaken, ofcourse, thattheretroaction problem need
not be definedin termsof discreteprimaryand secondary tasks.The
interpolated activity doesnothavetobea separate andunique task,the argument mightrun,sinceit isprecisely thefactor ofindistinctness and confusion betweentaskswhichis mostlikelyto be the explanation of
retroactive inhibition. Thishypothetical argument, however,begstheques-
tioninfavorofa radical formofthetransfer theoryofretroaction. There
isasyetnoproofthatit is thesingle andcomplete explanation ofthe phenomenon.
Thetheoretical approach oftheexperiment to bedescribed differs from thatexpressed by the Skaggs-Robinson function in that retroaction is
thought of as oneof theproblems of theinterrelations of tasksthat
Retention andtheInterpolated Task 17
problem dealing withtheeffect ofa second taskonthetestedretention ofa first.Thisemphasis ontask-characteristics carries withititsownmode of dealing withthesimilarity factorin retroaction. Twotasksaredesignated assimilar withreference to definite features which thetwohave
incommon. Such features would notbeindependent elements butaspects bymeans ofwhich twotasks could becompared. There areundoubtedly manysuchwhichcouldbe usedas a basisof comparison, butcertain features ofa taskwillstandoutasbeingmoreimportant thanothers foragenuine description ofitandconsequently fora significant statement ofsimilarity. Thesimilarity oftwotasks, then, willbeexpressed interms
of thesignificant features whichthetwotaskshaveincommon.
Thetwomostobvious features ofa taskaretheoperation which the
subjectis instructed to perform andthe material withwhichhe must
perform it.Thetermoperation willbeusedto mean thataspect ofa
taskbestindicated bythewordAufgabe, andthetermmaterial willbeused to indicate whatcouldbecalledtheperceptual data.Thesetwofeatures
arenotdistinct components. They areatleastpartly interdependent, since achange inthematerial ofa taskwould beaccompanied bysome change intheoperation, andviceversa. Forexample, theoperation inmemorizing
nonsense syllables is not the sameas in memorizing nonsense forms, althoughboth are memorial in nature.
Thisdistinction isnotnew. There have beenrelated onesranging from behavior asgoal-seeking andasinvolving behavior-supports. 5 Robinson himself, inanearlier experiment, distinguished between theprocessand thedistinction between actandcontent to Tolmans characterization of
thecontentofa taskandfound thatthe degree ofretroactive inhibition isa function ofsimilarity ofprocess aswellassimilarity ofcontent. 6 He
further demonstrated thatanother feature, which hecalled formofpresen-
tation,wasimportant forretroactive inhibition. Hisdata,however, didnot
show thatthese features were independently capable ofcausing retroaction. Asaconsequence, perhaps, hedidnotfollow upthislineofthought inhis subsequent research. With theexception ofthisexperiment ofRobinsons, allresearches onthesimilarity factor haveworked withsimilarity ofmaterial alone. When variations intheoperation oftheinterpolated taskhave been included inanexperiment, thisfactor hasnotbeenspecified norhas
its effectbeenisolated andmeasured. Littleis known, then,aboutthe influence ofoperational orfunctional similarity onretroaction. TheExperiment
Intheexperiment to bedescribed, ourproblem wasto decide upona setofinterpolated tasks which would enable ustoisolate andcompare the decrements inretention separately caused bysimilar operation andbysimi-
18
E. J. Gibson & J. J. Gibson
lar material.The relative importanceof the two factors for retroaction
couldthusbe determined. Specifically, we wishedto comparedegreesof
retentionof the primarylearningin the following cases:(1)whenthe secondary taskis likethe primarytaskin bothoperation andmaterial; (2)whenit is likein operationbut differentin material; (3)whenit is differentin operationbut likein material; (4)whenit is differentin both operation and material.
In order to makethe above comparisonsit was necessaryto selecta primary
learning taskandfourkindsofinterpolated tasksfulfilling theaboverequirements. On the basisof a preliminary experiment, learninga listof pairedconsonantswas
chosenfortheprimary taskTheinterpolated taskswere(1)learning anotherlist
ofpairedconsonants, (2)learning a listofpaired digits,(3)cancelling a specified pairofconsonaMs whenever it appeared ina sheetofpiedtype,and(4)cancelling a pairofdigitsinthesame manner. Afifthinterpolated task,ofthesortgenerally employed asa control orrest condition, wasalsoincluded. Itconsisted oflooking at pictures (moving-picture stills)andit wasbelieved to differfromtheprimary task even morethan did cancelling digitsin fact it was designedto be as differenta task as possible.Amongother features,it lackedthe motivational or work characteristics of the other four tasks.
For the sakeof simplicity, we shallcall the tasks,in order,Learning Letters,
Learning Numbers, Cancelling Letters, Cancelling Numbers, andPictures. Thefollowing schemadesignates by capitallettersthe arrangement of tasksin the fiveinter-
polated conditions. Therewere5 groups ofSs,onegroupcorresponding toeach condition.
Interpolated Condition
(Group)
Primary Learning
Secondary Task
V
LL
P
I II III IV
LL LL LL LL
LL LN CL CN
Experimental materials
Forprimary learning, allgroups studied alistof10pairs ofconsonants. Nopairs
wereincluded which made familiar abbreviations (such asPM); nopairconsisted ofconsecutive letters ofthealphabet; andnearly alltheconsonants ofthealphabet wereusedinonelist.Forthesecondary task,Group I learned another listof
consonants made upliketheonejustdescribed, Group IIlearned alistof10pairs
ofdigits. Nodigit appeared with disproportionate frequency, nordidconsecu tivepairs contain adigit more than once. Group IIIhadtoread through asheet ofpied consonants striking outthecombination KSwhenever itappeared. The combination occurred aboutthreetimesineverytwolines. Group IVcancelled adigit combination inthesame manner. Thepictures secured forGroup Vwere
Retention andtheInterpolated Task 19
photographs distributor. (SxIIin.) ofdramatic situations obtained from amotion picture Procedure
Five laboratory sections ofalarge course inpsychology served asthe5groups of intheassigning ofstudents tosections, thegroups were probably asnearly equal Ss.They averaged about 26Sseach. inview oftheabsence ofanyselective factor
inability asanythatcould befound. Theexperiment wasadministered inthe
manner ofagroup mental testwiththereading ofdefinite instructions bytheF andemphasis oncareful adherence tothem. Thematerial foreach group was bound inpamphlet formandpassed outtotheSs.Thefirstsheetcontained a warning nottoopen, thesecond sheet hadonit thelistforprimary learning, thethird wasblank, thefourth contained thematerial fortheinterpolated task, the fifth hadacolumn of10short lines forthewritten recall oftheprimary list,and forGroups I andTithesixthhadanother setoflines fortherecall oftheinterpolated list.Thepictures forGroup Vcould notbeincorporated inthebooklets, sothey were placed face down ontheSstables before theexperiment began. Theexperiment wasrunoffinthesame wayforallgroups, apart from the differences intheinterpolated activity. Two minutes were allowed forthestudy of
he primary listwithaviewtoreproducing it,incorrect order, later. Ssthenturned totheblank pagefor30sec.,during which theyweregiven instructions forthe
interpolated task. Three minutes were allotted totheinterpolated task; thentheSs turned tothefifth sheet and,after brief instructions, wrote asmany pairs of consonants fromthefirstlistastheyremembered during i mm.Finally, for Groups IandII,theSsturned tothesixth page andreproduced thesecondary list. Thetime-intervals allowed forprimary learning, forinterpolated task,andfor recall, were determined after preliminary experimentation. Scoring
Thefollowing system wasusedinscoring therecalls written bytheSs.Onecredit wasgiven ifaconsonant-pair wasreproduced correctly inthesame position which itoccupied intheprimary list,orifitwasreproduced correctly initsrelatively correct position (i.e.relative tothepairofconsonants which preceded it onthe list). One-half credit wasgiven ifapairwascorrectly reproduced inaposition one place removed fromitscorrect position (relative orabsolute). One-half credit was givenifone-half a pairwasreproduced initscorrect position. Nootherrecall received credit. Results
Table2.1isa summary oftheresults, showing thenumber ofSsin each
group, therange oftheindividual recall scores ineachgroup, theaverage
number ofconsonant-pairs recalled, andtheprobable errorsoftheaver-
ages.As onewouldexpect,Condition I, learninga secondlist of con-
sonants, results inpoorer retention thananyoftheotherinterpolated conditions. Conditions IIandIIIshowabout thesame average number of pairsrecalled. Thelossresulting fromeither common operation alone, or
20
E. J. Gibson & 1. J. Gibson
Table
2.1
Summary of Resuhs Group
No.Ss
I
II
III
IV
V
LL-LL
LL-LN
LL-CL
LL-CN
LL-P
22
25
30
28
27
Range Av.
P.E.av
09 3.7 –28
0.510 6.2
–36
0.510
6.3 –38
310
410
7.6 –.25
8.6 –21
commonmaterialalone,seemsto be about the same,i.e. similarityof
operation andsimilarity ofmaterial areapparently ofaboutequalimportance in determining decrements of retention.
In ConditionIV (LL-CN), whereneitheroperationnor materialis like
that in the first task, recallis better than it is in the two conditionsin whichone of these featuresis similar.This improvementis manifestin
boththeaverage recallandtherangeoftheindividual scores. Condition V (P),wheretheinterpolated taskwassimply looking at pictures, givesthe bestrecallof all5 conditions. Boththeaveragerecallandtherangeof the
individualscores show that there is an improvementeven over LL-CN,
whereneitheroperation normaterial werecommon to thefirstandsecond tasks.
Fig.2.1shows graphically thedifferent percentages ofthetotalpossible
recallwhichresultedfrom the 5 interpolatedtasks.It shouldbe remembered that the maximumpossiblerecallwas 10 consonant-pairs. Using
thesepercentages, wemaymakethecomparisons indicated asthepurpose of ourexperiment. Condition IV(LL-CN) wasdifferent fromtheprimary taskin both materialand operation,and 76%of the maximum recallwas
obtained by thisgroup.Condition III(LL-CL) waslikeIV,exceptthat the materialwas similarto that usedin the primarytask.Its recallwas 63%.Subtracting thisfrom76%wefinda difference of 13units.Hencewe
maysaythatsimilarity totheprimary taskinmaterial onlycauses a lossin
retention,in thissituation,of 13unitsof percentage. ConditionII (LL-LN) waslikeIV(LL-CN) in material,but in operationwassimilarto the primary
task.Theaveragerecallfor thisgroupwas62%.Subtracting from76% (recallforIV)wefinda difference of 14units.Hencein thiscondition we
maysaythatsimilarity totheprimary taskinoperation onlycauses a lossin retention of 14 units of percentage.
Now,if it wereassumedthat a taskis madeup of two actuallyin-
dependent andseparable components, material ontheonehand,andan operation ontheother,itmightthenbeexpected thatwithaninterpolated taskhaving bothoperation andmaterial similar to thefirsttask,theloss
Retention and the Interpolated Task
21
Figure2.1 Effectof different interpolated tasksonretention of a list of 10pairedconsonants . Reading fromleft to right, thefiveinterpolated tasksareLL( learning a second list of letters ), LN (learning a).listof numbers ), CL(cancelling letters ), CN(cancelling numbers ), andP (looking at pictures
03NIV .L80 llVJ3 ~ 3181SS0dlV .LO.L::fa .LN3J~3d
would be a sum of the lossescausedby these two independently. Since there is a loss of 13 with similarity of material and 14 with similarity of operation, the sum of the losses, in this case, might be supposedto be about 27. ActuallyI however, comparing Condition I (both material and operation in common) with Condition IV (neither in common) the loss is 39 units of percentage.The presumptionis that the effect of an interpolated task in which similarity of material and operation is combined is greater than the sum of the effectsof two taskseachsimilar in one of thesefeatures respectively.7 If this inferenceis correct, it would indicate that the material and the operation within a single task interact, each enhancing to some extent the effect of the other. Consequentlyone can be separatedout only with violence to the other, and in varying them independently, as we have done, the remaining common feature is no longer as similar as it would beif both remainedthesame.
In ConditionV (P) the interpolatedtask was differentfrom the first task in a variety of features . The percentagerecalled(86) was greater than in Condition IV (CN), where the interpolatedtask was different in operationandin material.Thegainin V ascomparedto IV suggests that
22
E. J. Gibson & J. J. Gibson
thereare otherfeaturesthan thesetwo with referenceto whichcompar-
isonsmaybemade, andthatoneormoreofthesefeatures, e.g.thework attitude, remained common totheprimary andinterpolated tasksinIV,but not
in V.
Theabovecomparisons canbemade, withatleastequallogic,byexam-
ining gains inretention asa function ofnon-resemblance oftheinterpolated task. Thegains are25unitswithadifference inmaterial only, 26unitswith a difference in operation only,and39unitswitha difference inboth.As before, thenon-additive character ofthelattergaincanbeexplained by supposing material andoperation to beinterdependent. Summary
Ourexperiment showsthatcomparison withreference to features of two
tasksis usefulin understanding therelationbetweensimilarity andretroactive
inhibition.
It hasbeendemonstrated that underthe conditionsof this experiment
theinterpolation ofa taskwhich issimilar to theprimary learning either inoperation orinmaterial results inpoorer retention thandoestheinterpolationof a tasksimilarin neither.In thissituation,the two featuresoperation andmaterial, seemto be aboutequallyimportantin theeffecton retention.
Operation andmaterial werechosen asthetwomostobvious features or characteristics by meansof whichtwotasksmightbe compared. That thesearenot the onlyfeatureson whicha comparison maybe basedis
suggested bythefactthatCondition V.which differed inotherrespects,
gavebetter retention thanCondition IVwhich differed inthetwofeatures of material and operation.
Finally, theoriginal suggestion thatfeatures ofa taskareinterdependent seems tobeupheld bytheprobability thatthesumofdecrements dueto operation andmaterial doesnotequal thelosscaused bysimilarity inboth. Notes
1.SeeE.S.Robinson, Thesimilarity factorinretroaction, thisJOURNAL, 39,1927,297 312.ForSkaggsearlier formulation, seeF. B.Skaggs. Further studies in retroactive inhibition,Psycho!. Monog.,34, 1925,(no. 161),160. 2. E. S. Robinson, op. cit.
3. T. A. Dreis,Two studiesin retroaction,J. Gen.Psycho!., 8, 1933,157172.
4.L.Harden, Aquantitative studyofthesimilarity factor inretroactive inhibition, 1.Gen.
Psycho!., 2,1929, 421432; andN.Y.Cheng, Retroactive effect anddegree ofsimilarity
J. Exper.Psycho!.,12, 1929, 444449.
5.F.C.Tolman, Purposive Behavior inAnimals andMen,1932,10f.
6.E.S.Robinson, Somefactors determining thedegreeofretroactive inhibition, Psychol. Monog.,28, 1920, (no. 128), 28.
Retention and the Interpolated Task
23
Sensory Generalization withVoluntary Reactions EleanorJ.Gibson
This experiment wasthefirst ofseveral undertaken formydissertation research at
Yale University. Mydissertation sponsor, Clark L.Hull, generously gave mehis notebooks toperuse over a long weekend, thinking I might find aproject that
appealed tomeamong anumber roughly outlined there. Hewas inthehabit of
writing a sortofdiary every Sunday morning, noting down events oftheweek justpast, especially ideas thathadoccurred tohim, including ones fornew experiments. Reading thenotebooks was fascinating, butI returned them with the conviction thatI hadtoformulate myown problem andthatI would continue in
thevein ofmymasters thesis, working onverbal learning and forgetting with human subjects. Hull greeted mydecision without much enthusiasm. IfIwanted towork with him, I hadtostay inhisterritory. Hetold metowrite upmy ideas asaproposal, andhewould consider them. Aprevious student ofHull, W.M.Lepley, had published amonograph onserial learning and forgetting based on conditioned reflex principles. Hull called this work to my attention thinking I mightdosomething similar. Lepley hadproposed thatrote learning might bethought ofasa series of conditioned responses, with each item intheseries serving asastimulus foralater
response, either a simultaneous conditioned response forthenearest item, oras intrusions, which would have tobesuppressed andwould give risetoinhibition ofdelay. Hull(1935) spunthisideaintoanelaborate deductive model and showed thatanumber oftypical phenomena ofrote learning could beaccounted for. Although theideas and reasoning were clever, they somehow didnotconvey afeeling ofwhat oneactually does ina learning situation. However, Hull was insistent ontheimportance ofextending analogies withconditioning toother learning situations. Two other conditioning phenomena, generalization anddifferentiation, appealed tomeasreasonable candidates, fortunately. Generalization
conditioned trace responses forlater ones. The latter would lead toanticipatory
Journal ofExperimental Psychology, 1939,24,237253.
26
E. J. Gibson
Theoretical contributio to the problem of learn hav rec b characterize by a new series of attemp at linkin the fac of co tionalleamin experime with the condit resp . The me o these systematic studies has been to use the empi law of the co -
Sensory Generalization withVoluntaryReactions 27
28
E. J. Gibson
This is a reaction time experime . That means that whe a ce signal occurs ,signal you are to react as quickly as you can apa response .The will be avibratory stimulu on the (lwith owe ,upp ,rig ,
Procedure The following instructions were given thesubject toread atthebegin ning oftheexperiment .
SensoryGeneralizationwith Voluntary Reactions
29
orleft) partofyourbackwhichwillbedemonstrated to youin aminute , and theresponse willconsist ofspeaking OUTintothisvoicekey.Whenever you feelthevibrator , sayOUTimmediately . Thenoiseof theapparatus starting will serve asa readysignal soyoucangetprepared to respond in a hurry . Youwillbegivensome practise trialsbefore yourreaction timeisrecorded . Youareto respond onlytothedemonstrated stimulus . Laterin theexperi menttherewill beotherextraneous stimuli . These stimuliwill be other vibrators located in different places onyourback . Youarenotto respond to anyexcept thedemonstrated stimulus onthe- partof yourback . Experiments onreaction timeshowthatyoucangofaster if youkeep your mindontheresponse youareto make , sokeepamotorattitude , thatis, be prepared to sayOUTtheinstant thisvibrator stimulates you. Whenthesubject finished reading theinstructions , thevibrator to whichhewas to respond wasdemonstrated to him.Thisvibrator wasalways atoneendof the series (extreme right,extreme left, top, orbottom ). Then50trialswiththestimulus to whichhehadbeen instructed to respond (the"practised " stimulus ) weregiven . Reactions weretaken atarateof aboutonein 20seconds . At thebeginning of a trial, theexperimenter started thedrum(thesoundof whichserved asa ready signal ), a variable interval of fromk to 21seconds intervened , thestimulus occurred , the subject responded , the drum was stopped , and the chronoscope reading taken bytheexperimenter . A practise series of 50 trialswasgivenfirstwith thedesignated stimulus . Then100moretrialsweregivenduringwhichthe3 "prohibited " stimuliwere interspersed atrandom intervals among trialswiththedesignated stimulus . Each of the3 prohibited stimulioccurred 9 timesduringthe100trials . Theorderof theprohibited stimuliwasvaried witheach subject according to a prearranged plansothatorderofoccurrence didnotfavoranyone stimulus when a groupof subjects wasconsidered together . Thisprecaution wastakenin case thereshould beatendency forfalseresponses to decrease withpractise .4A group of 9 subjects wasrequired to balance theorder . Eighteen subjects (GroupI) were runwiththevibrators placed vertically up anddowntheback(halfwith the practised stimulus atthetopoftheseries , theotherhalfwiththepractised stimulus atthebottom ); 18moresubjects (GroupII) wererunwiththevibrators arranged horizontally across theback(again halfpractised ateach extreme ). The subjects were all students at Yale University and were paidfor their serVIces .
Controls Following theexperiment proper , thesubjects weregiventwo briefseries oftests , onetocheck ontherelative intensity ofthestimuli , and theothertocheck ontheirlocalizability . Theintensity ofthevibrators was controlled mechanically in sofaraspossible , in thatthevibrators , their wiring , andmounting wereidentical ; asafurther precaution , each subject wasputthrough a series of judgments of comparative intensity of the stimuli , incase there should beatendency formore false responses tooccur to a stimulus of stronger intensity andthereby distortthecurve of fre-
Table .-Group 3 1 Perc ~ of Tim Jud Str tha St . ' A & ' . & . v ' .Nea Ne F Stim St ~ u lu ~ I~p A 22 % 49 % 3 % B 88 54 7 II %86 %6~
30
E. J. Gibson
a
Sensory Generalization with Voluntary Reactions
31
other point occu . Whe the vibr we dis up an d the back , more error of iden occ . Po 0 1 w confu 14 times out 108 cha ( 1 pe ) Ag , 3 0 n with point 2 and 3 . It is quit in kee wi th fa o s sensit that the vibra plac ver sho be ha to lo , since spatia thres are high whe stim are ap in a lo direct on skin than whe they are app tra . It is c howe , that any gene foun in Gro II , at lea , ca ha b due to mista local of stim . The main question of the experim was wheth or not gen was more apt to occur in proport as the prohi stim app imated spatially the practise one .This quest was ans by the
Figure 3 . 1 Frequency of response to vibrators as a function of proximity to practised point ( G I , roup vibrators placed vertically ) . Frequency of response in percentage is represe on the ordinate ,and stimulated points in order of proximity on the abscissa . .LN3:> ~ 3dNI3SNOdS3 ~ ::fOA :>N3nO3 ~ .::f
STIMULATEDPOINTS
Results
relative frequency of respons to prohib stimu .The res ar best presented graphica .In Fig .the 3.1 are prese com for
32
E. J. Gibson
l.N3:J 'd3dNI3SNOdS3'd ::IOA:JN3nO3'd::l
the 18 subjectswho had the seriesof vibrators placed up and down the back. The practised stimulus is designatedas 0, and the others are designated 1, 2, and 3 in order of proximity to it, regardlessof whether the 0 stimulus was at top or bottom. The 0 stimulus is almost invariably respondedto (98 percent). Falseresponsesare madeto the closestprohibited vibrator 25 percent of the time, to the next closest14 percent of the time, and to the farthest 9 percent of the time. In general, the curve shows a continuous gradient from point 0 to point 3, in accordancewith the expectation that a processanalogousto generalizationof conditioned responses occurswith voluntary responses .? The differencesbetween the frequencies at the four points have very satisfactory critical ratios (44.2 to 2.67), with the exception of the differencebetween point 2 and point 3 which has a critical ratio of only 1.48.8 The continuous downward trend, however, is unquestionable. Furthermore, when the data for the two sub-groups of 9 subjectseachare plotted separately, the resulting curvesareboth similar to the one above, showing a continuous slope from 0 to 3. The results for the 18 subjectswho had the vibrators placedacrossthe back are given in Fig. 3.2. Here the slope from point 0 to point 2 resembles 100
80
60
40
20
0
0
1 2 STIMULATED POINTS
3
Figure3.2 Frequency of responses to vibrators asa functionof proximityto practised point(GroupII, vibrators placed horizontally ).
Sensory Generalization with Voluntary Reactions
33
closelytheslopein Fig. 1, but theupturnat point3 is not typicalof the usualcurveof generalization . It mustbe remembered , however , thatthe vibratorshereweredistributedhorizontally , andpoints0 and 3 were consequently symmetrical . Perhaps thesymmetrical locationis responsible for thehigherdegree of generalization at point3 thanat point2. Anrep(1) reportsthatsymmetry is aneffective cause of generalization ; hefound,in fact(with dogsassubjects ), thatstimulation of thepointsymmetrical with the0 pointwasequallyaseffective asstimulation of the0 pointitself.The difference in thisgroupbetween points2 and3 hasa statistical reliability of only 1.16 timesits standarderror. But whenthe datafor the two subgroups areplottedseparately , they resemble the curvedrawnfrom combined data,showinganupturnat3. It seems unlikelythatthisisa mere chance resemblance . It isnotable , also,thatin Fig. 3.1, wheretherewereno symmetrical points , no suchupturnoccurred . It is interesting to compare percentages of generalization at point 1 in GroupsI and II keepingin mind the fact that stimuli0 and 1 were accurately localized by GroupII butnotinvariably by GroupI. It mightbe expected on common sense groundsthatlessaccurate localization should accompany a greaternumberof falseresponses , but suchis not thecase . ForGroupII, thefrequency at point1 is 36percent , whilein GroupI it is only 25 percent ; morefalseresponses occurred whenthevibratorswere easilyidentified . Generalization of responses , therefore , canhardlybeexplainedby theinabilityof thesubject to identifythecorrectstimulus . A seconddifference is that betweenresponse to stimulation at the 0 pointsthemselves in the two cases . Figure3.1 showsa slightlylower percentage of response at this point thandoesFig. 3.2 (98 percentas compared with 100percent ).9 Againfrequency of response waslowered whenabsolute localization waspoorer . Thisfactsuggests thatthedifficulty of discriminating points0 and1 in GroupI induceda stronginhibitory attitudein thesubjects - i.e., caused themto reactwith greater"caution ." If thiswasthecase , reactiontimeshouldbeincreased . An examination of thetimesreveals thatit was.Takingthefirst50practise trialsasa control , andcomparing with themall thetrialswith the0 stimulus aftertheprohi bitedstimulihadbeen introduced , anaverage increase in latencyis foundin thelattercaseof 63ms. in GroupI, and42ms. in GroupII. It is significant thatthegreateraverage increase in latencyoccurs in GroupI. Theinvariable increase in latencyafterintroduction of prohibited stimuli is interesting . It is related , for onething, to the difference alwaysfound betweensimpleanddiscrimination reactions in thereactiontimeexperi ment.Furthermore , it suggests theanalogous situation with Pavlov 's dogs whena previously differentiated stimulus is introduced amongtrialswith a conditioned stimulus . An inhibitoryprocess occurswhich'spreads ' or leaves anafter-effectservingto depress theensuing conditioned responses .
34
E. J. Gibson
It should bepointed outthatthepossible influence of fatigue hasnot beencontrolled in theabove comparison ofthefirst50reactions withlater reactions . It is mostimprobable , however , thattheincrease in latency is evenpartially dueto fatigue . Foronething,practise effects wouldnormally continue after50reactions andbalance anyfatigue ; furthermore , theeffects offatigue in thereaction timeexperiment areveryslight(see4, p. 46). Thereaction timeswereanalyzed to seewhether thegradient of frequency of response to thefourstimuliwasaccompanied by a correlative gradient of speed . Superficial consideration mightleadto theexpectation of a negative correlation between frequency andlatency , sincepositive evidence of suchacorrelation in thecase of simple conditioned responses isavailable (3, 9). Butsucha correlation hasneverbeendemonstrated for generalized responses in the conditioning situation . Furthermore , in the present situation two factors arepresent whichcutacross theexpected correlation - in fact,workin opposition to it. In thefirstplace , thereis voluntary inhibitionof long-latencyfalseresponses . Theeffectof such inhibitionwouldpresumably be to allowonlythefastest potential responses to occur . Sincethe subject doesnot inhibitresponses to the designated or practised stimulus whentheyareslow , noselection of fast responses wouldoccurforthisstimulus . Simply averaging thetimesof all responses madeto eachof thefourstimuliwould , therefore , showan artificially decreased latency of response to theprohibited stimuli ; whereas thesuperficial prediction wasthatlatency of theseresponses should be greater thanthoseto thedesignated stimulus , since frequency of response islower . Table3.2 shows theaverage latency of response to thefourpointsof stimulation in bothgroups . Theaverage latency at point0 is calculated fromall theresponses to thatstimulus aftertheintroduction of thefalse stimuli (it willberemembered thatreaction timesforthepreceding practise responses wereshorter ). Theresults showthattheaverage latencies byno means increase frompoint0 to point3, astheyshouldif a negative correlation between frequency andlatency exists ; onthecontrary , thereis a decrease in latency asthestimulus is displaced fromthepractised point. Ourassumption of voluntary inhibition of long-latency falseresponses is therefore probably correct .
Table 3 . 2 Average Latency in Ms ~vt "labc L.a~ll'-L 'yof VI .Response I.~ ""t'.,& .,-to -the --4 -Vibrators -~ - 1-Vibrator -,-. 2 -~ 7.1,--:3 Vibrator 0 Vibrator Vibra --Group I ( V ertical distribution ) 233 219 149 151 Group II ( H orizontal distribution ) 170 161 106 122 ~:uup 11 \1J.V ~I.LV.L\.U "'I-& ."..,_.& _& "/
SensoryGeneralizationwith Voluntary Reactions
35
The second complicating factor is the slight tendency for the subjects of Group I to confusepoints 0 and 1. It has already been pointed out that frequencyof responseat thesepoints was lower than in Group II, probably becauseof "caution" or a strong inhibitory attitude. It is as if the subject had to make a discrimination
reaction between
two confusable
stimuli ; and ,
asis well known to be the case,! 0 the reaction time lengthenedaccordingly. The average latency at points 0 and I for Group I may be seen to be abnormally high in comparisonwith the rest of the figures. The first of these two complicationsmay possibly be eliminated by the following procedure. For eachpoint, the times of only the fastest6 percent of the total number of possiblereactionsmay be averaged. The hypothesis is that the whole distribution of possible responsesshould be slightly faster, the greater the frequency of reaction produced by the stimulus (or the nearer the stimulus to the designated point). By taking the top 6 percent of the potential responses , we include only reactionswhich were fast enough to slip by (above the threshold of voluntary inhibition), and avoid the difficulty . There is no way of avoiding the influence of the secondcomplicating factor in presentingthe results.
8 o -
-
-
-
GROUP
I
GROUP
II
Figure 3.3
Average latency of the fastest 6 percent of responsesat the four stimulated points .
36
E. J. Gibson
Figure 3.3shows thecurves oflatency ofresponse to thefourstimuli
whenonlythefastest6 percent ofthepotential responses areincluded in
eachcase.CurvesforbothGroupsareplottedon the sameaxes.If there
werea perfectinverse relationship between frequency andlatency, the
curvesshouldshowa constantupwardslope.A glanceat themdoesnot indicatemuchtendencyfor an increasein latencyto accompany spatial
displacement fromthe0 point,asdoesdecreasing frequency. Butifpoints 0 and1forGroup I arediscounted (forthereasons givenabove), therewill be seento be a roughupwardtrend.A perfectslope,evenasidefrom the two discounted points,couldhardlybe expectedconsidering that6
percent ofthepotential reactions inthecaseofthethreeprohibited stimuli includes only10responses ateachpoint.Therefore, although it cannot be concluded that speedis correlated withfrequency underthe presentcircumstances, it maybe suggestedthat sucha trendis not entirelyabsent whenthecomplications introduced by difficult localization andvoluntary inhibition
are considered.
Discussion
Severalreferenceshavealreadybeenmadeto the reactiontimeliterature
pointing outparallels between thepresent experiment andthediscrimination reactiontime experiment.Someother factsfromthis sourceseemto
be relevant.Thefalse reaction,aneventgenerallyfoundbut so faraswe know never studied, would seem to be understandable in many cases as an
example of generalization. Withreference to somerecentattemptsto condition voluntaryresponses in a reactiontimeset-up,it hasbeensug-
gestedbyGibson (5)thatwhatweretakento beconditioned voluntary reactionsweremorelikelygeneralized reactions. Thefactthatgeneralized
responses havebeendemonstrated to occur ina voluntary situation lends plausibility to this suggestion.
Furthermore, phenomena reportedin studiesoftheeffectofa distracting
stimulus onspeedofreaction to a givenstimulus canbeinterpreted inthe lightof ourapproach. Evans(4)foundthatdistraction wasparticularly effective whenthedistracting stimulus belonged to thesamesensedepart-
mentas the stimulusto be respondedto. Thatis,if the distractingstimulus
wasanintermittent flashoflight,whilethedesignated stimulus wasa flash
oflightofdifferent intensity coming froma slightly different place, the
increasein reactiontimeswas greaterthan if the distractingstimulus had been a tone or a touch.Our data likewiseshow that reactiontime
increases significantly whenotherstimuliof the samedimension as the practised stimulus areintroduced, especially whenthenewstimulus isnot easilydiscriminated fromthe practised one.it maybe tentatively suggestedthatin suchsituations a typeof inhibition is setup comparable
SensoryGeneralization withVoluntaryReactions 37
totheinhibitory after-effects reported byPavlov (17,p.125) andbyAnrep (1)whena differentiated stimulus isintroduced amongtrialswitha condi-
tioned stimulus. Butsuchapossibility mustbeexplored inanexperiment
especially designedforthepurpose.
Theresultsof thisexperiment havea certainbearing on therelation
between conditioned andvoluntary responses. Incommon withtheexperi-
mentsspecifically directed atthestudyofthisrelationship, theresultsshow
bothsimilarities anddifferences. Generalization andthegradient ofgeneralization appearunderbothconditions. Buttheinstructions whichwere
necessarily introduced inthevoluntarysituation apparently produce a setorattitudeofcaution whichhastheeffectofshortening reaction times ofthefewresponses madeto thefalsestimuli. It is a question whether
generalized responses intheconditioning situation wouldshowanincrease
ora decrease. Itseems possible thatthelatencies ofgeneralized responses
wouldbelonger whilethediscrimination isbeingbuiltup,andshorter, asin thepresent situation, whenthefrequency of generalized responses has
beenloweredby differential reinforcement. Thatis,establishment ofdifferentiationwouldinhibitfalseresponsesin mostcases;the few whichoc-
curred mightbefasterthannormal responses, asinthepresent situation. Suchcasesmightbe producedby disinhibition, for instance.Butinves-
tigation, ratherthanspeculation, isobviously called forbythisproblem. Anewlineofexperimentation issuggested bytheprobability thatfalse reactions to a prohibited stimulus donotoccursimply because thesubject cannottellthedifference between thisstimulus anda designated one. Similarities of a superthreshold natureappearto playa part.Theexact natureoftherelationship between generalization andperceptual similarity andidentity isanunanswered problem, andonewhich should yieldto a technique combining psychophysical methods andmethods designed for the studyof conditioned responses. Summary
Subjects wereinstructed to respond verbally asquickly aspossible to a
designated vibratory stimulus, butnotto respond in thiswayto other stimuli. Practise wasgiveninresponding to thedesignated stimulus. Later othervibratory stimuli, distributed either longitudinally ortransversely on thesubjects back,wereintroduced, to seewhether generalized orfalse responses wouldoccurandwhether theywouldbe morefrequent, the nearerthe stimulus to thedesignated oneon the skin.Thefollowing conclusionsmay be drawn.
1. Frequency of response to theprohibited stimuli showed a gra-
dientofgeneralization analogous to thegradient foundin thecondi-
38
E. J. Gibson tioned response experiment, when the vibrators were distributed longitudinally. 2. When the vibrators were distributed transversely, an upturn at the end of the curve of response resulted. This upturn was probably causedby the symmetrical relations of the point stimulated and the designatedpoint. 3. Introduction of prohibited stimuli apparently producedan increase in latency of responseto the designatedstimulus. This increasewas particularly noticeable when one of the prohibited stimuli was not easily discriminablefrom the designatedstimulus. 4. The averagelatency of responseto the prohibited stimuli was less than that to the designated stimulus, probably becausethe longlatency falseresponseshad been inhibited. 5. When the differential effectsof voluntary inhibition were avoided, there was some indication that speedwas positively correlatedwith frequencyof response. 6. With referenceto the problem of the relation between conditioned and voluntary responses , the present results suggest that similarities exist, but that differencesdue to the instructions introduced in the voluntary situation are to be expected.
Notes
I.The writer wishes toexpress gratitude toProfessor Clark L.Hull forencouragemen and advice during theprogress ofthis research .The experiment isone ofaseries ofstudies presented totheFaculty oftheGraduate School ofYale University inpartial fulfillment oftherequirements forthedegree ofDoctor ofPhilosophy inPsychology . 2.This experiment was reported before theEastern Branch oftheAmerican Psychologic Association in1937 (6). 3.Generalization gradients have been demonstrated withthisstimulus dimension by Anrep (1)and byBass and Hull (2). 4.Such atendency didoccur .About 70percent ofallfalse responses came during thefirst half ofthese 100 trials . 5.Mvers gives thetwo -point intheregion ofthespine as5.4cmwhen the ." (16 ,)'-' - threshold stimuli are distributed longitudinally . 6.Thedesignation 0 isused torefer tothestimulated point which thesubject was instructed torespond to. The numbers 1, 2, and 3refer respectively tothevibrator nearest ,next ,and farthest from that point . 7.Forasimilar negatively accelerated curve ofgeneralization intheconditioned response situation see Hovland (10 ). 8.Reliabilities were calculated with theformula given byYule (22 ,p.269 )forthestandard error ofthedifference between twoproportions . 9.Although small , thedifference isstatistically reliable , itscritical ratio being 3.67. 10 .Henmon (8) found that thesmaller thedifference between twostimuli , thelonger the reaction time tothedifference .
SensoryGeneralization withVoluntaryReactions 39 References
1.Anrep, G.V.,Theirradiation ofconditioned reflexes, Proc. Roy. Soc., 1923, 94,Series B, 404425.
2.Bass, M.J.andHull, C.L.,Irradiation ofa tactile conditioned reflex inman, J.Comp. Psycho!.,1934, 17, 4765. 3.Campbell, A.A.andHilgard, E.R.,Individual differences inease ofconditioning,]. Exper. Psycho).,1936, 19, 561571. 4.Evans, J.E.,Theeffect ofdistraction onreaction-time, withspecial reference topractice andthetransfer oftraining, Arch. ofPsycho!., 1916, 37,pp.106. 5.Gibson, J.J.,Anoteontheconditioning ofvoluntary reactions, J.Exper. Psycho!., 1936, 19, 397399. 6.Gibson, E. J.,Sensory irradiation withvoluntary responses, Psycho!. Bull., 1937, 34, 5115 12. 7.Gibson, E.J.,ASystematic Application oftheConcepts ofGeneralization andDifferentiation to Verbal Learning, Dissertation, YaleUniversity, 1938.
8.Henmon, V.A.C.,TheTime ofPerception asa Measure ofDifferences inSensations, New York: Science Press, 1906.
9.Hilgard, E.R.andMarquis, D.G.,Acquisition, extinction, andretention ofconditioned lidresponses to lightindogs,J. Comp. Psycho!., 1935,19,2958.
10.Hovland, C.I.,Thegeneralization ofconditioned responses: I.Thesensory generalization ofconditioned responses withvarying frequencies of[one,J.Gen. Psycho!., 1937, 17, 125148.
11.Hull, C.L.,Thegoalgradient hypothesis andmazelearning, Psycho!. Rev., 1932, 39, 2543. 12.
, Theconcept ofthehabit-family hierarchy andmazelearning, Psycho!. Rev., 134152.
1934, 41, 3352; 13.
491516.
, Theconflicting psychologies oflearningawayout,Psycho!. Rev., 1935, 42,
14.Lepley, W.M.,Atheory ofserial learning andforgetting based uponconditioned reflex principles,Psycho!. Rev.,1932,39, 279288.
15.Marquis, D.G.andPorter, J.M.,Differential factors inconditioned voluntary and conditioned involuntary responses. Psycho!. Bull., 1937, 34,p.772. 16.Myers, C.S.,AText-Book ofExperimental Psychology, Cambridge: University Press, 1928, pp. xiv + 344.
17.Pavlov, I. P.,Conditioned Reflexes. Translated andedited byG.V.Anrep, Oxford UniversityPress,1927,pp. xv + 430. 18.Peak, H.andDeese, L.,Experimental extinction ofverbal material, J.Exper. Psycho!., 1937, 20, 244261. 19.Spence, K.W.,Thenature ofdiscrimination learning inanimals, Psychol. Rev., 1936, 43, 427449. 20.
, Thedifferential response inanimals tostimuli varying within a single dimen21.Stephens, J.M.,Theconditioned reflex astheexplanation ofhabitformation. 111. The sion, Psycho!.Rev.,1937, 44, 430444.
operation oftwohigher orderreactions inclose succession,]. Exper. Psycho!., 1936, 19, 7790.
22.Yule, G.Udny, AnIntroduction totheTheory ofStatistics, London: Charles Griffin andCo., Ltd.,1922,pp. xv + 415.
A Systematic Application oftheConcepts of Generalization andDifferentiation to Verbal
Learning EleanorJ. Gibson
The centerpiece ofmydissertation wasthisattempt tousetheconcepts ofgeneralization anddifferentiation todeduce anumber ofphenomena characteristically found inpaired associate learning studies ofthetime, such astransfer, interference, andretroactive inhibition. Toplease Hull, I used concepts indicating phenomena
found inconditioned response learning situations, butasI thought ofthem, the concepts actually referred toperceptual aspects ofthelearning process. However,
onedidnt speak much ofperception inpublic inthose days, sothere isa kind of
surreptitious meaning smuggled in.I wanted toemphasize structural features of items tobelearned, rather thansheer association. Perceptual differentiation, preliminary toassociation with another unit, whatever itmight be,was promising as theprocess thatmust beinvolved. Theterm predifferen ia ion evolved inthe process ofconstructing themodel andpulling outdeductions. Itbecame apopular term andwasthefocus ofconsiderable research (forexample, Arnoult 1953).
Inpreparing thispaper, I tried tofollow Hullslead(Hull 1935) presentingdefinitions, assumptions (called postulates, inemulation ofa geometrical proof), andthen deductions thatcould betested experimentally, with thecomplete argument given. Itnowseems tomesomewhat labored andpretentious, asdoes Hulls work. Spelling everything out, with explicit definitions andpredictions, has itsmerits. Even so,incomplete definitions thatallow unvoiced assumptions tocreep incanoccur, although itisremarkable how often they gounremarked. Ifind afew
nowinmyownmodel andwonder whynoonepointed themoutat thetime.
This miniature system, aswell asHulls miniature andmore major ones, are longsince passØ formostpsychologists, butit seems thatthatdoesnt include
everyone who isconcerned with models ofbehavior orcognition. Afewyears ago Igave a seminar attheUniversity ofPennsylvania whose participants included
promotors ofArtificial Intelligence androboticists. I askedthemto read,as
historical background, thepaper that follows andasample ofHulls work. Tomy
astonishment, they considered ita revelation, exciting andpromising ! Psychological Review, 1940,47, 196229.
42
E. J. Gibson
Introduction
Arecent trendinexperimental andtheoretical psychology hasbeenthe
attempt tosystematize thefacts ofagiven field onthebasis ofempirical principles derived fromstudyoftheconditioned response (14,15,21,39, 16).1 Itisreasonable toexpect thatthesame general lawsholdfromone learning tasktoanother, insofarasthesituations aresimilar, andit is possible thatthegreater thesimplicity ofthelearning, thebetter arethe chances oflaying baremechanisms which operate inalllearning. Itisnecessary, however, toguard against thechance thattheprinciples chosen forsystematic exploitation areartifacts of theexperimental situationfromwhich theywerederived, andhavenoimportance asgeneral
characteristics oflearning. Hullhassuggested a procedure fortesting the valueof a theorybuilton suchconcepts (14).Theessentials of such a testarethedevelopment ofa clear-cut hypothesis, exploration ofthe hypothesis forallitsderivatives, andexperimental testofsuchderivatives ashavereceived nodirectsubstantiation or disqualification. Thetheory muststandorfallwiththeexperimental evidence fororagainst thepredictions which it makes.
Thefollowing account isanattempt todevelop atheory relating various factsofverbal learning (such astransfer, interference, certain intra-list effects andretroactive inhibition), totwoexperimentally defined characteristics of theconditioned responsegeneralization anddifferential inhibition. The
theory willbeexplored fordeductions ofknown facts aswell asforimplications yettobetested, since ithasbeen undertaken principally inthehope
ofsystematizing many ofthefactsalready known. Thattransfer, interference, andretroactive inhibition arerelated haslongbeensuspected, asthe so-called transfertheoryofretroactive inhibition implies (2,p. 427if.).
Thehypothesis tobepresented islikewise classifiable asatransfertheory, butit aimstomake asexplicit aspossible themechanisms assumed tobe responsible forthetransfer, which isitself aphenomenon aspoorly understood as retroactive inhibition. Thehypothesis willfirstbestated inabrief, informal fashion. Then, for thebenefitofthereaderwhowishesto makea morescrupulous examina-
tion,thetermsusedwillbeexplained, assumptions stated, andarguments for variouspredictionspresentedin detail. Statementof the Hypothesis
Thehypothesis asserts thata major necessity ofverbal learning is the establishment ofdiscrimination among theitemstobelearned, andthatthis
process ofdiscriminating isactually a fundamental partofwhatiscalled generally thelearning process. If nodiscrimination between theitems
VerbalLearning 43
already exists, thentheearly partofthelearning process willseeanincrease discrimination. Learning timeshould beata maximum inthiscase. Ifsuch discrimination already exists, learning timeshould beata minimum. Posi-
inthetendency toconfuse theitems, followed bythedevelopment of
tivetransfer willoccurin situations wherethenatureofa second task
permits discrimination acquired inaprevious task tobebeneficial. Negative transfer willoccur when generalization witha previous taskoccurs, but where thesituation issuch thatdiscrimination between some aspect ofthe twotasksthemselves isrequired, aswellaslearning ofthesecond. Retroactive inhibition willoccur, similarly, ifa second taskgeneralizes withone
already learned, andifthesituation issuchthatdiscrimination between
someaspectof the twotasksmustbe produced beforethe first:canbe
recalledadequately.
Thereader familiar withthefacts ofconditioned responses willseeimtion anddifferentiation. 2 Thehypothesis will bestated asitapplies toverbal learning bythepaired associates method. Itisassumed thatwhen a listis being learned bythismethod, generalization mayoccur between thevariousstimulus items, sothata response learned toonetendstooccur, asa response toother stimulus items inthelistalso. Figure 4.1represents alist
mediately thepossibility ofstating these assumptions interms ofgeneraliza-
inwhich generalization isoccurring. Thedotted lines represent generalizationtendencies andthesolid arrows represent connections with theright responses, which aretobelearned. Insuch a listasthis,where thegeneralizing stimulus items allhave different responses, those responses occurringortending tooccur byvirtue ofgeneralization willblock theright
responses inproportion tothestrength ofthetendency togeneralization. Inordertoreduce thestrength ofthegeneralizing tendencies (toincrease
thedifferentiation), differential reinforcement through practice mustbe applied. Itisassumed, thatvarying degrees ofgeneralization mayoccur between oneitem andothers inalist, andconsequently thatlists may vary
asto theaverage strength ofgeneralization occurring in thelist.The Sa \ >-__ .,
\
Sb (
,/ /%.
\)(/ Rb /
Figure 4.1
K
\
\
44
F. J. Gibson
List1
List2
List1 R
Sa
Ra
Sc
* R
,.
Sa
Rb Sb
b
Sd .
_-ł .Rd
Sb ..-
łRb
Figure 4.2
necessity forgreater differentiation willthenserveto makethelearning moredifficult inlistswherea highdegreeofgeneralization occursthanin oneswherelessgeneralization occurs; andit willalsobe responsible for
poorer retention, since thedifferentiation, especially withlower degrees of
learning, decreases overa periodof time,allowing spontaneous recovery ofthegeneralization. Iftheabovehypothesis iscorrect, thegeneralization in eithercaseshouldbe apparentin actualconfusion of responses.
Furthermore, in anytwo-list situations suchas theonesusedto study
interferenceand retroactiveinhibition,it is assumedthat inter-listgen-
eralization mayoccur. Whena firstlistisfollowed bya second onewhich
includes stimulus itemssimilar to onesin the first,stimulus itemsin the secondlist willtendto produceresponsesfromthe first;andmembers likewise whenlist 1 is againpresentedforrecallandrelearning, response
fromlist2 willtendto recur.Figure4.2represents responses fromlist1
tending tooccur bygeneralization inlist2,andonesfromlist2 occurring bygeneralization when list1isagain tested. Ineachcase, thegeneralizin tendencies willblocktherightexcitatory tendencies inproportion to the
strength ofthegeneralization. Again, thegeneralization should beevident inactualconfusion ofresponsesinthiscase,errorsofreversionto the wrong list.
Thedegree oflearning ofanylistcannot bepredicted fromthenumber ofpractice repetitions alone, according tothishypothesis, since theamount ofdifferential reinforcement orpractice required to reducegeneralization to
a givenstrength increases withthedegree ofgeneralization between stimulusitems.Theextentof differentiation achieved as a resultof a given amountofdifferential reinforcement willvary,then,withthisfactor.More-
over,theextentofdifferentiation mustbecalculated alsointermsoftime elapsing sincedifferential reinforcement hasbeendiscontinued, sincespon-
taneousrecovery willeventually occur.Pavlovhasshownthatsponta-
neousrecovery occurs moreslowly, thefarther differential inhibition has beencarried 33,p. 123).Temporal intervals between learning andrecall thusassume importance inthelightofthehypothesis, andtogether with
VerbalLearning 45 Sa
St\V1 ,4 a P\ Rb St i I, I \
t \RC Rd
Figure 4.3
degree oflearning, mustbetaken intoaccount inpredicting inter-list Thehypothesis hasbeenstated interms oflistslearned bythemethod ofpaired associates, butthesame logic willapply throughout to lists learned byanymethod which allows astimulus-response analysis. Figure 4.3represents a listbeing learned bytheanticipation method. Itema functions onlyasa stimulus andmust produce thenextitem, Ra,asits response. Butwhenthisitemactually appears, it functions as a stimulus. Thedottedlinesrepresent generalizing tendencies within thelist.The hypothesis cannot besoeasily applied toother verbal learning situations, effects.
sinceinmostofthemthesubjects performance isnotsocontrolled asto
allow a detailed S-Ranalysis. Such would bethecasewiththemethod of
complete presentation, forinstance. However, there seems nogood reason
to suppose thatthehypothesis cannot holdwithsucha method. In the sections whichfollow, evidence willbequoted without comment whenit
hasbeenobtained bythepaired associates oranticipation method, but
where any method hasbeen employed, thisfactwill bespecifically referred to. other Theroleoftheresponses indetermining easeoflearning a single list orintransfer situations isnotspecifically delineated bythepresent hypothesis. Itisconceivable thatsomething analogous togeneralization of stimulus items might occur among response items, especially since every
response isina sense a stimulus aswell. Mullers substitution hypothesis develops thispossibility (30). According toMuller, activesubstitution wassaidtooccur when ideaawasconnected withideab,andideaA,which
wassimilar toa,alsoproduced b.This description follows thepattern of
sensory generalization whichhasbeendescribed here.ButMulleralso
described passive substitution; Bmight besubstituted forb,when Bwas
similar tobbutnotassociated with a.This, interms ofthepresent hypoth-
46
E. J. Gibson
esis,wouldamount to generalization of response items,if it occurred. Confusion errorswouldresultineithercase,anda priorimightbe dueto
eithertypeofgeneralization. ButrecentworkofThorndike (41)suggests thatthestimulus itemsof a listarechiefly effective in producing such errors.Hefoundthatstimulus members of a verylonglistevokedre-
sponses made toother stimuli like them nearly twice asoften astheyevoked connectedwithunlikestimuli.Buta stimulusmemberevokeda responses
word which was like the responseconnected with it little or no oftener than it evoked unlike responses.
Anexperiment byMcGeoch andMcGeoch (28)seems alsosignificant in
thisconnection. In a retroactive inhibition experiment, stimulus members ofcorresponding pairedassociates ina listI anda list2 weresynonymous; or response members weresynonymous; or bothmembers weresynonymous;or neitherwere.Whenthe stimulusmembers weresynonymous, therewasa consistent increase inretroactive inhibition overthecondition whereneithermember wassynonymous, butthiswasnotthecasewhen response
membersweresynonymous. Furthermore, whenboth members
weresynonymous, therewasnoconsistent increase inretroactive inhibitionascompared withthecondition wherestimulus members onlywere synonymous. Onthebasisoftheseresults, thepresent hypothesis will confineitselfto theexploitation oftheconceptof sensorygeneralization as originally stated. Relation to Other Systems
Therehasbeeninthepastat leastoneratherambitious attemptto systematizesomeof the factswhicharebeingconsidered, andtherearecon-
temporary theories which applyto thesamerealm. Theearlier attempt referred to is thatof G.E.Muller andhisstudents. MullerandPilzecker (30), besides presenting anenormous massofexperimental databased on a thorough-going exploitation oftheTreffermethode, posited a number of kindsof inhibition (e.g.,associative inhibition, retroactive inhibition) to
explain theirresults. Their substitution hypothesis, which rosefroman
exhaustive analysis oferrors, ina senseforeshadows thepresent theory. Mullerstheories, naturally, roseoutof hisresults; andthoughhistermi-
nology haslefta lasting impression ontheliterature, it seems antiquate today, because thefoundation laidbytheseearlier investigations makes itpossible todevelop a morecomprehensive theory which mayevensug-
gestthepsychological processes ormechanisms underlying histypes of inhibition.
Asystematic attempt toexplain various factsofmemory andlearning is
being made at present byGestalt psychologists, andhasgiven riseto experimental work, notably byKhlerandhisstudents (20, 33,32,35). The
VerbalLearning 47
essence ofthetheory isthattheGestalt laws ofspatial organization hold organization, theGestalt psychologists predict such effects asdifficulty in learning homogeneous series, retroactive inhibition, andproactive inhibition.Forinstance, thelawofsimilarity inspatial organization isassumed toholdformemory, andisdemonstrated ina series ofexperiments by
alsointhefield ofmemory; byanalogy withtheir laws ofperceptual
vonRestorff (35),inwhich retention ofhomogeneous seriesofitemsis
compared with retention ofheterogeneous series. Anitem inaheteroge-
neousseriesis saidto beina betterposition to beretained thanthesame
iteminanentirely homogeneous series, since bythelawofsimilarity there
isaggregation ofthetraces ofthehomogeneous items, thereby causing any single item toloseitsidentity. Also, anisolated item inoneseries may loseitssuperiority ifa second series follows which iscomposed ofitems similar toit(retroactive inhibition), while anisolated iteminasecond series willhavenoadvantage iftheseries preceding it ismade upofsimilar
items(proactive inhibition). Thetheoryis characterized by Koffka as a dynamic tracetheoryandhasbeenelaborated inhisbook(19).It could probably berestated intermsofdiscrimination, andinthissenseissimilar to the theorypresented in thispaper.Thechiefdifference liesin the concepts on which the two theories are based, and the methods bywhich they are developed. Athirdtheory, advanced byJames (17)toexplain interference, restslike
thepresent theory onPavlovian concepts. Histheory differs considerably
fromthepresent one,however, inthatthekeyconcept isexternal inhibi-
tion,ratherthandifferential inhibition. Thetheoryis extended to cover retroactive inhibition, butitisunfortunately notdeveloped totheextent of
predicting experimental results orspecific tests. Noexperimental work seems tohavefollowed it.Thetheory could consistently supplement the onesuggestedin thispaper. Thepresent-day associationists haveassembled a number of laws
which bear some relation tothecentral notions ofthepresent hypothesis. InRobinsons book (36), a Law ofAssimilation isincluded which isvery similar tothenotion ofgeneralization. Itisstated asfollows: Whenever an associative connection issoestablished thatanactivity, A,becomes capable ofinstigating anactivity, B,activities other than Aalsoundergo anincrease ordecrease intheircapacity toinstigate B. Thelawisextended tocover assimilation ofinstigated processes aswell. TheLaw ofAcquaintance, suggested inthesame work, issuperficially similar toaprinciple ofdiscrimination.Butthelawapparently means byacquaintance mere isolated repetitionofagiven item; acquaintance isrelevant tothepresent theory only insofarasitproduces discrimination from other items. Mere acquaintance doesnotseem toincrease efficiency oflearning (Waters, 42).Much closer tothenotion ofdiscriminability ofstimulus items isThorndikes Law of
48
E. J. Gibson
Identifiability , which states that IIconnections are easy to form in proportion as the situation is identifiable , distinguishable from others . . . " (40 , p . 87 ). In general , these laws seem to name results , rather than to form a coherent and interrelated system .
Explanation of Terms The following explained
terms are crucial to the hypothesis
in some detail . Illustrations
and are consequently
will be presented when necessary .
Generalization : thetendency fora response Ralearned to 5ato occur when Sb(withwhich it hasnotbeen previously associated ) is presented .3 A generalization gradient is saidto beformed when a number ofstimulus i~ems show decreasing degrees ofgeneralizatio witha given standard stimulus . Thehypothesis need make noassumption asto thetypeof stimulus continuum which willyielda generalization gradient , butit isconsistent withit to suppose that such agradient willbeyielded byagroup ofstimuli which canbe arranged along anydimension orscale withrespect tothepresence of some discriminable quality oraspect - inother words , stimuli which would beconsidered to varyin degree of similarity . Experiment evidence proves thatnotonlysimple dimensions such aspitchand intensity (11,12)yieldgeneralization gradients . Yum(44)found that when nonsense syllables , words , orvisual patterns were stimuli ina learning series , different butsimilar stimulus items would in a test series elicittheresponses learned , thestrength ofthetendency vary ingwiththedegree of similarity between testitemandoriginal stimulus item ; andGulliksen (10)found thatsimilarity oftestfigures to a training figure correlated significantly withtendency to give thetraining figure 'sresponse .Razran (34)has recently demonstrat generalization onthebasis ofsynonymity ofwords . 2. Multiple generalization : thetendency forresponses which arebeing learned to members ofa series of stimulus items to occur asgen eralized responses tomembers other than their"right " associate . The situation employed inYum 'sexperiment (44) wassuch astoyield multiple generalization . Gulliksen 'sexperiment (10)provides atransi tionbetween simple generalization andmultiple generalization , since a choice between twoantagonistic responses , each of which had been learned toastandard stimulus , was possible . 3. Right response : theresponse paired withaparticular stimulus item inalistpresented bythepaired associates method , ortherespons immediately following aparticular stimulus iteminalistpresented by theanticipation method , 1.
Verbal Learning
49
4. Learning : a list is said to be completely learned when the right response item is given in each case upon presentation of the stimulus
item. A given degreeof learningor criterion of learning meansthat a certainpercentageof right responsesis reproducedupon presentation
of the stimulus items .
5. Differentiation : a progressive decreasein generalizationas a result of reinforced practice with Sa-... Ra and unreinforced presentation of Sb.
6. Reinforcement : a processwhich occursduring verbal learning when a subjectseesa responseas he anticipatedit and thinks "that's right ." Differential reinforcement designatesthe situation wherein a right response is reinforced and a generalized response is not reinforced .
7. Excitatorytendency : tendency for a particular stimulus to evoke a particular responsein a degreegreater than zero. 8. Meaning : a characteristic of a verbal or visual item which serves to differentiate
it from other items .
Statementof Postulates
A number of postulates are basic to the hypothesis and should be made evident at the outset. In some casestheseprinciples have been empirically demonstratedin the conditioned responsesituation; in other casesthey are incapable of empirical demonstration, and must rest on confirmation of the relationshipswhich they predict in combination. 1. When lists are learned by a paired associates or anticipation meth od , S-R connections
are set up between
certain of the items (between
members of a pair in paired associates, and one and the next in the anticipation method ).
2. If there is a right excitatory tendency and also a generalizedone between the same5 and R, the resultant excitatory tendency will be stronger than either one alone; if a right excitatory tendency and a generalized one are aroused by the same 5 but lead to different R's,
they will interfere, the weaker tending to block the stronger in proportion to the strength of the weaker. 3. Stimulus items which generalizewhen presentedfor learning in a single list will also do so when they are presented in the form of two
lists, and with the samerelativedegreeof generalization. 4. Generalizationwill increaseto a maximum or peakduring the early stages of practice with a list , after which it will decrease as practice is continued . (This assumption has been confirmed by the author in an experiment to be reported , 7.)
50
E. J. Gibson 5 . In list -learning , all decrease in generalization reinforcement .4 6 . The amount tendency
of reinforcement
to a given strength
required
is due to differential
to reduce a generalizing
will increase with increasing
strength
of
the generalizing tendency . (This postulate is an " empirical " one , in the sense that it has often been demonstrated in the conditioned response situation , 33 .) 7. After the cessation of practice , differentiation
will decrease over a
period of time , leading to an increase (" spontaneous recovery " ) of generalization . (This , again , is an " empirical " postulate , 13 , 33 .) 8 . Spontaneous recovery will occur more slowly , the more differ ential reinforcement has been applied . (Demonstrated in the condi tioned response situation , 33 .) 9 . If differentiation has been set up among items , it will be easier to differentiate
a number
of stimulus
them again later , even though
they are paired with different overt responses than those learned when the original differentiation was set up . (This assumption might be called " transfer of differentiation ." ) 10 . If differentiation has been set up among a number of stimulus items , there will be less tendency for them to generalize with new stimulus items or for new stimulus items to generalize with them , the decrease in generalization being proportional to the amount of differ ential reinforcement given . (Analogous to this are the cases reported in Pavlov (33 ), of dogs who , having once learned to differentiate a circle from an ellipse , found it easier to differentiate
between
the circle
and other ellipses of different ratio . A common sense analogy would be the case of a child who , after at first generalizing all the animals he sees, learns that cats are not dogs ; after this , he does not have equal trouble
in learning
that rabbits are not dogs . He would
see every animal in the world animals .)
in order to differentiate
not have to
dogs from other
Some Propositions that Follow from the Hypothesis A number of propositions
which may be predicted
from the hypothesis
just
presented will be stated . These will not be formally derived in the strictest logical sense, but for each one the argument will be indicated as clearly as possible , with reference to the definitions and postulates on which it de pends . Examples will follow the argument when any clarification seems necessary , and evidence for or against the proposition will be reported . In some cases, the propositions are familiar facts of verbal learning . For the convenience of the reader , the propositions will be given titles and grouped under traditional
headings .
VerbalLearning
51
Similarity
Similarity willbeconsidered, inthepresent instance, asthatrelationship
betweenstimulusitemswhichcanbe indicatedandmeasuredin termsof theirtendency togeneralize. A priori,thisis a reasonable criterion of similarity, sincea groupof stimuliamongwhichgeneralization occurswould,
according to usualstandards, beconsidered similar (e.g.,a seriesoftones,
or ofspotsalongthefore-leg). Yum(44)foundthatjudgesratings of perceptual similarity ofvisual patterns andsynonyms to theirrespective standards coincided withthedegree towhich theforms orsynonyms were
capable ofproducing a response originally learnedto thestandard. Furthermore,it seemsobviousthata common wayofspeaking ofsimilarity is in termsofdiscriminability, or tendency notto generalize.
Acomplication interminology isintroduced bythefactthatgeneralizationwillvarywithotherfactors thanoriginalgeneralization tendency. Ashasalready beenpointed out,itwillvaryalsowithdegree oflearning andwiththetimeintervalsincelearning wasleftoff.Thesetwofactorsare
ina sense secondary determiners ofdegree ofgeneralization, andoriginal generalization tendency willbeusedtherefore to denote thedegree of generalization apparent before differentiation hasbeensetupthrough
differential reinforcement. Undertheheading ofsimilarity,then,willbe
included propositions which haveastheirmajorvariable thedegree of
originalgeneralizationof stimulusitems.
I. Intra-ListTransferand ConceptFormation
Ifmultiple generalization occurs during thelearning ofa list,andifthelist isconstituted sothatthegeneralizing stimulus items arepaired withthe
sameresponses, learning shouldbeeasierin proportion to thedegree of
generalization.
Theargument forthisproposition would beasfollows: Suppose thata lististobelearned inwhich multiple generalization mayoccur, andsupposethatthosestimulus itemswhich aresimilar (i.e.,generalize toa high
degree)actuallyhavethe sameresponses to be learnedto them.The
generalizing excitatory tendency andtherightexcitatory tendency, inthis case,willcoincide. According topostulate 2,ifthereisa rightexcitatory tendency anda generalized onebetweenthesameS andR,theresultant
excitatory tendency willbestronger thaneither onealone. Now,according
to the definition of generalization, generalization mayexistin different
degrees, sothattheaggregate generalization ina listmayvaryfromlow tohigh.Then, inthepresent case,thehigher thedegree ofgeneralization
inthelist,thestronger willbetheresultant excitatory tendencies afterthe sameamountofpractice, andtheeasierwillbethelearning.
52
E. J. Gibson
Sc"' .... ~ Rc ....... ............... >!! !!p J.o aaJ6aa
The situation describedis of coursethe one which resultsin "retroactive inhibition." It is an acceptedfact that retroactive inhibition is a function of similarity of original and interpolated tasks (see Britt, 2, p. 389 ff.). However, degreeof similarity hasusually beenestimatedon an a priori basis. An experiment by the writer (8) has checkedthe prediction that a generalization gradient will correspondwith a gradient of retroactive inhibition. Degreeof Learning X. Interference and Degreeof Learning Difficulty of learninga secondlist will vary with the degreeof learningof the precedinglist, increasingto a peak and then decreasing as degreeof learningof list 1 increases .
Supposethat two lists, the items of which tend to generalizewith one
another , are to be learned in immediate succession. The tendency for items
in the first list to generalizewill increaseto a maximum during the early stages of practice of the list , but from this point on, the tendency to
generalizewill decrease (postulate 4). Now a secondlist will be harder to
learnasthe strengthof the tendencyfor itemsof list 1 to generalize with it increases(proposition VII ), so that difficulty of learning list 2 should
A
B Degreeof learning of list 1
Figure4.9
c
59
Verbal Learning increase tion
with
. But
begins list
, and 1,
list
the
2 , the
A
will
theoretical
of
list
the
list . In
the
due
to
curve
is
erated
to
check
of
this
Retroactive
and
The or
second been
was
of
equal
list
- list
by
the
to
B and of
the
, since
to
the
later
from
the
increasing decrease
B to
,
C . The accel
-
' s ( 7 ) demon
during
the
that on
a positively author
the
-
progress
experiments
in the
stimulus
members
generalizing
of
degree
practice
due
fall
in
decrease
assumes of
B , and
up
amount
of
prediction
to
-
with
2 will
2 as a function
generalization
than
of
field in
. Melton
this
of
learning
above prediction
since
point
of of
should ; and had
McQueen
this
- Irwin
, are
to
be
to be tested of
relearning
learning list
. of
1 should
generalization
2 increases
situation
been
. The
occur there
be
, argu
of
not
drawn
-
( 29 ) definitely
. But
find
time in
(23 ) has
with of the
-
any con -
that
interpolated
learning
earlier
on
indication
increase of
come
, and
. McGeoch is some
degree
degree
will
learned
cannot
did the
then
X .
1 has
retroaction
items
maximum list
and .
1 is then
if degree
and
curve
upheld
at which
of list , and
in
list
theoretical
been
point
and
the
a maximum 2 increases
stimulus
constant
to
well
to list
recall
generalization
have ,
that
in proposition
predicted of
increase
interpolated
in recalling
2 up
how
Learning
generalizing
degree
an accurate
the
1 will
of an
, and
given
on
higher
list
of Interpolated
1 is kept
list
maximum
half
beyond
a
inhibition
as that
might
carried
learning
intra
of Degree
as the
same
first
half
A
set
the
list
difficulty A
from
to
cessation
in
from
is available
for
of
inflection
the
after
been generalize
learning
list
experiment
lists , having
practice
, so that
case , the finned
for
succession
depending
intervals
rise
, rather
, retroactive
point
later
soon
an
of learning
two
decrease
is the
a sharp
prediction
learning
with
then
ment
learning
rise
to
proportional of
generaliza
, differentiation
has
members
, is represented
C , since
inhibition
that
of
its
Fig . 4 .9 . The
the
point
differentiation
increase
by
sort
as degree
2 is varied
increase
early
identical
in immediate
If degree list
, the
as a Function
decrease Suppose
in in
maximum
.
XI . Retroaction
learned
difficulty
is begun
this
used
if
of this
for
list
B to of
-
point
1 passes
being
plotted
show
from
( 18 , 3 ) have the two lists
apply
differentiation
strated a curve of learning . No
of
1 is
curve
drawn
drop
the
list
tendency
is represented
increasing
to
of
generalization
second
generalization
1 up
. Therefore , the difficulty list 1 is carried further .
curve of
will
less
in
list
learning
10
given in
learning
first
of
be
decrease
learning
of
stage
postulate
there
reinforcement as differentiation
of
practice
as the
the
learning degree lists
of
1 and
inflection
2 .
60
E. J. Gibson
XII.Retroaction as a Functionof Degreeof OriginalLearning
Retroactive inhibition fora listI willdecrease asdegree oforiginal learning ofthelistisincreased beyond thepointofmaximum generalization.
Suppose thattwolists,having generalizing stimulus items, areto be
learnedin immediate succession andthatrecalloflist1 is thento be tested;
degree oflearning oflist2 istobekeptconstant while degree oflearning oflist1 isvaried.Now,theinitialincrease ingeneralization aslist1 is given
morepractice willcause nocorresponding increase inretroactive inhibition in thissituation,becausethe effective generalization causedby thisincrease willbe fromlist 1 to list 2 ratherthan fromlist 2 to list 1 (i.e.,wrong
responses willtendincreasingly tooccur inlist2,butlist1willnotbecome moresusceptible togeneralization ofresponses fromlist2,because degree
oflearning inthatlistremains constant). Ontheotherhand,differentiation in list 1 willincreasewithdegreeof learningafterthe peakof generalization has beenreached(postulate4);and if differentiation has been set up
among a number ofstimulus items(inlist1,here), therewillbelesstendency fornewstimulus items (from list2,here)togeneralize withthem, the
decrease in generalization beingproportional to theamountof reinforce-
mentgiven(postulate 10).Since alsoretroactive inhibition isproportional to thestrength ofthetendency formembers oflist2 to generalize with members oflist1 (proposition LX), it follows thatretroactive inhibition will
decrease as degreeof learning of list1 is carried beyondthepeakof generalization.
McGeoch(22)has shownthat degreeof retroactiveinhibitionvaries
inversely withdegreeoflearning ofthefirstlist,whenthelistsaremade upofnonsense syllables andlearned bytheanticipation method. Length of Interval
Manypredictions canbemadewithreference totheeffect ofvarying the lengthoftheinterval between learning andrecall ofa list,orbetween a firstlistanda second.Onlya fewof thepredictions willbe included here
asexamples. 7 These predictions depend principally onthenotion ofspon-
taneousrecoveryof generalization. But sincethe rate of spontaneous
recovery willvarywiththeamount of differential reinforcement which hasbeenapplied, thepredictions mustalways be statedin termsoftwo variableslength oftheinterval andamount ofreinforcement or,roughly, degree of learning.
XIII.Interference asa Function ofLength ofInterval Between Tasks
Theinterval afterwhicha list2 mustbeintroduced inordertoobtainthe maximum interference inlearning it willvarywiththedegree oflearning of
Verbal Learning
61
list1,being zero ifdegree oflearning has been carried only as farasthe peak of generalization or below , and thereafter occurring the later , the higher the degree oflearning oflist1. Suppose thattwolists aretobelearned , and thatthere ispotential generalization oftheitems ofthetwolists . Now , iflearning oflistI is carried only tothe point ofmaximum generalization , orbelow ,maximum interference with list2willoccur immediately ,since there can inthis case be norecovery ofgeneralization inlistI,and since ,according toproposition VII,interference willbeproportional tothe degree ofgeneralization . Ontheother hand , suppose that learning oflist1iscarried beyond the peak ofgeneralization .Differentiation willincrease aspractice iscontinued beyond this point (postulate 4), and differentiation willprotect theitems from generalizing withanew list(postulate 10 ) inproportion tothe amount ofdifferential reinforcement given . Butspontaneous recovery of thegeneralization willeventually occur (postulate 7). Itwilloccur more slowly , themore differential reinforcement has been applied (postulate 8). So , the farther differential reinforcement has been carried , the later the maximum in~erference willoccur . Figure 4.10shows thetheoretical relationship .Maximum interference in learning list2willbeobtained ifitisintroduced atonce ,when the first list has been learned only tothe peak ofgeneralization orbelow ; thecurve of interference would probably falloff , then , astheinterval increases . But if - - - List learned to peak of generalization or below -
Figure4.10
List learned to high degree
62
E. J. Gibson
list 1 hasbeenlearnedto a high degree , little interference would occurat once, but ratherwould be postponeduntil spontaneous recoveryof the generalization occurred ; it might thenfalloff again.Therateof riseandfall represented in the curvesis perfectlyarbitrary, sinceit dependson the rate at whichspontaneous recoveryoccursandthe rateat whichgeneralization tendencies areforgotten, andquantitativevaluesfor thesefactorscanonly be determinedexperimentally . No thoroughtest of thesepredictionsexists, but an experimentof Lepley's (21) indicatesthat, with a high degree of learning,interference will increaseto a maximum , andthen decrease as the intervalbetweenlist 1 andlist 2 increases . Thisresultis consistentwith the secondhalf of the aboveprediction.Underthe conditionsof Lepley's experiment , the peakof interferencecarneat about 3 hours. In a more recentexperimentby Bunchand McCraven(5), no tendencywas found for the deQree '-' of transferto changewith varying length of interval, althoughsyllablelists were learnedto a high degree ; but no testswere madebetweenthe zero intervaland 48 hours, so that thereis really no conflictwith Lepley's results.8 XIV. Retroaction asa Function of Intervalbetween Tasks Theintervalafterwhichmaximum retroactive inhibitionwill occurfor a list 1 will vary with thedegree of learningof list 2, beingzeroif degree of learningof list 2 hasbeencarriedonlyasfar asthepeakof generalization or below , andthereafter occurring thelater, thehigherthedegree of learning of list 2. Supposethat two lists are to be learnedin immediatesuccession , and that thereis potentialgeneralization betweenthe two lists; andthat retention of list 1 is to betestedaftera variableinterval. If thedegreeof learning of list 2 is low (reachingthe peakof generalization or below) maximum retroactiveinhibitionshouldoccurimmediately , by the argumentin propositionXIII above. But if the degreeof learningof list 2 is high, maximum retroactiveinhibitionshouldoccurlater, againfollowing the argumentin propositionXIII. A similarpredictioncanbe madeif the degreeof learningof list 1 is varied, insteadof list 2. No test of the temporalcourseof retroactive inhibitionasa functionof the degreeof learningof the two lists hasbeen made. But McGeoch(24, 25) andMcGeochandMcKinney(26, 27) have investigatedthe courseof inhibitoryeffectswith time, andhavenot found a consistenttendencyfor retroactionto vary with time. Two of their experiments gavesomeindicationof anincrease of retroactionwith lengtheningof the interval, oneshowedno tendencyfor retroactiveinhibitionto vary uniformlyin any direction, anda fourth showeda slighttendencyfor
VerbalLearning 63
retroactive inhibition to decrease withtime.Different materials, learning methods, anddegrees oflearning were employed inthese experiments. The
apparentinconsistency of the resultspointsto the need for a further experiment in whichtemporalcourseof retroactiveinhibitionis studiedin relationto degreeof learning.
On the basisof this proposition, and assuming a constantinterval
between learning listI andrecalling it,itispossible tomake predictions for
thewell-known point ofinterpolation problem, withretroactive inhibitionasa function ofbothinterval anddegreeoflearning (9).Results from
a number ofexperiments onthisproblem aresuperficially conflicting (2, 399If.;38).Highdegrees ofretroactive inhibition havebeenobtained by
oneinvestigator oranother withthesecond listinterpolated atalmost any
pointintheinterval between learning andrecall oflistI.Thediscrepancy maybe due,among otherthings, to thevariable degrees of learning employed indifferent experiments onpointofinterpolation. Theexpectation willvary,depending onwhether learning iscarried toa lowora high degree. MeaningfulversusNonsenseMaterial
Thisgroup ofpredictions isrelated totherestofthehypothesis through thedefinition of meaning whichwehaveaccepted. Meaning hasbeen
definedas onecharacteristic of a verbalor visualitemwhichservesto
differentiate it fromotheritemsin otherwords, generalization is at a
minimum formeaningful material, unless ifoccurs ona secondary basis,as
in thecaseofsynonyms (34).Proposition )(Vis included herebecause the
pattern oftheargument issimilar tothatinthepredictions actually dealing with meaningfulmaterial. XV.Learningof Pre-differentiated Items
Ifdifferentiation hasbeen setupwithin a list,lessgeneralization willoccur
inlearning a newlistwhich includes thesame stimulus items paired with
different responses; andthetrialsrequired tolearnthenewlistwilltendto
bereduced byreduction oftheinternal generalization. Suppose thata list1hasbeenlearned, andisfollowed bya list2 whose
stimulus items arethesame asthoseinlistI, although theresponses are different. Lesstotalintra-list generalization should occur inlearning list2 thaninlistI, because ifdifferentiation hasbeensetupamong a number of stimulus items,it willbe easierto differentiate themagainlater,even
though theyarepaired withdifferent responses (postulate 9).Furthermore, since thenumber oftrials required tolearn a listincreases asthedegree of
E. J. Gibson
64
List 2
List 1
/
Ra
Sa.L ::::~_ .. .. Rx Rb
Sb L. . Ry /
Rc
Sc~:~ :__........ Rz Figure 4.11
generalization
2
within
should
tend
A
clear
because
as
ing
cannot
new
factor
reduced
2
;
.
but
as
only
list
1
to
impossible
in
two
the
list
would
The
first
cases
lists
.
)
XVI
half
of
A
of
in
be
was
verified
in
represented
absent
an
is
of
within
to
use
the
stimulus
in
It
actually
weights
tradi
the
the
-
.
comparing
)
.
be
, "
by
as
which
generalization
same
errors
found
as
be
would
the
2
list
responses
group
inhibition
to
can
errors
learned
quantitative
associative
list
in
(
of
( confusion
such
before
control
knowing
as
-
represented
trials
reduction
2
generaliz
would
learning
responses
prediction
-
A
conflict
of
II
.
list
first
number
and
experiment
of
second
the
writer
'
s
.
.
Ease
of
Learning
Meaningful
trials
nonsense
will
Suppose
two
of
,
meaning
one
a
are
list
syllables
is
to
than
lists
one
nonsense
8
Lists
required
,
that
items
are
be
syllables
stimulus
tion
list
without
learned
original
tendencies
-
effect
generalization
reduction
first
2
,
in
inhibited
their
list
prediction
present
are
list
of
against
the
overt
More
list
inter
in
above
is
the
generalizing
but
,
of
the
,
shows
in
facilitating
pitted
of
successfully
number
conflict
. 11
excite
the
,
represented
be
for
half
tendencies
to
have
The
been
generalization
.
responses
4
represented
whether
case
Figure
stimuli
with
.
name
of
8
the
second
of
have
of
trials
generalizing
are
would
say
factors
tional
the
contend
to
the
(
2
the
.
no
decrease
learning
conflict
these
ones
list
)
reduced
;
,
new
list
conflict
1
for
the
learned
list
list
for
-
the
of
made
inter
Therefore
tendencies
well
in
-
,
number
be
,
in
undertaken
increases
the
intra
tendencies
is
list
reduce
case
a
well
the
to
to
are
,
characteristic
in
learn
a
which
be
in
are
learned
meaningful
or
list
they
a
verbal
visual
items
words
in
those
of
According
item
that
the
to
which
are
. 9
only
while
.
or
stimulus
vary
,
meaningless
of
the
meaningful
which
words
relatively
which
serves
the
other
defini
-
to
Verbal Learning
65
differentiate it fromotheritems . Furthermore , if stimulus itemshaveonce beendifferentiated , it will beeasier to differentiate themagainlater , even thoughtheyarepaired withdifferent responses thanthoselearned when theoriginaldifferentiation wassetup(postulate 9). Therefore , moregeneralization will tendto occurin thelist whosestimulus itemsareless meaningful ; andsinceasthedegree of generalization withina list is increased , moretrialsarerequired to learnthelist (proposition IV), more trialswillberequired to learnthelistwhose stimulus itemsarelessmean ingful . Thefactthatalistofmeaningful wordsiseasier to learnthanonemade upof nonsense syllables isnotorious , andit seems certain thata check of thesituation described above wouldyieldpositive results . XVII. Interference andMeaningfulness More trials will be requiredto learna secondlist whenthe stimulus itemsof bothfirst andsecond list arenonsense syllables , thanwhenthey aremeaningful words .
Suppose thata second list is to belearned followinga firstlist, andthat easeof learninglist 2 is to becompared whenstimulus itemsof bothlists arenonsense syllables , or relativelymeaningless , withthecasewhenstimulusitemsof botharemeaningful words.In theformercase , generalization fromlist 1 to list 2 shouldbe greaterthanin the latter, sincemeaning serves to differentiate theitems . And sincemoretrialswill berequired to learna second list in proportionto thestrengthof thetendency for items of afirstlistto generalize withit, morelearning trialswill berequired when thestimulus itemsarerelativelymeaningless . In a checkof this prediction , it wouldbe necessary to preventsuch factorsas overlearning in the nonsense syllablesituationfrom cutting across themainvariable(relativemeaningfulness ) andconfusing theissue . If learningwerecarriedto a criterion , manymoretrialswouldpresumably begiventhefirstlistwhenstimulus itemsweremeaningless , thusintroduc ing a second variable , andincidentally tendingto evenup thetwo situa tionsas regardsdegreeof differentiation reached in the first list. This difficultymightbeavoidedby delayingpresentation of list 2 for a length of timesufficient for spontaneous recoveryto occur . Also, comparisons shouldin eachcasebemadewith controlgroupswhichhavelearned only list 2, but are otherwiseidenticalwith the experimental situationsdescribed , sincelist2 wouldpresumably beharderto learnwhenstimuliwere nonsense syllables , evenin theabsence of list 1. Degree of relativenegative transferwouldbe calculated separately for the two situations , andthen compared .
E. J. Gibson
66
XVIII. Retroactionand Meaningfulness Retroactive inhibition will be greater for a first list when the stimulus
itemsof both first and secondlist are nonsense syllablesthan when they are meaningfulwords. Supposethat retention of a first list is to be tested after interpolation of a second list in two
cases: one , when
stimulus
items
of both
lists are
nonsensesyllables; and two, when stimulus items of both are meaningful words. More retroactive inhibition would be expectedin the first case, by an argument similar to the one above . Again , a number of variables other than the crucial one will tend to cut
acrossthe experimental situation,andit would be absolutelyessentialin a test of the predictionto provide control groupsin which interpolated learning was omitted , so that actual percentages of retroactive inhibition could be calculated for the two cases. In an experiment by Sisson (37),
retention was comparedfor list 1 when both lists were madeof syllablesof 100 percent associative value with a case in which both contained only
syllables of 0 percent associativevalue. Presumably, a higher associative value is correlated with greater meaningfulnessof the item. Sissonfound slightly poorer retention in the first case; but percentagesof retroactive inhibition could not be calculated, sinceno control groups were provided. More learning presentations were given 0 per cent lists, since a learning criterion was used. Time intervals between tasks were brief . This experi -
ment should be repeatedwith adequatecontrol groups, since it has certainly been generally assumedby investigators that the surestand highest percentages of retroaction are to be found with nonsense material .! 0
PossibleExtensions of theHypothesis It has been possible to present only a sample of the predictions and implications of the hypothesis advanced. Many extensionsmay be made, especiallyif one or two new postulates are added. Among the potential extensionsis a group of predictions involving various phenomenarelating to overlearning. These predictions require a definition of overlearning, as well as the new assumption that the variability of trials needed to learn
individual items will increasewith aggregate generalization. Six or more predictions follow , among them these: increaseof overlearning with greater generalization; later maximal loss as a result of high generalization; various length-of-list phenomena, such as decreasein retroaction with increasinglength of list; decreaseof retroaction as generalizationwithin a first
list increases
.
If a postulate concerning IIdisinhibition" is added, a number of predictions regarding the effect of shockor distraction in learning and retroaction
VerbalLearning
67
situations willfollow. Afurther group ofpredictions ensue ifanassumption regarding theeffects ofcaffeine beincluded. These particular extensions
havebeenmentioned, because theyhavebeenworked outin(9). Evaluation
Ahypothesis relating certain factsofverbal learning tothemore general ingthis,a number ofpropositions dependent onthehypothesis were outlined in somedetail.Sincethepotential number ofdeductions froma hypothesis maybealmost infinite, itdoesnotconstitute a perfect evaluationofthathypothesis tocalculate thepercentage ofverified propositions concepts ofdifferentiation andgeneralization hasbeenpresented. Follow-
which havebeenshown tofollow fromit.Yet,it iscertainly ofinterest to
examine theevidence sofarobtained, sincenegative results, at least,area goodindication thatsomething iswrong.Whereevidence relevant to the proposition wasavailable, it hasbeenreferred to already. To summarize,
conclusive positive evidence exists foreight ofthepropositions presented;
somepositiveevidenceexistsfor fourothers,thoughnot conclusive; no
evidence existsforfour;andfortheothertwo,theexisting evidence is
conflicting andnotentirely relevant. Inother words, none ofthese proposi-
tionshasbeenshown tobefalse. Furthermore, ofthepropositions which
havebeenworked outbutnotincluded here,nonehasbeenshownto be
false.
Ahypothesis maywellbeevaluated withreference to itsscope, also and inthisconnection, thetypesoflearning phenomena which the
hypothesis isnotadaptedto explain, aswellasthosewhichit canhandle well,shouldbe mentioned. Obviously, thehypothesis is framed to deal withthataspect ofverbal learning inwhich development ofdiscrimination
between items isvital; andit islimited to a stimulus-response typeof
analysis. Factsrelating tovarious typesoftransfer-effect, bothintra-list and inter-list, constitute itsprincipal applications. Manyotherfactsof verbal learning arenotpredictable fromthehypothesis asif stands. A number of
thesefacts(serial position effects, reminiscence, etc.)canprobably bestbe subsumed under a hypothesis which develops theimplications oftheinhibitory aspects ofmechanisms postulated asbasicto learning, andthe present hypothesis hasleftunexplored suchpossibilities. 11 Otherfactsof verbal learning notsubsumed bythepresent hypothesis would befactsof recognition-memory, andmemory changes.Herea perceptual analysis, perhaps ofthetypemade bytheGestalt psychologists (19,20,32)might bemoreprofitable thana stimulus-response analysis. It maybethatthe problems ofverbal learning must beapproached from boththese angles,
atthepresent timeatleast.Inspiteoftheselimitations, thewriterfeelsthat
therange ofthepresent hypothesis issatisfactory, since nomore compre-
68
E. J. Gibson
hensivetheorycenteredaroundthoseproblems of learningwhichare broadly classifiable astransferphenomena seems tohavebeenproposed. Thevalueof emphasizing discrimination as a factorin verballearning seemsunquestionable; andthatgeneralization anddifferentiation arecon-
ceptssuitable foritssystematic development seems almost aslikely. Notes
1.Thisarticleis a partof a dissertation presented to theFaculty of YaleUniversity in
partial fulfillment oftherequirements forthePh.D. degree. Thewriter wishes toexpress thanks to Professor Clark Hull for his advice and assistance.
2.Thetermdifferentiation willbe usedin placeof differential inhibition, sincethe term
inhibitionimplies a theoryofdifferentiation whichthepresent hypothesis neednot assume.
3. It willbe notedthat thisdefinitionmakesno referenceto nervousprocessesor to any
physiological explanation. Thisusage isinaccordance withthatfollowed byBass and Hull(1)andHull(16).Thatgeneralization willoccurwithvoluntary verbalresponses in theformofa typical generalization gradient hasbeendemonstrated byGibson (6).
4.Thereis a possibility thata slightincrease in specificity mightbeobtained bypure practice ofa single connection. Thispossibility hasnotbeenincluded inthepresent
system because thegreatoverelaboration necessary would beoutofproportion tothe importance of the possibility.
5.From thispointon,itwillbeassumed thatthegeneralizing stimulus items havedifferent responses, andtheassumption willnotbespecifically mentioned. 6.Thereis a furtherpossibility of explaining a decrement in recallof list1.Aslist2 is learned, itemsfromlistI tendto occur, andmustbedifferentiated fromthelist2 items. Suppose thatthisdifferentiation actually involves inhibition ofthegeneralized responses. ThenthemorelistI tendsto generalize withlist2, themoreitemsfromlistI willbe inhibited. Whenlist 1 is returnedto, not onlywilllist 2 itemstendto occurby
generalization, butalsotherewillbespread ofinhibition fortheresponses oflistIin thiscase,thedesiredresponses. Whether thispossibility shouldbe takenseriously depends partlyonwhether generalization ofresponses fromlist2issufficient toaccount fortheloss.Acomparison ofthelossinthissituation withthatinthesituation described
inproposition VIIIshould berelevant, forinthatsituation therecouldbenoadded inhibition withwhichto reckon. Ifrecallof list2 is notaspoorasrecalloflist1 after thesameinterval, thensomefurtherfactorsuchasinhibition mustbe at workin the present situation.
7. For others,see dissertationon filein YaleUniversityLibrary(9).
8.InBunch andMcCravens experiment, thetotaltransfer effectwaspositive; thiswould not,however,preventrelativedegreesof interference fromshowingup.
9.Anexception tothisprediction should beacaseinwhich thestimulus items aremutually synonymous meaningful words. Thissituation hasactually beenshown, intheSmith College Laboratory, toincrease learning difficulty ascompared withunrelated stimulus words. 10.If thematerial is reallyhighlydifferentiated, beingnotonlymeaningful words,but, furthermore, wordsmaking upprosepassages or poerty,rathersmallpercentages of retroactionare apt to be obtained(26,27).
11.TheLepleyhypothesis (21,14)covers someofthesefactsbyexploitation ofa postulatedinhibition of delay.
VerbalLearning 69
Bibliography
1.Bass, M.J.,&Hull, C.L.Theirradiation ofa tactile conditioned reflex inmanJ.comp. Psycho!.,1934, 17, 4765. 2. Britt, S.H.Retroactive inhibition: a review of theliterature. Psycho!. Bull., 1935,32, 381440. 3.Bruce, R.W.Conditions oftransfer oftraining. J.exper. Psycho!., 1933,16,343361. 4.Bunch, M.E.,&Winston, M.M.Therelationship between thecharacter ofthetransfer andretroactive inhibition. Amer. J.Psycho!., 1936,48,598608.
5.
, &McCraven, V.G.Thetemporal course oftransfer inthelearning ofmemory
material.J. comp.Psycho!., 1938,25, 481496.
6.Gibson, E.J.Sensory generalization withvoluntary reactions. 1.exper. Psycho!., 1939, 24, 237253.
7. 8.
. Intra-list generalization asafactor inverbal learning Manuscript inpreparation. . Retroactive inhibition as a function of degree of generalization. Manuscript in preparation.
9.
. Asystematic application oftheconcepts ofgeneralization anddifferentiation to verballearning. DoctorsThesis, YaleUniversity, 1938.
10.Gulliksen, H.Transfer ofresponse in human subjects. J. exper. Psycho!., 1932,15, 496516.
11.Hovland, C.I.Thegeneralization ofconditioned responses: I.Thesensory generalization ofconditioned responses withvarying frequencies oftone.1.gen.Psycho!., 1937, 17, 125148.
12.
. Thegeneralization ofconditioned responses: II.Thesensory generalization of conditioned responses withvarying intensities oftone.J.genet. Psycho!., 1937,51,
279291. 13.
. Thegeneralization ofconditioned responses. Ill.Extinction, spontaneous recovery, anddisinhibition ofconditioned andofgeneralized responses. 1.exper. Psycho!., 1937, 21, 4762. 14.Hull, C.L.Theconflicting psychologies oflearningawayout.Psychol. Rev., 1935, 42, 491516. 15. 16.
46,
. Mind,mechanism, andadaptive behavior. Psycho!. Rev.,1937, 44,132. . The problem of stimulus equivalence in behavior theory. Psychol. Rev.,1939, 930.
17.James, H.E.0. Theproblem ofinterference. Brit.J.Psycho!., 1931,22,3142.
18.Kline, L.W.Anexperimental studyofassociative inhibition. J.exper. Psychol., 1921, 4, 270299. 19.Koffka, K.Principles ofGestalt psychology. NewYork: Harcourt, Brace andCo.1935. 20.Khler,W.,&vonRestorif, H.Analyse vonVorgngen imSpurenfeld. II.ZurTheorie der Reproduktion. Psycho!. Forsch., 1935,21, 56112.
21.Lepley, W.M.Serial reactions considered asconditioned reactions. Psycho!. Monogr., 1934, No. 205. 22.McGeoch, 1.A.Theinfluence ofdegree oflearning upon retroactive inhibition. Amer. 1. Psycho!.,1929, 41, 252262. 23.
. Theinfluence ofdegreeofinterpolated learning uponretroactive inhibition.
Amer.J. Psycho!.,1932, 44, 695708. 24. 25.
. Studies inretroactive inhibition: I.Thetemporal course oftheinhibitory effects
of interpolated learning. J.gen.Psycho!., 1933,9,2443.
. Studies inretroactive inhibition: II.Relationships between temporal pointof interpolation, length ofinterval, andamount ofretroactive inhibition. J.gen. Psycho!., 1933, 9, 4457.
70
E. J. Gibson
26., 27.
& McKinney, F. Retroactive inhibition in the learning of poetry.Amer.J. Psycho!., 1934, 46, 1 34.
. Thesusceptibility of proseto retroactive inhibition. Amer.1. Psycho!.,
1934, 46, 429437.
28.McGeoch, J.A.,&McGeoch, G.0. Studies inretroactive inhibition: X.Theinfluence of similarity of meaning between listsofpairedassociates. J. exper. Psycho!., 1937,21, 320329.
29.Melton, A.W.,&McQueen-Irwin, J.Inter-serial competition ofresponses duringthe
relearning ofserialverbal material intheretroactive inhibition experiment. Psycho!. Bu!!., 1938, 35, 691692.
30.Muller, G.E.,&Pilzecker, A.Experimentelle BeitrgezurLehrevomGedchtniss. Z. Psycho!., Erg. 1, 1900.
31.Muller, I.ZurAnalyse derRetentionsstrung durchHufung.Psycho!. Forsch., 1937,22, 180210.
32.Ortner,A.Nachweis derRetentionsstrung beimErkennen. Psycho!. Forsch., 1937,22, 5988.
33.Pavlov,I. P. Conditioned reflexes(trans.by C. V.Anrep).OxfordUniv.Press,1927.
34.Razran, G.H.S.A quantitative studyof meaning by a conditioned salivary technique (semanticconditioning).Science,1939, 90, No. 2326, 8990.
35. von Restorif,H. Analysevon Vorgngenim Spurenfeld. I. Uberdie Wirkungvon Bereichsbildungen imSpurenfeld. Psycho!. Forsch., 1933,18,299342. 36.Robinson,E.S.Associationtheorytoday.CenturyPsychologySeries,1932.
37.Sisson, E.D.Retroactive inhibition: theinfluence of degreeof associative valueof originalandinterpolated lists.J. exper. Psycho!., 1938,22,573580. 38.
.
Retroactiveinhibition:the temporalpositionof interpolatedactivity.J. exper.
Psycho!., 1939, 25, 228233.
39.Spence, K.W.Thedifferential response inanimals tostimuli varying within a single dimension. Psycho!.Rev., 1937, 44, 430444.
40. Thorndike,E. L.Humanlearning.CenturyPsychologySeries,1931.
41. A note on assimilation andinterference. Amer.J. Psycho!., 1937,49,671676. 42. Waters,R. H. The law of acquaintance. J. exper.Psycho!., 1939,24, 180191.
43.Whitely, P.L.,&Blankenship, A.B.Theinfluence ofcertain conditions priortolearning uponsubsequentrecall.J. exper.Psycho!., 1936,19,496504.
44.Yum,K.S.Anexperimental testof thelawof assimilation. J. exper. Psycho!., 1931,14, 6882.
Retroactive Inhibition as a Function of Degreeof Generalization between Tasks
EleanorJ. Gibson
Looking backat thisexperiment, thenovelaspectof it seemsto meto be
thepreliminary experiment inwhich a measure ofgeneralization among a setof
items wassecured, independent ofthelearning andrecall called forinthemain experiment onretroactive inhibition. Confusion between items wasthemeasure of
generalization, alsoa novelty. Inchoosing material tobesampled fordegree of generalization among items oftheset,I selected pictorial instead ofverbal material.Similarity wastobejudged, as wellas theactual responses toan item confusing it withanother. Nonsense forms were suggested asappropriate byan
earlier experiment ofmyhusbands U.Gibson 1929; histhesis research, actually), inwhich hestudied memory forforms andfound thatthememory ofa form
changed overtime. I copied some ofhisforms indevising myset. Myfirstexperiment, using thismaterial ascue items forpaired associates in a retroactive inhibition experiment, wasperformed asa group experiment, using
students infivelaboratory sections oftheintroductory course atSmith College. Theresults wereinlinewithpredictions andthusquite satisfactory, butHull distrusted group experiments, andsoitwasrepeated withindividual subjects. The
results ofthefirstexperiment werereplicated, including a finding thatstrict analogy withconditioning would hardly have predicted: thata setofitems, once differentiated, doesnotgeneralize againwhen newresponses mustbelearned to
them,andthatdifferentiation obtains withinthesetas a whole, notas an acquisition ofa particular item.Intralist generalization wasgreatly reduced in
interpolated learning, even when theitems were notthesame ones presented for original learning. What occurs, itseemed tome,wasa kind ofperceptual learning, butnotlearning attributable todistinctiveness ofparticular responses learned to each individual item, asa burgeoning theory ofacquired distinctiveness ofcues
assumed (Miller andDollard 1941). Differentiation, onceachieved, is notlost again,noris it a specific resultofattaching particular responses to individual JournalofExperimental Psychology, 1941,28, 93115.
72
E. J. Gibson
items. A setofconfusable itemsisdifferentiated bymeans ofdiscovering the features thatareshared tovaried extents byitems within theset.Thisidea surfaced yearslaterwhen I wasgroping fora theory ofperceptual learning. Thereweretwomorepapersin thisseries, notreprinted here(onebyoneofmy
graduate students). A hiatus followed dueto World WarIIandcoping witha family along withteaching. Butformedifferentiation asafundamental process in development andlearning wasfirmly established, tobecontinued later .
A hypothesis relating verbal learning to twoconcepts fromtheliterature of conditioninggeneralization and differentiationhasrecentlybeen
presented bythewriter (4)12Thishypothesis hasadvanced thegeneral notionthatdiscrimination is anessential processinverballearning, andthat
thespecial difficulties referred tobysuchtermsasinterference andretroactive inhibition can be understood,at leastin part, as casesof relativelyiow
discriminability of learning material. Whenthisnotionis combined with theprinciple of establishment of differentiation between thepreviously undiscriminated items,many implicationsfollow.These implicationsare
statedasspecifically aspossible intheformofpredictions in thearticle referred to.Figure 5.1below mayclarify themechanism assumed tobeat work.It represents a listof pairedassociates in whichgeneralization is assumed to be takingplaceduringlearning. Generalization is defined as the tendency foraresponse Ralearned to5a tooccur when5b (with which it hasnotbeenpreviously associated) is presented(4,p. 204).In other words, 5a 5b andS possess lowdiscriminability whentheyarepresented
separately. Generalization tendencies (represented bybroken arrows) will
tendto blocktherightresponses (solidarrows) in proportion to the strength of thetendencies. Competitive blocking thusoccurs, rendering learning difficult.
Ina two-listsituationalso,generalization andcompetitive blocking may
occur. Figure 5.2represents therelationships. Whenthesecond listis learned, itsstimulus members maygeneralize withthoseofthefirstlist,and Sa
*Ra / \
s.
/
\_,
Sb f
,,-_\ .
S Figure 5.1
\/\
/%.
Rb
A
Rc
RetroactiveInhibitionas Function
73
theresponses ofthefirstlisttendtooccur, ortoblock therightresponses, during thenewlearning. Likewise, ifList1 isrecalled afterList2,generalized responses from List 2mayinterfere with recall, producing competitiveblocking or overtgeneralization (reversion to thewronglist).The right excitatory tendencies willbeweakened, itisassumed, inproportion tothedegree ofinter-list generalization. This argument leads tothepredictioninthehypothesis (4,p.215)thata firstlistwillbemorepoorly recalled as thestrength of thetendency foritemsof a secondlistto generalize withit increases. Thepresent experiment isdesigned totest thisproposition; thatretroactive inhibition is a function ofdegree of generalization between a primary andaninterpolated list.Theexperi-
mental situation willyielddatarelevant to theprediction ofinterference fromListI to thelearning ofList2 aswellasretroactive inhibition. The
hypothesis predicts thatmorerepetitions willbe required to learna
second list,inproportion tothestrength ofthetendency foritemsofa first
list to generalizewith itemsof a secondlist.
Numerous experiments havedemonstrated that retroactive inhibition
varies withthedegree ofsimilarity between anoriginal andaninterpolated task(1,p.339if.).Thepresent hypothesis isobviously related tothiswork, butwould maintain thatindistinguishability, rather thansimilarity orhomo-
geneity (14)is theimportant variable. Similarity implies a sophisticated phenomenal relationship involving boththe likeness andthe discriminability oftheitems inquestion. Butit istheindistinguishability ofthe items,andthenecessity fordifferentiating themwhichformsthebasisof thepresent hypothesis. Attheoutset oflearning, items usually havevaryingdegrees ofstimulus equivalence, particularly intheverbal learning situation where items arenotpresented sidebysideforperceptual comparison. Itisreasonable tosuppose thatthisequivalence orindistinguishability isessentially similar totheclassical factofgeneralization which appears in
theearlystagesofconditioning.
Sincethe purposeof the presentinvestigation wasto discover the
relationship between generalization andretroactive inhibition, a prelimi-
naryexperiment wasdesigned tostandardize material interms ofdegree of List1
List2
List1
Sa
0 Ra
Se
..R
Sa o-Ra
Sb
ORb
Sd
.Rd
Sb -
. Rd
Figure 5.2
.-Rb
74
E. J. Gibson
generalization. Themethod ofstandardization andtheresulting dataare presented inExperiment I.Thestandardized material isutilized inExperimentII,a retroactive inhibition experiment, to formseriesofprimary and interpolated listsdiffering indegree ofgeneralization. Experiment IIIisa repetition ofExperiment IIusingindividual subjects instead ofthegroups used
in II.
Experiment I.Determination ofDegree ofGeneralization ofStimulus Items Procedure
Fordetermining degreeof generalization of items,a technique previously em-
ployed byYum (15) wasused. Aseries ofstandard forms wasselected, tobeused ascuesinpaired associates learning. Anumber ofvariations fromeachofthese formswasthendrawn. Twenty-four hoursaftera groupofsubjects hadlearned the
listofpaired associates, arecall series wasgiven them, inwhich each cueform was either theoriginal standard, ora variation substituted forthestandard. Themeasure ofgeneralization wastaken astheextent towhich thesubjects responded tothe variation asifitwere thestandard. 3 Yum wasinterested inrelating thefrequency of response tothevariation withthedegree ofperceptual similarity between the variation andthestandard. Perceptual similarity wasdetermined byjudgesratings. Tomake thepresent experiment comparable, a number ofgroups offorms, each group consisting ofa standard andseveral variations, weredrawn oncards and presented totenjudges. Theywereasked toputeachvariation withthestandard whichit resembled most.Whenthiswasfinished, theywereaskedto rankthe variations according to degree ofsimilarity to thestandard. Thirteen groups of formsweretreatedinthisway.Eachfinally consisted ofa standard formandtwo variations fromit.Therewereavailable thenthreeclasses offorms: standard, first
degree similarity, andsecond degree similarity. Included alsointheexperiment wasa fourthclassofformswhichborenoresemblance to anyofthestandards, as determined by the judges.
Relative degrees ofgeneralization forthevariation-forms werenowdetermined.
The13standardformswereusedasthecuemembers of a listofpairedassociates.
Eachstandard waspaired witha monosyllabic word(again inaccordance with
Yums procedure). Theorder ofthepairs wasvaried witheachpresentation, sothat noserialassociations wouldbeformed. Fourtestlistsfordetermining degreeof
generalization were prepared, each composed of13forms, likethelearning list,but
formsofall4 classes wereincluded amongthem.Ina typicaltestlisttherewould
be3 ofthestandard forms, 3 forms highly similar totheirstandards, 3 forms less
similar to theirstandards, and3 formswhichresembled noneof the standards.
(Actually, therehadinanylisttobe4 forms fromoneoftheclasses tomake up 13.)There werenecessarily 4 different testlists,toinclude thetotalof54forms. Novariation everappeared in thesamelistwithitsstandard. Fourgroups of subjects tookpartintheexperiment, varying innumber from19to29.Each group hadtheoriginal learning list,butadifferent testlist.Thegroups werefoursections in an elementary psychology course.
RetroactiveInhibitionas Function
75
Aspecial serial exposure device was used topresent thematerial tothesubjects,
consisting ofaJastrow memory apparatus placed behind aprojector which threw alarge image ofeachform-word pairona translucent screen. Theforms hadbeen printed bymimeograph sothatperfect copies were available forthedifferent lists,
and theresponse words were printed inblack India ink. Exposure time foreach pair
was2 seconds. Five learning trials oftheoriginal listwere given; immediately following each presentation, recall was tested, theforms alone being presented in adifferent order each time. Thesubjects hadtorecord theproper response words
ina prepared booklet within 3 seconds. Thesubjects werethenreleased but returned thenextdayatthesame hourandweregiven oneofthefourtestlists
fordetermining degrees ofgeneralization. This testwas similar inprocedure toone
oftheprevious recall trials. Following thistestseries, anordinary recall testwas given using the originalforms.
Thefollowing instructions were given thesubjects atthebeginning ofthe
experiment:
You willbeshown agroup offorms, each onepaired withaword. Study
these pairs sothatwhen aform isshown alone, youcanwrite theappropriate
word. Youwillbeshown onlyonepairata time. Donottrytolearnthese pairsinanyparticular order, because theorderwillbechanged everytime.
Thepoint is,toassociate aparticular word withtheform withwhich italways Aftereverypresentation oftheforms paired withwords, youwillbe shown theforms bythemselves inordertoseewhether youremember the appropriate word ornot.ifyoudoremember it,write it intheblank proappears.
vided, being sure toputitintheproper space. Donotleave anyblank spaces; ifyoudonotremember theproper word, draw aline through thespace. After wehavefinished oneofthese testtrials, turnthepagesothatyoualways havea blank pagebefore youduring thelearning. Onthesecond day,before thetestseries, these instructions were given: Today Iwant youtolook atsome forms astheyappear andputdown the
wordforeachonewhenever youcan.Youwillbeshown forms bythemselves; tryto writetheappropriate words, justasyoudidinthetestseries yesterday. Youmaynotremember themallthefirsttime, butyouwillbe given another chance later.
Thesubjects could hardly failtorecognize atsome pointintheseries thatthiswasnottheoriginal list,since 3or4oftheforms were totally unfamiliar ones bearing noresemblance tothestandards. This awareness probably hadtheeffect (tojudge from later questioning) ofcausing them towrite responses onlytothose forms which theyfeltsureof,sothatingeneral whenvariations wereresponded toasiftothestandards, theitems were functioning asequivalent stimuli, notasmerely recognizably similar stimuli. Results
Degree ofOriginal Learning Since variations inthedegree oflearning attained areprobably accompanied byvariations inthedegree ofgen-
76
B. J. Gibson
Table
5.1
MeanNumberof ResponseWordsCorrectlyRecalledafter5 Presentations
Group I
Group II
Group III
Group IV
12.5
12.0
11.9
11.9
eralization exhibited, 4 the meannumberof responsewordscorrectlyre-
calledafterthefifthlearning trialforeachgroupofsubjects ispresented in Table5.1.Thedegreeoflearning attained washigh,beingapproximately 12 out of a possible13 correctresponses. Thisimpliesthat manyindividualswereableto recallall 13 responsescorrectly.The differences betweentheaverages of the4 groupsaresatisfactorily small.
Degrees ofGeneralization forIndividual FormsThemainpurpose of the
resultswasto establish fourlistsofformsvarying intheirtendency to call forththe sameresponses as didthe originallist of standardforms.To obtainsuchlists,theresultswerescoredasfollows. First,thenumber of
subjects responding correctly to eachstandard formin thetestlistwas
obtained. Then,thenumberresponding to eachof the variations as if it werethestandard wascalculated. Whentheselatterfiguresareexpressed
aspercentages ofthetotalnumber taking partinthegroup concerned, they
constitutemeasuresof the degreeof generalization with the standard.
Finally, it wasdetermined whether anyformrateddissimilar to thestandardswasresponded to asifit wereoneofthestandards; andwhether anyvariation froma standard wasresponded to asifit werea different standard.
Figure 5.3shows theforms grouped inclasses according to theirtendency to generalize withthestandards. Thestandard forms themselves areClassI. Theformsin ClassII arethosevariations whichproduced
thehighest frequencies ofresponses belonging to theirstandards. Those grouped inClass IIIproduced lower frequencies ofthestandard responses. In eachcase,theformsin ClassII andClassIIIarelistednextto their standards. Thepercentage ofthegroupwhichresponded appropriately is
ineachcasegivenopposite theform.Themeans showa widedifference between Classes I,II,andIIIintermsoftendency to callforththestandard
response. Theforms grouped asClass IVarethosewhich werejudged dissimilar. Fourof theseformswereresponded to by a fewsubjects, but withno consistency as regardsthe responses given.If the meanpercent
response iscalculated regardless ofthenature oftheresponse, thefigure is 2.8percent. Theorderinwhich forms ofthisclassarelisted isnecessaril arbitrary. Noneofthevariations (Classes IIandIII)wasresponded towith anyconsistency asifitwereastandard otherthantheonebeside which it islistedinthechart.Therewerea totalof10wrong responses eachfor
Retroactive Inhibition as Function
77
CLASSIV
() ~. ;J-.-J.j . .
. .
. .
. 84.5
41.1
~ 9Q ~ bb gg % 0 ~ C5~ ~
. ~ I1II Av .
::;J/ / ~~~2:~1.j :;[J ,I,,I 6~-'~ ..D cyu ~ 9.7
Figure 5.3 Stimulus fonns inclasses according tothe percentage ofsubjects giving the stan dard responses .grouped
78
E. J. Gibson
ClassII andClassIllabout 10percentof the totalpossibleresponses in each
case.
Twoof the formsincluded in ClassIIIwerenot responded to at all,
although thesetwoformshadbeenjudgedsimilar to thestandards next which theyarelisted. Besides thesetwocases, therewereonlytwodiscrepancies between similarity asratedbythejudges andgeneralization as determined by quantitative score.Thesediscrepancies causeda change fromClassIII to ClassII in the caseof forms10 and 12. Therewas
therefore a highcorrelation between judgments ofsimilarity andobjective
tendency to generalize. Toshowthiscorrelation, theresults canalsobe scored according toamethod usedbyYum. Hecalculated foreachtestlist themeannumberofformsofeachsimilarity groupresponded to correctly,
ratherthanscoring eachformindividually ashasbeendoneabove. Ifthis
calculation is madein the presentexperiment, the dataareconsistent for thefourtestlistsin showingthehighestmeanresponseto the standard
forms, nexthighest toforms judged offirstdegree similarity, andleasthigh to thosejudgedof seconddegreesimilarity. 5
Experiment II.Degree ofRetroactive Inhibition andofInterference asDependent on Degreeof Generalization Procedure
Sinceit wasproposed to testtherelationship between degreeof inter-list generalization ontheonehand,anddegreeofretroactive inhibition ontheother,the conditions oftheexperiment wereplanned soastoinclude threedifferent degrees
ofgeneralization between primary andinterpolated lists(Conditions II,IIIandIV below). Another condition contained identical stimulus itemsin thetwolists
(Condition I),anda fifthserved asa control forcalculating retroactive inhibition (Condition V).Inallconditions theprimary learning listwasmadeupof the standard (Class I)forms paired withnonsense syllables. Thefourclasses offorms
astheyaregrouped inFig.5.3constituted thestimulus items forthefourinterpolated lists.Theplanof theexperiment wasthenasfollows. Theresponse Learning of
Learning of
InterpolatedList
Retention of
Primary List
Condition
Stimulus Forms
Stimulus Forms
Stimulus Forms
I
ClassI (Standard) ClassII
Standard
II
Standard Standard
III
Standard
Standard
IV
Standard
ClassIII
V
Standard
(Nolearning)
Standard
ClassIV
PrimaryList
Standard Standard
RetroactiveInhibitionas Function 79
members oftheinterpolated listswere nonsense syllables different from those of theprimary list,butidentical inallfourinterpolated lists. Conditions I,II,III,and
IVdiffered, therefore, only astostimulus forms. The syllables forboth primary and interpolated listswere chosen according totheusual rules from Hulls list(7) Only those oflowassociation value were included. Incondition V.thesubjects
were given copies oftheNew Yorker toreadduring theinterpolated interval, with Thedetails ofprocedure were similar tothose ofExperiment I.Since thenature
instructions to selectthebestandtheworstcartoon.
ofthefive conditions precluded using asubject more than once, five large, equivalent groups ofsubjects were secured. The sections ofanelementary laboratory course offered themost available andsuitable subjects. Thesame technique of
exposure wasemployed, again allowing twoseconds perpairforlearning, and threeseconds peritemfortherecall trialwhich followed. Theorderwasvaried withevery learning trialandeverytesttrial. Fivelearning andtesttrials were given. 6 Theprepared booklets inwhich responses were recorded were collected
immediately following theoriginal learning, andnew booklets passed out.Instructions fortheinterpolated learning followed andthistaskwasbegun atonce. The interpolated listwas also presented five times, with tests following each presentation.Assoonastheinterpolated learning wasfinished, books werecollected, others distributed, further instructions given, andrecall andrelearning ofthe original listwasbegun. Inallfive conditions, theinterpolated period between the endofthefifth testoforiginal learning andthefirstrecall testofitlasted fourteen minutes. Seven relearning repetitions ofthefirstlistweregiven. Thenumber was limited to seven, owing to the termination of the class period; it did not allow learningto complete masteryin mostcases.
Theinstructions were almost identical with those given inExperiment I and
need notberepeated. After the5presentations ofthefirstlist,subjects were told: Weshall return tothislistlater on,andgive enough more trials foreveryone to finish learning. When thenewrecord books hadbeen handed out,theywere told: Nowthere will beanew list,ofthesame type, tobelearned. Theprocedure will beexactly similar tothatwehave justfollowed. Please tryagain towork ashard aspossible. Thefinal instructions given before recall ofthefirstlistwerethese: Nowweshall goback tothefirstlistandrepeat ituntil everyone haslearned all ofthepairs. Weshall startwithatesttrialtoseehowmuch youremember. Write every syllable which youcanrecall intheproper space when itsform isshown, and
draw aline intheothers, asusual. Then weshall proceed with alternate learning
and test trialsas we did before. Results
Degree ofOriginal Learning Since thevalidity oftheresults depends on theequivalence ofthe5groups ofsubjects asregards degree oforiginal
learning, themeannumber ofsyllables correctly recalled onthefinaltest
foreach condition ispresented inTable 5.2.Themeans arefairly uniform,
andthevariation could notberesponsible forthedifferences inrecall which
will appearlater.
80
E. J. Gibson
Table
5.2
Degree of Original Learning Condition
Mean No. Learned
No. of Ss
Table
1
II
8.21
III 8.05
7.86
24
22
22
IV 7.92 39
V 7.91 23
5.3
Degree ofLearning andMean Number ofReversion Errors inInterpolated Learning Condition
MeanNo.Recalled atTrial5 Mean Total No. Recalled
Mean No.Reversion Errors
II
I
7.67 22.75
.83
III
9.68
30.91
.45
Iv
10.00
31.82
.41
9.82
33.95
.41
Degree ofLearning ofInterpolated ListsThedegree ofinterpolated learning
achieved inconditions I,II,III,andIVisinteresting ontwoscores. First,the
hypothesis predicts thathigher degrees ofgeneralization between a first
anda secondlistshouldmakethesecondlistharderto learn.Andsecond,
variation inthedegree ofinterpolated learning achieved mayinitselfaffect thedegreeof retention of theprimary list,sincedegreeof retroactive inhibitionhas been shownto be a functionof degreeof interpolated
learning (9,12).Table 5.3shows themeannumber ofsyllables correctly recalledafterthe fifthtrialof the interpolated list in eachof the four conditions, as wellas the meantotalnumberrecalled on all fivetests
ofinterpolated learning. Onlythedifferences between Condition I andthe otherconditions approach statistical reliability, 7 but the totalsshowa definite tendency formoresyllables tobecorrectly recalled asdegree of generalization decreases fromCondition I to Condition IV.Thelearning curvesforeachofthefourconditions arepresented inFig.5.4.Condition I
is lowerthantheothersthroughout thefivetrials.In theearlytrials, Condition IVisnoticeably higherthantheotherthree,butinthelatertrials itscurveis closeto thecurvesforConditions II andIII.Thechieffact shownisthemuchlowerdegreeoflearning achieved inCondition I,which
isinkeeping withthehypothesis under consideration. Itmight alsoaffect thedegree ofretroactive inhibition forthiscondition bylowering it,since
it hasbeenshownthatretroactive inhibition decreases withlowerdegrees ofinterpolated learning (within therangeofpartiallearning).
Whentheresponses givenduring interpolated learning areanalyzed for
reversion errors(casesof overtgeneralization withthepreceding list)it
shouldbe expected, according to thehypothesis, thatmoresucherrors accompany thecondition where learning ismostdifficult. Themeannum-
81
Retroactive Inhibition as Function
a311~J3~ All .J3~~OJS3SNOdS3 ~ ::jO~38l1 'JnN
(g.x- a0- - - - . 0- --
Figure
5 .4
Learning
curves
number
ber
of
of
of
that
more
conditions
according
I
Degree
of
be
ately
is
the
low
Retroactive
the
Table
number
of
groups
show
5
correct
.4
first gives
significant
at
the degree
of
first
of
of
!
( Recall
retroactive
second
the
) ,
the
comes
from
) ,
and
inhibition
. 8
all
the
second
from
All
,
.
list
immedi
recall
a
,
overtly
-
test
relearning
based
four
on
the
experimental in
is
the
first
test
inhibition
test
It
errors
at
recall
the
.
between
of
first
.3
reversion
retention
2
5
between
occur
the
retroactive
and
of
with
,
of
from
Table
difference
errors
degree
trial
percent
terms
coinciding
measures
( Recall
relearning the
I ,
such
:
in
in
greatest
feature
measures
plotted
given
reversion
which
The
is
also
The
striking
with
several
.
and
A
learning
recalls a
.
learning
.
Condition
series
recall
II
interpolated
follows .
both
in
that
Inhibition
from
of trial
are
occur in
frequency
calculated
Degree
each
conditions
errors
learning
.
after
four
Condition
following
which
the
to
and
lists
recalled
such of
Condition
,
interpolated
for
difficulty
however
four correctly
errors
obvious
score
the
responses
such
greater
may
CONDITIONIV CONDITIONIII CONDITIONII CONDITIONI
terms
of
both
,
82
E. J. Gibson
Table5.4 RecallScores , Percentof RetroactiveInhibitionMeasuredby RecallScores , andMean
Number ---------of - Trials toRelearn 9Items Conditions --------- I II III IV V Mean Recall 1 3.04 3.00 4.18 5.79 7.57 - ---- of 9.78 8.36 7.00 6.55 5.92 Mean of Recall 2 Retroactive Inhibition , Recall ! Retroactive Inhibition , Recall 2
Mean ----Trials to'Relearn '
60.6% 41.7%
4.08
59.9% 32.6%
3.73
45.8% 29.6%
3.86
23.5% 14.6%
3.00 2.35
IV Figure 5.5 Degrees of retroactive inhibition plotted as a function of degree of generalization between stimulus forms of original and interpolated list. The Roman numeralsunder the percentages of generalization on the abscissaindicate the experimental condition previously described.
Retroactive Inhibition as Function
83
measuresof recall. Furthermore, the degree of retention varies with the degree of generalizationbetween the original and the interpolated leaming.9 Figure 5.5 shows degreeof retroactive inhibition plotted against the degreeof generalizationas obtained in the standardizationexperiment. As the degree of generalizationincreases , so also does degree of retroactive inhibition, as the hypothesis predicts. When the two curves are averaged, the result is a smooth, negatively acceleratedgradient. It is noteworthy
that Condition I (identical stimulus items) does not
show a significantincreasein retroactive inhibition over Condition II (Class II stimulusitems), particularly in RecallI. A greater differenceis apparentin Recall 2, but even here the critical ratio is only .74. This result may be due
to the lower degree of interpolated learning for Condition I. In other words, had degree of interpolated learning in Condition I been as high as that in the other three series, a higher degree of retroactive inhibition might have been found. Another feature of the curves is the sharp rise from Condition IV to Condition III Condition
from 0 generalization to 10 percent generalization . Yet
IV itself shows 23 .5 percent
retroactive
inhibition
in the first
recall. It may be questioned whether such a degree of retroaction for a condition of supposedly zero generalization is consistent with the theoreti -
cal prediction. Two explanations suggest themselves. The first is that syllable responseswere learned in the interpolated list as well as in the primary list, and that something analogous to generalization may take place among response membersas well as stimulus members. Doubt is cast on this possibility by the McGeochs' study (10), which showed that variation in degree of similarity of responsemembers of interpolated paired associates to response members of a primary list was without
effect in
determining degree of retroactive inhibition. This fact is consistent with Thorndike's finding (13) that similar responsemembersare not frequently substituted
for one another in recall , but that similar stimulus members do
frequently evoke one another's responses . The second possibility is that there was actually generalization between stimulus members of original
and interpolated lists in Condition IV, in spite of the negligible tendency established in the standardization
experiment . It will be remembered
that a
very high degree of original learning was achievedin the standardization experiment , while only 8 out of 13 responses, on the average, were learned
in the presentexperiment. It is suggestedthat generalization, after reaching a maximum, decreasesas learning increases ; and that as generalization decreases , the stimuli
most discriminable
from the standard are the first to
ceaseproducing generalizedresponses . In this case, the forms belonging to ClassIV were ineffective in producing generalizationin the standardization experiment becausethe degree of learning was high and becausethey
84
E. J. Gibson
varied considerablyfrom the standards.But in the present experiment,
theywereeffective to someextentbecausethe degreeof learningwas muchlower.Evidence to supportthisexplanation maybe foundin thefact that somecasesof errorsof reversionto the interpolatedlist actuallydid occurin ConditionIV,thoughnot as manyas in the otherconditions. But
beforethisexplanation canbe accepted, it willbe necessary to studythe curveof generalization in relationto degreeof learningundercomparable conditions.
The tendencyfor a greaterdecrement to accompany higherdegrees
of generalization is furthersupported by the numberof trialsrequired to relearnto a set criterion.Thismeasureis of doubtfulvalue,sincelearning was not carriedto a criterionoriginally.But it shouldbe of some sign-
ificance, sincethe groupswereequatedas to numberlearned,on the average. TableIVgivesthemeannumber of trialsrequired in eachcondition to relearnto a criterionof 9 out of the 13 items. (A higher criterion couldnot be set, sincesomesubjectslearnedno morethan 9 in the seven
relearningtrials.)ConditionI requiredmoretrials,on the average,to
relearn9 items,thandid any of the others.Thereis not a perfectgradient
here,sincethe averagefor Condition IIIis slightlyhigherthanthat for ConditionII.It is interestingto notethat the difference betweenCondition IIIand ConditionIV is againthe greatest,as wasthe casewiththe two other measures
of retention.
Errorsin Recall In the light of the above results,it is interestingto
examinethe meannumberper subjectof errorsof reversionto the inter-
polatedlistwhichoccurred inrecalling thefirstlistinthefourconditions. Thefiguresforthefourconditions are:Condition I, 1.00;Condition II,1.09; ConditionIII,.91;ConditionIV, .36. As was the casewhen retentionwas
measured by firstrecall,thereis littledifference betweenCondition I and
ConditionII.Whatis especiallyinterestingis the largedifference between ConditionIII and ConditionIV, which coincideswith the great difference between these conditions in all three measures of retention.
It is obvious that actual reversionsaccount for only a smallpercentage
of the errorswhichoccurin either interpolatedlearningor recalland
relearning oftheprimary list.Superficially, thisfactmayappearto be out
of linewiththe hypothesis undertest.Butwhenthe natureof the other errorsis studied,their occurrenceseemscompatiblewith the hypothesis.
Omissions arethe mostfrequenttypeof error.Whenthe relearning trials in ConditionI are examined,omissionsaccountfor 67.9 percentof the errors. Theoretically,omissionsmay occur becausethere is no strong
excitatory tendencyto respond,or becausethetendencyor tendencies are
inhibited.The formerexplanationmay accountfor a numberof the omis-
RetroactiveInhibitionas Function
85
sions, since theprimary listwasnotlearned toa veryhighdegree originally.Thesecond explanation isinlinewiththehypothesis, since it would assume thatwhena correct response anda generalized response, or two generalized responses bothtendtooccur atonce,blocking mayresult. Itis
alsopossible thata response, although learned, wasextinguished during theinterpolated learning andisthereby inhibited. Such a possibility re-
quiresan extensionof the hypothesiswhichwillbe considered later.
Theerrorsotherthanomissions maybedivided intothreetypes
reversionsor casesof overt generalization between lists,casesof overt
generalization within thelist,andmisspellings. During therelearning of Condition I,3.2percent oftheerrorswerereversions; 5.6percent were cases ofovertgeneralization within thelist;and23.4percent weremisspel-
lingsofthesyllable responses. Thequestion arisesasto whether theerrors of misspelling areconsistent withthehypothesis. Whentheseerrorsare examined carefully, a veryinteresting discovery emerges. Manyofthem appearto becompromises resulting fromconflicting excitatory tendencies. Forinstance, twoformswithina listhaveas theirresponses ruvandtah, respectively; thesubject, instead ofresponding correctly to eitherofthem, mayrespond withtuvto both.Or,a formintheinterpolated listwhichhas
beenshown previously togeneralize witha formintheprimary list,may haveasitsresponse kiv;theformintheprimary listmaybepaired withhaj.
Butwhenthesubject isaskedtorecalltheprimary list,hemaywriteneither of theseresponses, butinsteadkij.In suchcases,the stimulus formseems
tohavetwoexcitatory tendencies, anassociated oneandageneralized one, whichresultina compromise response ratherthanblocking. Underwhat conditions compromise willoccurinstead of blocking is an intriguing
question. It seemspossible thatthestrengthofbothtendencies maybethe
keyto thisproblemthattwoequalbutweaktendencies mayleadto compromise, strongonesto blockingbutfurtherstudyis required to
solve it.Probably notallthemisspelling errors areofthistype;anattempt wasmadeto estimate thepercentage falling inthisclass, butit proved impossible to classify manyofthecaseswithcertainty.
Theerrorsofgeneralization within thelistwhichoccurred during the learning oftheprimary andtheinterpolated listsshowaninteresting trend. There isamean decrease of59percent insuchcases from primary learning to interpolated learning. Thisdecrease occursconsistently in allthefour
conditions. Thereduction inCondition I is consistent witha prediction regardingthe learningof pre-differentiated itemsIf differentiation has beensetupwithin a list,lessgeneralization willoccur inlearning a newlist which includes thesame stimulus items paired withdifferent responses (4,
p. 222).Thereductionin the otherconditions is hardto understand unless
something liketransfer ofdifferentiation ispostulated.
86
E . J. Gibson
ExperimentIII . A Repetition of ExperimentII Using Individual Subjects Procedure
Sincethe third experimentwasundertaken principallyasa checkon the reliability of the results obtained in Experiment II, conditions were kept as nearly the same as possible. In general, only those changeswere made which were demandedby individual experimentation in place of a group procedure. The two major changes were omission of Condition V , and the use of a learning criterion rather than a set number
of trials . Condition
V (the control
for retroactive
inhibition
in which
the
subjectsread during the interpolated period) was omitted, becausethere could be no doubt of the fact that retroactive inhibition
occurred and we wished only to
verify the gradient observed from Condition I to Condition IV . The subjects always learnedto a criterion in the primary learning, the interpolated learning, and the relearning so as to insure equal degreesof learning in the four conditions. The lists learned by the subjects in Conditions I, II , III , and IV were identical
with those of Experiment II, except that one pair was omitted, making them 12-unit instead of 13-unit lists. The forms were pasted on heavy paper, and the syllables were typed beside them. The learning material was presented on an electrically driven Chicago-type memory drum. Eachpair was again exposedfor 2 seconds. But when the forms were presented alone, in the test trials, they now appearedfor 2 secondseach instead of 3 seconds, since the subjectshad only to speak the right associatesinstead of to write them. Six random orders were preparedfor eachlist, and the order was changedeachtime for both learning and test trials . All the subject 's responses were recorded by the experimenter .
The procedure was as follows. As soon as the subject had received the instructions, which were the sameas those of Experiment II, the first list was presented. When the list had been exposed, the forms alone were presentedand the subject attempted to recall the appropriate responses.This sameprocedurewas continued until the subjecthad recalled8 of the 12 responsescorrectly in a single trial. Then the subjectwas given a four minute rest period while the list was changed. In order to prevent rehearsing , he was given a magazine to read. At the end of this
period, the interpolatedlearningwasbegun. Theprocedurewasexactlythe same as during original learning, and was continued until the subject had learned 8 of the 12 responses.Another rest period ensued, and the subjectread, as before, until the time interval was up. The total time allowed for the interpolated interval was 20 minutes . The subject then tried to recall the responses of the first list , and
relearning followed. Relearningwas carried to a criterion of one perfect recall trial. The experiment required about one hour. Fourteen subjects were secured for each condition . Since it was important to
equatethe groups for number of trials spent in learning the first list, an attempt was made to secure similar distributions for the four conditions in this respect. That is,
an attempt was made to distribute the fast learnersand the slow learnersequally over the four conditions. This could easily be done since it was possible to assign a subject to a condition after shehad completed the originalleaming . The subjects
were all studentsfrom elementarypsychologycourses , and comparableto the subjects in Experiment II .
Retroactive Inhibition as Function
87
Results
Trials toLearn Original List Themean number oftrialsrequired tolearn
theoriginal listto a criterion of8 outof12rightassociates foreachofthe fourconditions areasfollows: Condition I,6.86;Condition II,7.00;ConditionIII,7.07; Condition IV,7.14.Themeans areverysimilar, andtheslight
differences which existareintheopposite direction fromtheexpected differences inretention. Itisinteresting thattwomore trials were required
here,ontheaverage, to learnSresponses, thaninExperiment II,where onlyfivetrialsweregiven.Thedifference maypossibly bedueto the shorterrecalltimepersyllable in thisexperiment, or to theinfluence
of socialfacilitationin the earlier one.
Trials toLearn Interpolated ListThemeannumber oftrialsrequired to learntheinterpolated listtoa criterion ofSoutof12rightresponses is presented foreachofthefourconditions inTable 5.5.Ingeneral, thedata
corroborate thoseofExperiment II.There, fewerresponses werelearned in Condition I thanintheotherthreeconditions, which didnotdiffer reliably
fromoneanother. Here,moretrialsarerequired to learnto thecriterion in ConditionI thanin any othercondition. ButConditionIV alsodiffers widelyfromtheotherthree,requiring fewertrials.Thedifference between Conditions II andIIIis whollyunreliable. 11 Theseresultstendto confirm
theprediction thatlearning ofa second listwillbeharder asthedegree of generalization betweenthefirstandthe secondlistincreases.
Retention of Original List Theretention of the original listin thefour conditions maybemeasured infourdifferent waysby a firstrecallscore, a secondrecallscore,andby twodifferent relearning criteria. InTable5.6
are presentedthe resultsfor Recall1 and2. Thetrendsin the resultsof
Experiment IIareconfirmed beyonda doubt,sincea gradient inrecall from Condition I to Condition IVis againapparent in bothRecall1 and2.12
When thereciprocals ofthese figures areplotted, thecurves areremarkably similar totheonesobtained inExperiment II.Inspiteofequivalence in
degree ofinterpolated learning, Condition I again shows onlyslightly
worseretentionthanConditionII.Thedifference betweenthe two averTable 5.5
Number ofTrials Required toLearn theInterpolated Listtoa Criterion of8 Correct
Responses
Condition MeanTrials to Learn
I
aav.
– 1.07
7.43
II 5.43
III 5.71
IV 3.93
–65
–77
–28
88
E. 1. Gibson
Table
5.6
RetenUon of theOriginal ListasMeasured by Recall1 andRecall 2 Condition 1 II III MeanNo.Retained, Recall 1 2.93 3.00 3.71
IV 6.21
a av.Recall I – .45 – .46 – .64 – .34 MeanNo.Retained, Recall 2 5.00 5.71 7.29 8.36 a av.Recall 2 – .56 – .62 – .67 – .43 11.Thecritical ratios(D/crdlff ) ofthedifferences are:I and11,1.60; I andIII,1.31; I andIV, 3.18;II and III,2.8;II and IV,2.12;IIIand IV, 2.18. Table
5.7
ReteMionof the OriginalListas Measuredby Relearning
Condition MeanTrialsto Relearn8 outof 12
I
aav.,Soutof12 MeanTrialsto Relearn12out of 12
a av.,12outof12
II
3.79
–.80
2.79
–48
III 2.00
IV 1.36
–41
–21
8.29
6.43
5.43
4.79
– 1.30
– 1.10
– .69
– .53
agesisunreliable. 13 Thedifferences between theotherconditions areall
fairlyreliable byonemeasure or theother.Therecallscoresthusbearout ineveryrespecttheresultsof theprevious experiment andtheexpectation of thehypothesis thatretentionshouldbe pooreras degreeof generalization between the two lists increases.
Relearning scoresintermsofthenumberoftrialsrequired to relearnto oneperfect repetition, andthenumber required to relearn S outofthe12 responses maybecalculated. Onlythelatteriscorrectly calledrelearning, since the first list was learnedoriginallyto this criterion.The results
obtained bythesetwomeasures areshowninTable5.7andFig.5.6.Both relearning measures areconsistent withtherecallmeasures inrevealing a gradientin retention fromCondition I to Condition IV.Thedifference betweenConditionI and ConditionII is relativelylargeras measuredby
relearning thanwhenit ismeasured by recall,thoughit is stillnotstatisticallyreliable. 14 The slopesof the curvesbasedon relearning are more
gradual andnearlylinearthanthosebasedon recall, butthetrendof the results could hardly be more consistentlysupported.
Analysis ofErrorsTable5.8showsthefrequency ofoccurrence oferrors
of reversionto the previouslist in interpolatedlearningand relearning. Totals,not averages,are presented,sincethe absolutenumbersare very
small. Thefigures areinconsistent, andshownogeneral trend,exceptthat
the failureof any sucherrorsto occurin ConditionIV is probablysig-
Retroactive InhibitionasFunction 89
o- - .
120UTOF 12 8 OUT OF 12
IV DEGREEOF GENERALIZATIONIN PERCENT
N~V313~ 0.1a3~lnO3~ SlVI~1. ~O~381 /\.jnN
Figure 5.6 Trials torelearn List Iasafunction ofdegree ofgeneralization between the primary and the interpolated lists . The two curves represent number of trials required to relearn to two different criteria .
Table 5 . 8 Frequency Reversion Errors in Interpolated Leaming and Relearning ,T .,.1..1..of . 0 ---~~ '~ & 'bIII Condition I II IV . I L " 1 .V Interpolated Learning 2 4 7 0 Relearning 14 23 -- 4 ~ 0 v-
90
E
.
nificant
J .
.
larger
Gibson
The
most
total
than
interesting
number
in
original original
case
the
.
probably
two
in
McKinney
not
cases
,
for
in
and
question
.
and
spontaneous
that
tain
such
fairly
an
as
of
cases
51
original
of
such
,
criterion
Verbal
Reports
of
that
it
a
of
during
,
as
in
stimulus
between
greater
the
as
have
The
a
cer
list
.
-
-
hypothesis
Experiment
within
condition
on
interpolated
the
learning
original
can
As
in
were
be
A
during
in
Experiment
interpolated
learning
lists
,
the
average
.
during
interpolated
II
members
generalization
and
than
would
.
in
time
) .
each
during
was
lists
the
,
the
.
some
interval
;
too
omission
that
,
short
must
responses
any
case
be
learning
all
the
conditions
learned
to
that
for
decision
to
,
since
and
had
they
could
,
the
same
in
none
was
blocking
case
made
among
trend
a
,
,
of
the
caused
strong
at
by
results
occurred
.
to
tendency
ordinarily
some
respond
responses
sometimes
a
Therefore
the
distributed
the
not
reported
to
of
were
such
had
so
learned
think
responses
they
done
had
not
subjects
two
.
that
who
influenced
was
but
Those
these
that
reported
.
have
there
due
,
,
not
reported
response
was
cues
could
subjects
simultaneously
make
In
subjects
rehearse
form
and
the
periods
to
visual
cues
conditions
all
them
to
of
rest
really
way
four
Almost
Most
during
impossible
without
the
Subjects
the
particular
easily
of
207
was
requires
generalization
and
longer
I
explanation
errors
in
- list
than
original
.
List
in
,
test
a
list
of latter
.
thought
in
15
intra
less
the
p
of
occurred
of
consistently
,
the
generalization
and
frequency
although
( 4
- list
recall the
permitting
a
-
the
equal
recall
of
retro
given
that
thereby
This
to in
about
single
fact
within
inter
of
errors
learning
the
.
a
from
repetitions
just
recall
accomplished
percentage
a
the
, 15
generalization
against
of
and
in
list
are
only
to
learning
assumption
high
classified
total
of
function
make
A
,
related
learning
differentiation
protective
does
experiment
be
interpolated
recovery
sumption
is
' s
) ,
errors
they
a
relearning
transfer
reversion
number
experiment
may
( 11
"
for
in
interpolated in
in
tendency
occur
Ilovert
the
increase
difference
interpolated
learning
,
mere
present
increase
between
original
the
the
the
to
McGeoch
less
from
the
is
list
noticed
than
for
McGeoch
The
intervened
II
reason
results
previous
and
also
list
The
these
the
McKinney
,
interpolated
one
'-'
is
the
.
experiment
to
the
to
learning
inhibition
of
reversions
interpolated
active
aspect
of
,
least
active
to
because
of
the
the
errors
competition
of
.
Discussion
The
ment
that
consistency
II
when
retroactive
of
repeated
inhibition
the
results
and
with
the
individual
is
a
complete
corroboration
subjects
function
of
leaves
the
degree
of
no
room
of
for
generalization
Experi
doubt
-
Retroactive Inhibition as Function
91
between primary andinterpolated list. Butaconsideration ofthedynamics oftheprocess isagain inorder , since thisfactdoes nottellustheactual role ofgeneralization ; nordoesproofof thisrolenecessarily exhaust thepossi bilitiesof explaining decrements in retention . Theprocess assumed to occurby thepresent hypothesis is briefly asfollows . Generalization mayoccurfromlist to list- therewill bea tendency forwrongresponses to occur inlearning anewlistorinrecalling anoldoneafterinterpolation . Activeintrusion of wrongresponses will thusbea factorleading to decrement . Butsincetheseintrusions areinfrequent , importance fallsonthenotionofcompetition ofresponses , which mayobviously leadto omissions or compromises . Generalized andright responses willtendto blockoneanother . Asfarasourowndatago, active generalization andcompetitive blocking seem tobesufficient explanations . A recent articlebyMeltonandIrwin(12) furnishes evidence , however , thatsomeprocess besides theonedescribed aboveoccurs in thetypical retroactive inhibition situation . Thisevidence is in thenatureof a curve showing therelationship between theamount ofretroactive inhibition and thedegree of learning of theinterpolated material . Whena separate curve wasplottedto showthedecrement directlyattributable to intrusions , the curveincreased to a maximum andthendecreased asthedegree of interpolated learning wasincreased . Thisis therelationship predicted forgeneralization by thepresent hypothesis (4, p. 217 ). Competitive blocking , according to thehypothesis , should followthesame curveasovertgeneralization , sincestrength of thegeneralized tendency is thecrucial factor inboth.Asdifferentiation isincreased , blocking aswellasovertgeneraliza tionshould decrease . Buttheresidual retroactive inhibition , afterthecurve yielded by intrusions hadbeensubtracted fromthetotal, didnotfollow thesamerelationship . Somefactoror factorsotherthangeneralization andcompetitive blocking musttherefore havecontributed to thetotal inhibitory effect . Thepresent experiment throws nolightonothercontributing causes of retroactive inhibition . Thewriterwouldlike, however , to examine the possibility ofextinction orunlearning ofthetypefoundin theconditioned response situation asafactor , since suchanhypothesis hasbeenattributed to her(12). Theconditions for extinction arepresent duringinterpolated learning , in sofarasdifferential reinforcement takesplace . If thesubject givesageneralized response , it willnotbereinforced andanother response will beindicated asthecorrect one.In thissituation , therefore , thegeneralized response maybeextinguished through differential inhibition . But thisresponse itselfis notgenerally inhibited , of course . Whena salivary response to a bellis extinguished , thereis notinhibition of salivation in general , butonlyof salivation in thatsituation or similar ones . If aparallel isdrawnin thepresent situation , thegeneralized response hasbeenextin-
92
E. 1. Gibson
guished asfarasthestimulus intheinterpolated listisconcerned; another response hasbeenlearned asthepositive oneforthatstimulus. Buthasthe extinguished response beeninhibited foritsownstimulus ofthefirstlist itspositive stimulus? Generalization ofinhibition is a well-known phenomenon(6,p. 178if.),butto thewritersknowledge it hasnotbeendemonstratedin a discrimination situation.Thereis not evenan analogue,then,to
supporttherequired assumption thattheinhibition ofgeneralized responses
duringinterpolated learning willitselfgeneralize to thepositive stimulus itemsof the firstlist.Supposesuchan assumption is made,however.It is stilla factthatveryfewgeneralized responses actuallyoccurred overtly
during interpolated learning, sothattheopportunities forinhibition of
generalized responses werefew.Itmightbeargued thatnon-reinforcement of anywrongresponse willproduce inhibitory effects, or thatimplicit generalized responses mightbe extinguished. Asfortheeffects of extinguishing anywrongresponse, it seems clearthatsuchextinction couldnot contributeto retroactiveinhibition,becauseunlessthe punishederrors were
dueto generalization, theywouldnotbe at allrelevantto recallof List1. It wouldbe necessaryto assumea conceptof free-floatinginhibitionto
finda wayforsuchnon-reinforcement to affectrecallofList1.Asforthe possibility thatinhibition mayresultfromthenon-reinforcement ofimplicit generalized responses, thefollowing arguments presentthemselves. If a subject thinksofa generalized response, butdoesnotspeakit,oneoftwo
things isprobably happening. Ifitistheonlyresponse hethinks of,hedoes notspeak itbecause healready knows ittobewrong, inwhich casehehas differentiated it. In the writersopinion, the essential condition for retroactiveinhibitionis therebyremoved.If it occursalongwith anotherpos-
sibleresponse,the situationwhichhas beenreferredto as competitive blocking results.In so faras thecompeting generalized responseis nonreinforced, it maybe extinguished in thatsituation. Thepossibilities of extinctionare thereforelimitedto overt generalizedresponsesand compet-
inggeneralized tendencies. Sincein boththesecasesit is necessary to assume generalization ofdifferential inhibition to thepositive stimulus, the theoreticalstructurerequiredis elaborateandnot too plausible. Othertheoretical possibilities forretroactive inhibition mustbe keptin
mindat thepresenttime,especially becauseit hasbeenshownto occurin
certainsituationswhichare difficultto analyzein the presentterminology.
Retention offormsinprimarymemory, especially, maybementioned. Here Gibsonand Raffel(5)have shownretroactiveinhibitoryeffectsfor the
reproducing ofa givenvisualformwhenif hasbeensucceeded byother forms.The retroactiveinhibitoryeffectincreasesprogressivelywith an increasein the numberof differentinterpolatedforms.Subjectsreporteda
blotting-outofthememory imageasnewformsmadetheirappearance. In this situationit seemslikely,therefore,that the principallocusof the
Retroactive Inhibition as Function
93
decrement occurred while theinterpolated impressions werefaking place, andthatinterference atthetimeofrecall played a small part.Norcould
inhibition resulting from differential reinforcement beinvolved. Itispossiblethatwhencontentretention isprimarily involved, asinthissituation (and presumably tosome degree inanylearning situation), aweakening of theoriginal impression during interpolation isanimportant factor. The weakening effect does notnecessarily depend upon theinterruption ofperseveration, ashasoftenbeenassumed; it mighttaketheformofdistortion
or deletionof the originaltrace.
Inconclusion, thewriter feels thatadetermination oftheroleandweight
ofthevarious factors suggested canonlybemadeaftercareful considera-
tionhasledtoprediction oftheeffects which theyshould produce when
theusualdimensions (similarity, degree oflearning, etc.)arevariedinthe
retroactive inhibition situation. Conclusions
1. Whensyllableresponses havebeenlearnedto a listof stimulus
forms, variations fromtheseforms maybeshown togeneralize with theoriginal forms, since thevariations areresponded toina testseries asiftheyweretheoriginals. Gradients ofgeneralization intermsof
frequency of response to thevariations maybe demonstrated, and
suchgradients correspond withgradients ofperceptual similarity. 2. Asthedegree ofgeneralization between corresponding stimulus members ofafirstandasecond listisincreased, when responses inthe
twolistsaredifferent, thereis a tendencyfor the secondlistto be harderto learn.Thesecondlistishardestto learnwhenthestimulus
membersof the two lists are identical.
3. Thedegree ofretroactive inhibition ofa firstlistvaries directly
withthestrengthof thetendency forstimulus members ofaninter-
polated listto generalize withthoseofanoriginal list. 4. Errors of overtgeneralization, or reversion to theprevious list occur during bothinterpolated learning andrelearning oftheprimary list.There issometendency forfrequency ofsucherrors tovarywith the degreeof generalization betweenthetwo lists.
5. Errors ofintra-list generalization occur during thelearning ofboth
theprimary andthe interpolated list,but sucherrorsdecrease considerably fromprimary to interpolated learning. Notes
1.Thisexperiment isoneofa series ofstudies presented totheFaculty oftheGraduate
School ofYale University inpartial fulfillment oftherequirements forthedegree of
94
E. J. Gibson
Doctor ofPhilosophy. Thewriter wishes toacknowledge thevaluable advice andassistanceof ProfessorClarkL.Hullduringthe progressof thiswork.
2.Thisexperiment wasreported at the1938meeting of theAmerican Psychological Association
(3).
3.Thiscriterion is appropriate to ourdefinition ofgeneralizafionthe tendency fora
response Ralearned toSato occur when5b (withwhich it hasnotbeenpreviously associated) is presented (4, p. 204).
4.Hilgard andMarquis (6,p. 182)present evidence to showthatwithconditioned responses thecurve ofgeneralized responses differs withdifferent stages ofpractice.
5.Themeansforthefourlistscombined are:Standard forms,2.47;firstdegreesimilar-
ity,1.12; second degree similarity, .33.Themaximum response would be3,since each testlistcontained formsfromeachofthefourgroups. Thedataarepresented indetail in (2).
6.Fiverepetitions werechosen onthebasis ofa preliminary experiment. Itallowed only partial learning ofthelists(about Soutof13items) which wasdesirable inorder to securea highdegreeof retroactiveinhibition(5).
7.The criticalratio(D/o ff)
of the difference betweenConditions I and IV is 2.84
whendegree of learning is measured bythenumber recalled at trial5, and3.75 whenmeasuredby the total numberrecalled. 8.Theretroactive inhibition scores havebeencalculated bytheusualformula fromthe percent retained, ratherthandirectly fromthenumber recalled. 9.Critical ratios(D/adff) basedonRecall 1are:I andIII,1.69;I andIV,3.83;I andV,5.56 IIandIII,1.56;IIandIV,3.51; IIandV,5.17;IIIandIV,2.11; IIIandV,3.96.These statistical reliabilities areingeneral satisfactory, considering thata gradient isinvolved. Criticalratiosfor Recall2 rangefrom1.18to 3.07whenalternateconditions are compared.
10.Thisproblem hasbeentakenupinaninvestigation tobepublished. 11.Thecritical ratios(D/o ) 1 ofthedifferences are:I andII,1.60; I andIII,1.31,I andIV, 3.18;II and III,.28;II and IV,2.12;Ill and IV,2.18.
12.Critical ratios(D/ad ff) between alternate conditions are:I andIII,.99firstrecall, 2.61 secondrecall;II andIV,5.61firstrecall,3.54secondrecall.
13.Since interpolated learning wascarried toacriterion, andthetotalinterval ofinterpolationheldconstant, lesstimeintervened between cessation ofinterpolated learning and recall oftheoriginal listinCondition I thanintheotherconditions, giving lessoppor-
tunityforspontaneous recovery ofgeneralization in thiscondition. If spontaneous
recovery isanimportant variable, itwould thentendtolower therecall scores ofthe otherthreeconditions ascompared withCondition I, thusbringing themallcloserto ConditionI, wheredifferentiation wasmostrecentlyachieved.
14.Thecritical ratio(D/o ff) is 1.07for8 outof 12,and1.09for12outof 12. References
1.Britt,S.H.,Retroactive inhibition: a reviewof theliterature, Psycho!. Bull.,1935,32, 381440.
2. Gibson, F.J.,A systematic application oftheconcepts ofgeneralization anddifferentiation to verballearning, Dissertation on filein YaleUniversity Library.
3.Gibson, F.J,,Retroactive inhibition asa function ofdegree ofgeneralization, Psycho!. Bull., 1935, 35, 626.
4.Gibson, E.J.,A systematic application oftheconcepts ofgeneralization anddifferentiation to verballearning,Psycho!. Rev.,1940,47, 196229.
Retroactive InhibitionasFunction 95
Retrospectand Prospect:Are TheoriesRecycled?
Some of the concernsthat were motivating psychologists in the thirties are in one way or another alive and well today . If we moved away from them, we have moved back again with some revision . Functionalism, for
example, is one of them. It was discovered, during the fifties, that conditioning does not happen fortuitously and inevitably according to some simple formula for attaching a responseto a stimulus with frequent repetition. Some situations are definitely favored for conditioning to certain
responses dependingon the speciesof animaland its way of life. Garcia , investigating learning of taste aversion in rats, emphasizedthe evolutionary basisof favored kinds of learning for a species(Garcia, McGowan and Green 1972). An animal learnswhat is functional and adaptive for its species.Present-day ecologically oriented views of animal learning abound (Bolles and Beecher 1988, Johnston and Pietrewicz 1985). On
the other
hand , connectionism is also with
us again , in an even
sterner form than Thorndike or Hull could have imagined . Neural nets are
thought to be shapedby repeated exerciseof pathways in the network, without (as I understandit) any recourseto ultimate function or adaptive necessity (Rumelhart and McClelland 1986). This new form of connection -
ism emphasizesmicrostructureand mechanism , and contrastssharply with the functional, evolutionary approachof the ecologicalbiologists. The preoccupationof the thirtiesand fortieswith learningtheory gave way as the IIcognitive
revolution " came in in the late fifties and the sixties and in -
formation processing took over . The boxes in the flowcharts were labelled with terms like attention and memory (indeed, a proliferation of memories ),
but never learning. The cognitive psychologiststo this day have neglected learning, leaving it to the biologists, neuropsychologists, and Skinnerians. Cognitive psychology seemsto derive more from a structuralist legacy than a functional one. As the ecological view strengthens today and spreadsto humans, concern with learning can be expected to strengthen
too I believe . Thistime, perhaps , perceptual learningwill be in theforefront.
II Comparative Research on Learningand Development(1952- 1970)
Introduction to Part II
- ~ In the introductionto part I, I pointedout the significance of a comparative approachfor functionalpsychologyin Americaand the prevalenceof research on animalsaspsychologybecame accepted asa science . It wasthis domain,in fact, that first attractedme to psychologyasan undergraduate at Smith College. It was a severedisappointment to arrive at Yale as a graduatestudentandfind that its primatelaboratory,underthedirectionof Yerkes , wasout of boundsfor women.Thereweretwo youngerprofessors in Yerkes 's laboratory , Henry W. Nissenand CarlyleJacobsen , who were sympatheticwith my second -classstatusand found me minor ways to maintaina concernwith comparative psychology . I watchedandeventuallyassisteda bit in brainsurgeryfocusedon the motor systemin cats (extirpationwas the style of the day) with a Dr. Marshallin the medicalschool. I alsoconductedexperiments on maternal behaviorin micefor a Dr. LeBlond , a postdoctoralresearcher from France who wasvisiting at the medicalschool. The projectinvolvedthe role of prolactinin instigationof maternalbehavior.The ideawas to inject prolactin in maturevirgin femaleanimalsand test their behaviordaily for evidenceof ensuingmaternalbehavior.The testinvolvedpresentingthem with pups of variousages(borrowedfrom other cages ) and observing retrieval to the nest. The older the pups retrievedand the larger the numberretrieved , the strongerthe maternaldrivewaspresumed to be. But the adultmicewerenecessarily exposedmoreandmoreto the presence of infant animalsas the experimentcontinued , so I introduceda control groupof adultfemalesthatreceivedno prolactinbut wastesteddaily in the samemanner . All the animals , controlsand experimentals , exhibitedincreasinglystrongmaternalbehavior.LeBlondpublishedthe data, but without my name,perhapsbecause of a contretemps that mayor maynot have beenmy fault. Themicebelongedto a colonybredfor someotherproject, I think involving cancer , andit wasimportantto returnthe litters to their
102
PartII
proper home cagesafter using them for testing. Somemonths after 1began working in the lab, a litter of mice with brown coats grew up in a cage with
white parents. Although 1 had never cleanedthe cagesand had been very careful (I thought) in restoring litters to their home cages, 1 was accused of negligence. 1 was pretty sure the negligencewas on the part of service people who cleanedthe cages, and who had a high turnover. 1 finished the experiment but received no thanks and retired from the scene.
Although 1taught a coursein comparativepsychology for a year or two after returning to Smith, there were inadequatefacilities for animal research there . 1 had to wait until after World
War II , when we moved to Cornell
in
1949, for my chanceto do researchwith animalsagain. 1 accompaniedmy husband, although not as a faculty member (Cornell had nepotism rules in those days). SinceI had no laboratory to work in, 1 was glad to acceptan invitation
to work
as research associate in the laboratory
dell , at a sort of farm known
as the Behavior
of Howard
Lid -
Farm . The animals available
were farm animals, sheep and goats. Liddell's research was supported lavishly by the RockefellerFoundationand it was supposedto focus on the
so-calledexperimental neurosis . Thatwasnot whatI wasinterestedin, but 1was happy to have the opportunity to work with animalsagain, even if 1 could only do so at someone else's invitation .
Themethodemployedfor producingexperimental neuroses in the sheep and goats was devised by Liddell from the classical conditioning
para-
digm- the animal was given a signal such as a buzzer, followed by shock (inescapable ) to a forepaw. A daily routine of this procedure supposedly resultedin a neurosis, diagnosedby rapid heart rate and irregular breathing. Recordswere accordingly taken daily of heart rate and breathing, and these were indeed disturbed, but the animalswould also struggle to get away when they saw an experimenter coming , which seemed to me a mark of
quite reasonabledislike of a very uncomfortableprocedure. I managedto perform a study of conditioning with avoidable and unavoidable shock (reprinted below), which fit nicely into a controversy current in learning theory at the time and so reinstated my earlier interest in learning . My main interest during the two years I spent at the farm was in a
project I beganin the secondyear on maternalbonding of newborn kids to their mothers. Goats have the convenient habit of producing young in
pairs, sothe twinscouldbe splitup in experimental andcontrolgroupsand given different rearing treatments . 1 attended the births (often on cold February nights ) of eight pairs of twins , destined to be reared with their own mothers , or foster mothers , or a peer group , or alone (the latter two
groups fed artificially from a nipple pail). A series of observations were madefrom birth at frequentintervals to observeevidenceof imprinting and maternal-kid interactionsof various kinds. A couple of interesting observations came out of the research before it came to a distressing (for me) end.
Research on Learning andDevelopment (19521970) 103
I wasparticularly interestedin chemical information as a factorin bonding,andso at somebirthsI removedthe newbornkidfromthe mother
beforeshecouldtouchit andbeforetheappearance oftheafterbirth.The
kidwasthenbathedin a detergent. Ononesuchoccasion, I hadjust
completed thefirstkidsbathwhen itstwinbegan tomake anappearance. What todowiththefreshly bathed oneinahurry? Thefarm manager, who
waswatchingfroma half-door, saidput it on thestand. Thestandwasa
veryhighcamera standwitha pedestal surface abouta footsquare. I said, Butwontit falloff?Heassured methatit would not.I stoodthedamp littleanimal onthestand,andit remained there,upright, looking around theroom, untilIcould carryitofftoitsassigned place. Goats areprepared frombirthtosurvive onaprecipice, andthislesson wasagoodpreparation for the visual cliff research later on. It wasalsoquiteclearfroma numberof observations that the maternal
goat,if deprived of its offspring forevena fewhoursafterbirth,didnot
welcome it andinfactwould buttit rudely awayif it approached and
attemptedto nurse.Lickingthe newbornkid and otherchemical inter-
changes areimportant (asanysheepfarmer couldprobably havetold
me).Thekid,on the otherhand,wouldapproachany adultfemalefor
several daysafterbirth.Imprinting didnotoccurveryearly,butit would
eventuallyto its peergroupor a humancaretaker, if its ownor a foster
motherwas not provided.
Thisresearch ceasedin the springof my secondyearat the Behavior
Farm whenI discovered aftertheEaster weekend thatsomeofmycarefully rearedsubjects hadbeengivenawayasEaster presents by thefarmmanager.It wasnevercompleted andso notwrittenup,although I havea demonstration movieto showforit.Thatdisappointment turnedmeto a moretraditional kindof research thatI describe in partIII.My next periodof comparative research beganfouryearslaterafterRichard Walk
came toCornell. Weembarked together onaseries ofrearing experiments, thistimewithrats.Atleast, noonewanted themforEaster presents. Inthelate1940sa stronginteresthadgrownintheroleofenvironmental
conditions during earlyrearing fordevelopment ofsensory processes and
perception. Hebbsworkwithdark-reared rats(1937) wasa precursor of thisconcern; in 1947,Riesens reportontwochimpanzees rearedindarknessfrombirthto sixmonths waspublished andproduced a realstir.If
impoverishment ofrearing conditions could havedamaging effects onperception, might notenrichedconditions havebeneficial ones? Howmight environmental conditions function to constrain perceptual development? WalkandI,withtwosuperb research assistants, planned andcarried outa series ofrearing experiments withhooded rats,providing quite specifically enrichedenvironments forthemtoviewastheygrewtomaturity, and
after three months testing their perceptual competence withaveryspecifi-
104
Part II
callyrelateddiscrimination task.Moreof thislater,whenI commenton theseexperiments. Theystruckme aftercompletion as the not too productiveresultofjumpingon a bandwagon, but nowin retrospect I think theyprovidea usefulmoral.Development is enormously flexible. Theenvironmental milieu does indeed constrain development, but perception is
not constructedover timefromelementarypieces.Two of the fivepapers
stemming fromthisresearchthefirstandthelastare presented with a backwards glanceandreconsideration of whattheymean.Latein the program ofrearing experiments, WalkandI decided toraisea groupofrats in the darkso as to providea majorcontrastwithour enrichment groups.
Darkrearing animals is a greatdealoftrouble. It is nojoketo cleanthe cagesandfeedandwateranimals in totaldarkness. Wedecided thatwe
shouldmakeall this troublepay offby testingthe dark-rearedrats in more than one situation.What shouldthat be, in additionto the usualdiscrimina-
tionlearning? Walksuggested depthdiscrimination, andhereminded meof
a storyI hadtoldhimaboutdrivingeastfromCalifornia withour two children andworryingthattheyounger,onlytwoyearsold,wouldfall
overa cliffas we picnickedon the edgeof canyons.Thuswasconceived the idea of a clifftest for our dark-rearedrats:wouldthey walkover the
edgeofa cliff, forlackofseeing edgesduring theirearlydevelopment? This
experiment wasthemostfunofanyI haveeverhada handin.Wewent onto perform a trulycomparative experiment, withasmanyanimalspecies as we could collect.
Webeganwithhoodedratsbecause thatwastheanimal wehadbeen rearingin the dark.Wemovedon to albinoratsto compare animals witha similarlifestylebut poorervision,andthenbranchedoutto observe
babychicks, ungulates (lambs andkids),puppies, andkittens. Thebaby chicksand the kids were particularlyinteresting,becausethey were pre-
cocialanimals, capableof locomotion at birthor minutesthereafter. They couldbe observedon the cliffbeforethey had an opportunityto learn,
withoutbeingdeprivedofa normalenvironment whiletheywereattaining
independent locomotion. Boththechicks andthekidsshunned thedeep side of the cliff,but moved onto and freelyaround the shallowside.
Obviously, therearespecies thatavoida drop-off withoutlearning from postnatal experience ofanykind.Forthem,onecanpresumably claim that perceptually controlledavoidanceof a drop-offis innate.
Naturally, wewenton to observehumaninfantson a visualcliff.We
advertised in the localnewspaper for crawlingbabies,providinga tele-
phonenumberforparentswhowishedtheiroffspring to takepartin an experiment. Several colleagues (including myhusband) predicted thatwe wouldgetno answers, thatparentswouldfeartheirbabieswouldreceive shocksor unpleasant tests.Butthetelephone rangfuriously andafterthe experiment wasexplained to them,mostparentsbroughtthebabiesin.We
Research onLearning andDevelopment (19521970) 105
observed onlycrawling infants inthoseexperiments, though onsetand length ofcrawling experience varied. Only three outofatotalofthirty-six infants crawled offonthedeepside,although allparents beseeched their babies tocome overit.While wecould notclaim thatthetendency to avoidthedeepsidewasinnate, sincetheinfants hadenjoyed sixto ten months ofgrowing upinalighted, social environment, itseemed extremely unlikely thattheyhadlearned to avoidthedeepsidethrough fallsor punishment ofanykind. Parents, when questioned, seldom reported opportunitiesfor suchlearning. Iflearning isimplicated inavoiding a cliff inhuman infants, it seems
likely tomethatithappens when aninfant firstattempts locomotion, inthe
courseofexploring a surface (before movingontoit)to discover whatkind
ofsupport it affords. Thevisible andpalpable qualities ofthesurface are sampled andpropulsion oftheentire bodyontoasurface occurs onlywhen thereisa surface perceived asaffording support. Mylaterresearch (dis-
cussedin partVI)is relatedto thisearlierresearch withthecliffwhich
showed, perhaps forthefirsttime, theclose tiebetween perception and action inthehuman infant. Thiswasanexperiment onperceived affordances foraction, although wedidnotconceive ifprimarily thatwayat thetime. Perception doesnt havetobecome meaningful through associative
learning, nordoesif guideactionbybeingassociated viaS-Rbonds. It is worthmentioning thatthroughout allthisresearch wewerenever struck byananimals apparent fearofthedeepsideofthecliff. Allthenon-
human animals, to theextentthattheycouldseeit,simply avoided it.
Humaninfantsoccasionally cried,butthatwasattributable to a frustrated urgeto get to theirmothers,whowerecallingto themacrosstheinvisible
surface. Laterworkby Campos (Campos, Langer, andKrowifz 1970),
supported theobservation thatfearofdrop-offs waslearned, probably after self-initiated locomotion iswell under way. Butbythattime, itcould easily
belearnedfromanxiousparents. DuringtheseyearsI hadnolaboratory ofmyownandhadto collabo-
ratewitha colleague onthefaculty, I wrotegrantproposals andhad
graduate research assistants, butcould notsigntheforms asprincipal investigator or adviser.I hada tinyofficeon the fourthfloorof Morrill Hall, where thegraduate students andtheanimals were housed. I enjoyed thecompany ofboththese groups. Itwasa goodatmosphere forresearch
to flourish, evenwithoutanempireof onesown.
The Role of Shock in Reinforcement EleanorJ. Gibson
Traditionally, a classical conditioned reflex issupposed toconsist oftheoriginal
unconditioned reflex becoming instigated bysome previously neutral stimulus (the
conditioned stimulus)afterbeingpairedmanytimeswiththeunconditioned
stimulus anditsensuing unconditioned response. Myearlier experience with finger withdrawal toshock (paper 1)hadtaught methattheformoftheresponse was notinvariably thesame andthatitwaseven transferable toanother appendage ormuscle group. Watching thebehavior ofanimals persistently shocked ona forelimb wasnevertheless a revelation. Theanimals thatwesubjected toconditioning exhibited a number ofpatterns ofbehavior, sometimes going through a
gamutof responses that appeared almostritualistic. Thebehavior wasnotme-
chanical, buttheanimals could notgetaway.Whatwasgoing onwhen shock
wasdelivered inescapably, overandover?Theconditioned stimulus wasa warning,buthowcould it beuseful andthusleadtoadaptive behavior? Theessential contradiction inconditioning withshockthatcannotbeavoided occurrence oflegflexion totheconditioned stimulus thatisimmediately punishedbyensuing shockhadalready beenthesubject ofconsiderable debate and wasresponsible fora two-factor ortwo-phase theory ofconditioning (Mowrer 1947). Theideawasthattheretakes place, first,classical conditioning ofinternal
emotional responses (autonomic responses such asquickened heart rate,respira-
tion,etc.)andthatmotor responses (e.g., legflexion) areinduced bythisstate.The motor responsebehaviorwould presumably bedefense reactions characteristic
ofthespecies, suchasfreezing orrunning away. Ifthisbehavior were successful
inavoiding theshock, it would recur,become habitual. Ifnot,theanimal should runthrough whatever repertory ofdefense reactions itsnatureandthesituation
permitted. Thesituation I hadavailable wasideal forobserving theresponse repertory andtheanimals behavior as theconditioning procedure wasadmin-
istered andcontinued daily, because theanimals were notrestrained ina frame Journal ofComparative andPhysological Psychology, 1952,45,1830.
108
E. J. Gibson
and couldmoveaboutwithinthefair-sized roomwherethe experiment took place.
Theanimalsdid indeedexhibitan extensive repertory of actions,suchas backwardlocomotion fretreat),rearingand head-lowering, crouching, freezing
(inhibition ofmovement), andotherpostures thatcouldbeunderstood asmanifes-
tations ofdefense oravoidance. I would reword mydescription ofwhattook place
now,usinga terminology taken frommyhusbands theory ofaffordancesthese animalswerelearning whattheconditioned stimulus (andindeed, thewhole situation) afforded. Theirbehavior washighlycomplex andvariable as they responded withdefense andavoidance actions, gradually settling downtogreater economy ofaction astheylearned thatattempts atgrosslocomotion wereineffective.Thebehavior wasa far cryfroma simpleconditioned reflex.
Onemightaskwhether theanimals werelearning to be afraidofthesitua-
tion.Weseemto thisdayto understand ratherlittleaboutfears(maybe about
emotions ingeneral). Thisexperiment certainly convinced methatfearscanbe learned.Later researchwith the visualcliffand with a looming situation
gradually inclined metotheviewthatmost fearsarelearned, although I doubt thatconditioning (asinthepresumed learning oflittleAlbert)istheanswer. Building a theory of learning ona concatenation ofconditioned reflexes is
clearly insupportable ifa so-called conditioned reflex isitself a complex learned actionthat canshiftitsformin an animalseffortto achieve adaptation to
circumstances that it cannotcontrol.A goodtheoryof howperception of affordancesis learnedandguidesbehavioris still badlyneeded . Does electricshockact as a reinforcingagent to strengthena conditioned
response? 12 Thispurelyexperimental question hasfar-reaching implications.Shockmayberegardedaspunishment. Theeffectsofpunishment are of intenseinterestto socialand childpsychologists; but thereare as yet no
commonly accepted, well-founded principles oftheoperation oreffects of punishment. Thewriter became interested inthisproblem inthecourse of training animals inoneoftheroutines usedat theCornell Behavior Farm fordeveloping neurosesin animals. Theroutineinvolves a classical (5)
Pavlovian conditioning technique, withshockto a forelegastheinevitable unconditionedstimulus.As the observer watches the relentless succession
of metronome-beats followedby shock,the questionpresentsitselfmore
andmoreinsistently, Whyshouldtheanimalflexitsforelegto thesound
of the metronome? Thereactionyieldsit nothing,apparently, but another shock.Yetthisroutinehasbeena traditional methodof establishing conditionedresponses. Froma Pavlovianstandpoint, a conditioned response
(legflexion) ispredicted. Theshock regularly produces theflexion; hence, in termsof a contiguity-substitution theory,themetronome shoulddo so aftera numberof pairedstimulations. Butsuchan expectation is wholly
Roleof Shockin Reinforcement 109
contradictory toeffect theory. From thestandpoint ofthetraditional law ofeffect, shock isunpleasant andistherefore anegative reinforcement; it
should tend tosuppress anyaction preceding it.Iftheanimal begins tolift
itslegto themetronome, andtheshock follows, theactionshould be discontinued.
This paradox hasbeen thesubject ofseveral experiments designed evitably with conditioning when theCRavoids theshock. Schlosberg (17,18,8)performed experiments withwhite ratscomparing inevitableshock withavoidance andfound little difference intheefficiency withwhich theprocedures produced a conditioned motor response (e.g., tailorfootwithdrawal). Infact, theCRdeveloped veryslowly to compare theeffectiveness of conditioning whenshockoccursin-
andremained unstable underbothconditions. Butevidence ofemo-
tional conditioning (sharp inspiration andsquealing) wasclearly pre-
sent in both cases.
Twoyearslater,Brogden, Lipman, andCuller(2)foundshock
avoidance much more effective thaninevitable shock forproducing
conditioned running (wheel-Lurning) intheguinea pig.Anavoidance grouplearned veryquickly toa criterion of100percent; butthenonavoidance groupneverroseabove 50percent andshowed, furthermore, anerratic learning curve. In1918, Sheffield (20)suggested that
thepigsactual response toshock might often have been stopping (freezing) rather thanrunning. Insuch anevent, asubstitution theory would notpredict thatrunning should constantly prevail asa CR. Sheffield applied Guthries postremitytheoryto thesituation and attempted totesttheprediction thata CRona given trialshould
repeat theURofthejustprevious trial(running orotherbehavior). Hefound thatwhether theanimal ranorstopped toshock increased ordecreased, respectively, theprobability ofa conditioned runtothe
CSonthefollowing trial.Butconditioned running roseonlyto52 Brogden, Lipman, andCullers study (2)itroseto50percent without suchselection ofcases. Also, conditioned running deteriorated inthe percentwhencasesfollowing othertypesof URwereomitted. In
laterstages oftraining withunavoidable shock. Since otherfactors are
required toexplain these results, contiguity seems unpromising asa singleexplanatory principle. These twotheoretical positions (contiguity-substitution andeffect) do equivalent to drivereduction (6)andisthereby reinforcing. A recent experiment reported byMowrer (16, pp.278if.)renders thisposition less attractive, however. There is, finally, the possibility of a two-phase or two-factor theory(19,3,21,13,15). notexhaust thepossibilities. It canbeassumed thatcessation ofshock is
110
E. J. Gibson
Thewriterproposes to develop theimplications ofa two-phase theory
for the unavoidableshocksituation.(a) It is assumed,first,that an emo-
tionalresponse, involving the typicalinternalemergencyreactions, is
quickly conditioned to theCS.Tosupport suchanassumption, thereis evidencefromthis laboratory, as wellas others,for the occurrence of increasedheartrate,irregularrespiration, and psychogalvanic reactionto
theCS,anticipating theUS(10,pp.510if.).(b)It isassumed, second, that
thisemotional stateinducesaction,in allprobability an escapeor defense reactioncharacteristic of the species.It is likelythat a repertoryof such reactions existsforanygivenspecies, perhaps ina hierarchical relationship. 3
Inanavoidance experiment, theshockplaysa singlerole.It reinforces the conditionedemotional reaction to the signal,which,in turn, instigates one
of theresponses in theanimals escape-defense repertory,as weshall tentatively nameit.If theresponse is successful in avoiding theshock,it willpresumably recur.Itsrecurrence maybe dueeitherto lackof interference bya competing reaction, inasmuch asnopunishment results, orto
positive reinforcement byfearreduction (14).Ontheotherhand,with inevitableshock,the shockplays a dual role. It functionsto promote one
oftherepertory of emergency reactions, but,also,it tendsto suppress or
inhibit thisveryreaction assoonasithasoccurred. Aftera number ofsuch inhibitory effects, a particular reaction maybe eliminated entirely, and anothermemberof the emergency repertorytakesover.Theresultwould
beanappearance oftrialanderror.Ifourassumptions arecorrect, a number ofpredictions canbemadeforthesituation whereshockoccurs inevitably 100percent ofthetime.(a)Thereshould bea rotation orshiftofresponses
through theanimals escape-defense repertory (akindoftrialanderrorof
histotalrepertory); (b)thereshould bea number ofcasesoftotalinhibition ofresponse, aftera motorresponse to CShasfirstappeared; (c)inhibitions
ofthistypeshould terminate ina shorttimebecause theprimarily conditionedemotionalstateis constantly strengthened by the sameshockthat
tendstosuppress themotorresponse; and(d)thereshould begreatfluctuationor variability ofresponse intheunavoidable shockgroup,contrasted
withincreasing uniformity inanavoidance group. A qualitative studyof theresponses astheydevelop inthetwosituations should provide acheck of these predictions. Method
Fourteen younggoats(between 1 and2 months oldat thebeginning ofthe experiment) wereconditioned in a relatively freesituation. Theywerenot re-
strainedin a Pavlovframe,but couldmovenormallyto any part of the experi-
mentalroom,whichwas10ft.squareandbareexceptfortheobservers chair. 4 TheCSwasa darkening oftheroomfor10sec.preceding theUS,a mildshock
Role of Shockin Reinforcement
111
delivered tothekidsright foreleg. Eight oftheanimals were given practice with inevitable shock, following theclassical Pavlovian procedure (group Inev), while
four were allowed toavoid theshock bysustained lifting oftheright foreleg shock was unrelated toacritical avoidance response, making itinevitable, although (group Av).Twoanimals weretrained with25percent random shock. Here, the
itoccurred only25percent ofthetime. Ourprincipal dataconsist ofdetailed, qualitative, written observations and photographs of the development of behavior through manystagesoftraining.
Theshock wasdelivered bywayoftwoelectrode bracelets ofsaline-soaked
cloth, placed oneabove theother ontheright fore-ankle. Metal clips attached the bracelets towires which coiled around theright foreleg andpassed through a buckle ontheanimals back. Thewires coiled upward, forming aspring attachment toa4-ft. long stick suspended from thecenter oftheceiling, allowing completely freelocomotion toanypartoftheroom andmaking thenecessary electrical
connections through a swivel jointsothatnoamount ofturning would leadto tangled wires. Theequipment wasnotuncomfortable, andtheanimal could walk
normally. Intensity oftheshock was regulated byavariable resistance and ranged from 10to20V.; duration was 0.2sec. The shock was adjusted soastobestrong
enough toproduce areliable foreleg flexion, butnotexcited running andvocalizationoftheanimals coat, andsometimes it wasstrong enough toproduce a jumpingmovement (bothforefeetoffthefloor)in theanimal. The overhead suspension served asecond purpose. The 4-ft. stick which, owing tion.Theintensity did,nevertheless, varyslightly withfactors suchasthecondi-
toitsspring attachment totheanimal, always pointstoward it asit moves about
theroom, ismade tomove twoself-synchronous motors, oneforeach ofthetwo dimensions oftheplane ofthefloor. Each ofthese ispaired withasimilar motor ina recording instrument andmoves a pencil toconform withthemovements of thestick. Agraphic recording wasthussecured oftheanimals totallocomotion
about therooma kind ofminiature map ofitsroute and movements throughout the experimental hour. Thesignal (CS) wasa dimming oftherooms illumination. During intervals between trials, theroom was illuminated bya300-w. bulb; atthebeginning ofthe
signal, this light went off, and adim light, barely enough toallow theexperimenter
toperceive theanimals movement, remained on.Theduration ofthissignal was 10sec., followed immediately bytheshock, butnotoverlapping it.Signals and shocks were timed anddelivered byanautomatic clocking device atregular 2-mm. intervals. Theshock fortheavoidance subjects wascontrolled bymeans ofan experimenters key. Theexperimenter watched theanimal closely, raising thekey whenever theanimal raised itsleg,sothatthecircuit wasbroken iftheanimal had itsfootoffthefloor attheendofthesignal. Theanimal wasnotallowed toavoid theshock iftheforeleg wasreturned tothefloor again before thelight came on. Thekeywasalso used foranimals receiving 25percent shock, though inthiscase
theshock wasgiven according toarandom arrangement without regard tothe Atthebeginning oftheexperiment, thekids were brought totheexperimental
animals behavior.
room andthewires attached fortwoadaptation periods without signals orshocks. Theanimals veryquickly became adjusted totheroom andtheharness andceased
112
E. J. Gibson
tostruggle during harnessing. When training began, theyweregiven 20signals perday,theexperimental period lasting 40ruin. Asa ruletheywereruntwoto threetimesa week,thoughthistiming variedslightly withtheseason, fortraining wenton intothesummer. Allanimals received 25 daysof training(500trials). Somewerecarried to 1,000trials,butonlythefirst500willbeanalyzed here.The animalsliveda normallifewiththe restof the herd,in pastureor barn,except during experimental sessions. Results
Varietiesof Reactionto the CS
Ourfirstquestion is,Whatkindsofresponse weremadetotheCS?Dida family or repertory of defense reactions emerge, or wastheremerely foreleg flexion? Theresults definitely indicate a varied repertory. Aclassificationof thedifferent patternsofresponse to theCSthrough25daysof
training hasbeenmade fromprotocols oftheInevgroup. Thefirstdayof training hasbeendisregarded inthedatapresented, because it washardto judge atwhatmoment during thefirst20trials theanimals behavior began to be affected by theCS.Theobserver, of course, alwayshadto makea
judgment astowhether theanimal wasaffected atallbytheCS,butthis wasveryeasyafterthefirstday, for spontaneous activity rapidly decreased. observed follows: A catalogue of the responses
1. Walking or runningbackward, i.e.,retreatingfromthe apparent locus of the shock
2. Walkingor runningforward.
3. TA/heeling (circling)to right or left. 4. Side-stepping.
5. Independent leg-movements without locomotion, e.g.,pawing thefloor, tapping,markingtime,steppingin place.
6. Flexion of eitherforeleg.Flexionvariedas to pattern,but we defined it asliftinga singlelegoffthefloorwithbentknee,whether thelegwasliftedforward or retracted backward (seeFig.6.1). 7. Extension. Extension alsovariedin direction. Wedefinedit as stiff or rigidliftingof the legfromthe shoulder(Fig6.1).
8. Humping withheadlowered to ground. A slightly bowedhead wascharacteristic of manyreactions, but an exaggerated loweringof theheadusuallyaccompanied a humpedback.Thispeculiar postureis
occasionally observed inthepasture ina frightened orangryanimal. 9. Rearing. Rearingis alsocharacteristic of a curious, excited,or
frightened goat.Oneanimal (seeFigure 6.1)reared andwalked back-
wardon its hindlegs.Rearingsustained by leaningwithforefeet againstthewallwasmorefrequently observed.
Role of Shock in Reinforcement
113
Figure 6.1
Fourofthevarieties ofreaction to theCS:a,backward flexion; b,forward extension;
c,rearingwithbackward locomotion; d,humping withloweredhead.
Crouching, lyingdown,andbuttingwereinfrequently observed. Combi-
nations among thenineresponses abovedidoccur, forinstance, wheeling alternately fromrighttoleft,orwalking backward followed byflexion. It isclearthatmanyforms ofaction appear inresponse tothesignal whenthe
animal is not restrained in its movements.
Inhibitionof Response
It waspredicted thatcasesof totalinhibition of motorresponse should sometimes occurduringtrainingin the Inevgroup.Suchinhibition did,in
fact,appearin 24 percentof the trials.Included in thiscountwerecases
where theanimal madeslight headorearmovements butnolegmovement or grossposturaladjustment. Thisabsenceof responsedid not looklike
lackofanyreaction, forevenwhentheanimal stoodcompletely still,itwas frequently notedtolookrigid,tense, oras ifitwerepressing hardonthe
114
E. J. Gibson
floor." Inhibition seldom occurred on more than four successivetrials, and it was generally followed after one or two trials by action again. Frequency of ReactionPatterns There was in the repertory of reactionssome suggestionof a hierarchy of dominance. A comparisonof the four most frequent types of reaction in the Inev group is provided in Figure 6.2. It may be seenthat walking backward is dominant over all other reactions in early training, being the most frequent responseto CS on the second, third, and fifth day of training. On day 2, it was the dominant response for seven of the eight animals. Walking or running forward appearedearly in the courseof training, being second most frequent on day 2, but its curve falls steadily thereafter. Inhibition was actually the most frequent responseon day 4. Not until day 6 did flexion of the shockedleg gain top frequency. Its curve rises as the curves for running fall, but it never becamethe exclusive reaction, occurring only on 52.5 percent of the trials over-all. In general, locomotion was reducedas training went on, both to signals and during intervals between them. Figure 6.3 shows locomotor maps for one animalmadeon the third day and the tenth day of training. By day 10, the animal has cut down both its range and amount of movement. Still later, the animal, unless disturbed, usually stood still at the wall without moving for the entire 40 min. and made single limb movementsto the CS.
120 .
11 0 Z 0
.
/ '
'
.
" " - ,' .
.,. ' _ .. / 1 00
-
.
.-
90
U -
18
Q
.' 16
a:
W
I
\
I , \
\
1
\
"
~ 14 I 1-
I
12 ' "
\
1
\
r: - -.
\
.
~ :I:
I' 1
'
I
; 8\
' .'
I
,
'.
.
\' - ' ) '
.
25 % Random
lA"
I
~ Reinforcemen
' V'
Inevitable Shock
~ 10 V "'\ .
ffi8 :J 6 ~ 4 ~ 2
\J\
m
~
.
.
\ / .
. \V', .~. I.' .- / \ .1'. ' " .~-- Avoidance .
0 2
4
6
8
10 DAYS
Frequency
of Shift
from
12 OF
14
16
18
20
22
24
TRAINING
One Reaction
Pattern
to Another
Figure 6 .4 Frequency of shifts from one reaction pattern to another under three conditions of shock .
Animal
40 shows
rather
consistent
reduction
in locomotion
and some
stereotypy on day 25, although there is one regressionto walking backward . Animal 7 exhibits no tendency at all to uniformity
and shows a
variety of reactions and frequent shifting on all four days. Avoidance animals, long before day 25, fixated one responseand repeated it consistently . It may be concluded that the data show clearly the extreme variability of reaction under inevitable shock as contrasted with increasing uniformity under avoidance . Change of Reaction with Avoidance There have been several discussions
of the form of conditioned
reaction
shown in shock avoidancewhen the shockwas presentedto the foreleg (3, 9, 22). According to our presentassumptions,the reaction learnedmight be any responseincludedin the animal's escape -defenserepertory. To produce fixation of a given response,the experimenterneedonly wait till the animal makes that response and then consistently omit the shock whenever it occurs
.
As a test of this prediction, we allowed one of our avoidance animals to escape shock by rearing; when the rearing to CS became absolutely regular, the animal was given inevitable shock. After variability of reaction had reasserted
117
Role of Shock in Reinforcement
Table 6.1 Sample Daily Protocols . Animal 40 Animal 7 Day 13 Day 20 Day - 3 Day - 25 Day ~3 Day J13Day - --of20 -- Day - -J25 -I Fl Fl,R Fl H I WB WF WF 2
FI , WB
R
3
WF , WB
WB , FI
4
WF
WB
5
I
WB , FI
FI
H
FI
H
WB
,
FI
PI FI
I
6
I
WB , FI , R
FI
H
7
I
WB , FI (L )
FI
FI
8
FI
WB , R
WB
9
WB
WB
FI
10
WB
WB
WB
11
WB , FI
WB , Wh
12
WB
WB , FI
FI
WB
WF H H LM H I WB WB I
,
FI
FI FI
,
,
FI
FI
FI
WF I FI
FI
FI
WF
I
I
Wh
H
FI
Ex
FI , Ex
H
H
H
LM
WF
FI
Ex
FI
FI
WB
I
WF
FI
I
FI
13
FI , WL
WB , FI
FI
FI
I
WF
14
WB
WB , Wh
FI
FI
I
FI
I WB
I
15
WB , Wh
16
WB , WF
WB
FI
FI
WB , FI
FI
FI
17
WB
WB , FI
FI
FI
H
18
WL
WB , FI
I
H
H
FI
FI
I H
,
FI
WF
,
Wh
WF
,
FI
WL I H ( L )
Ex H Ex I I I FI FI I
Wh
,
WB
Wh
,
WB
Wh
,
WB
,
WF
I
Wh
Wh
I
WF
WF
WF
20 WB ,Wh I : FI FI I I -H - -I .FI Key to abbreviations -flexes IH-head -inhibition FI ( L ) -flexes left leg movements only WB -walks backward Wh -wheel . C \ WF -walks forward Ex -extends leg WL , WR -sidesteps to left or right R -rears LM-independent leg movement (other than flexion or extension ) 19
WB , FI
WB , WF , FI
FI
FI
I
I
E. J. Gibson
118
16 - Avoids shock byrearing
14
I
.
---- Avoids shock byextension of A. forelimb
12
~~ ,
,
! ~,~J~.
i
,
'V
.
I
,
~I
8
,
, .
I
10
.
,
I
.
,
.
,
.
,
I
~ A
I
I
r
'
.
,
6
8
10
12
14
16
18
DAYS
20 OF
22
24
26
28
.
'
I"'
\ I , .~
I
30
\ ,
V
, , ,
4
,
"
~ "
,
2
r "
~ "
I
, I , , I ,
32
34
36
'.
38
TRAINING
Figure6.5 Frequency of shiftsfrom onereactionpatternto anotherfor oneanimalduringtwo phases of avoidancetraining.
A'v'O~3dSJ .:lIHS:lO ~38ViJnN
itself, the animal was again given avoidancetraining, but this time it could escape the shockonly by a sustainedlift of the right foreleg alone. Figure 6.5 portrays the number of shifts of reaction for eachday's training for this single animal. Through day 21, the shock could be avoided by rearing, which the animal beganperforming on its first day of training. Shifts of responseincreasedin the early stages(through day 8) becauserearing was part of a complex pattern involving lying down between trials. When the CS occurred, it rose to a stanceand then reared, placing the forefeet on the wall. As it began to avoid shock, it tended at first to eliminate rearing and to retain the rising from lying position only. After the eighth day, however, the animal gave up lying down, and rearing becamegradually stereotyped. On days 22, 23, and 24, it receivedinevitable shock, with an accompanying rise in the number of shifts, as we should expect from our original assumptions. After day 24, it was allowed to avoid the shock by sustainedlifting of the right foreleg and the number of shifts again diminished until a stereotyped reaction was -achieved ----- - - ---. The - lee was extended forward and waved, usually rhythmically, until the u - rearing light came on. Three days of inevitable shock cut out the reaction completely, and it did not occur again. But a number of other responsesdid, and presumablyit would have been possible to fixate anyone of them.
RandomShock One effect of 100 percent shockis to produce variability of reaction. If the number of shockswere to be decreasedfrom 100 per cent, but in a random fashion so that no one reaction consistently led to avoidance, would variability decrease ? In order to answerthis question, an experimentwas set up with 25 per cent shock, using two animals. This proportion was chosenin order to approximate, at least roughly, the number of shocksreceivedby
Role of Shock in Reinforcement
119
the avoidanceanimals, and it seemed,as an early guess, about right. A new random
distribution
of the five shocks in each group
of 20 trials was used
for every training day. In these animals the reduction of locomotion and the developmentof individual limb movement proceededvery slowly. The animals made a large number of IIconflictful" responses , jerking the head from side to side, circling alternately from right to left, or hitching backward and forward during the signal. But the most interesting result concerns variability. The reducedproportion of shockswas not accompanied by reduction in variability of reaction to the CS, as Figure 6.4 makesclear. Shifting from one reaction to another was very frequent. It is as if trial and error were at a maximum for an insoluble problem, which, indeed, it was. In a sense, the problem is even more baffling than inevitable 100 percent -
shock , for avoidance
occurs without
consistent
.&
relation
to action . Absence
of shock as such, consequently , cannot be the reason for increasing uni -
formity in avoidancetraining. Uniformity is createdby the situation. Consistent avoidance of the shock functions as positive reinforcement . The present
experiment
does not permit
determination
of the nature
of this
reinforcement, but it might be, for instance, either fear reduction or confirmation
of an expectancy .
Discussion
In general, the theoretical predictions made are clearly borne out by the results. But several questions arise concerning additional results which were
not predicted. The theory demandsthat actions following the CS be part of the animal's escape-defense repertory . Some of those observed are
clearly such: locomotion, which begins as running away; rearing and headlowering, which are associatedwith startle or fright in the animal's life history; wheeling, which is a circling type of withdrawal; crouching and butting, which are defensiveand offensivereactions, respectively. But what of the individual limb movements, the segmentalreactions, such as singleleg flexion or extension, or stepping in place? The significanceof these movements can be understood if it is recognized that they are actually
segmentsof the original locomotor response. To quote Liddell: "May it not be that the conditioned limb movements graphically recorded in our experiments are stepping movements , and that these precise reactions represent an experimentally curtailed , or " symbolic " manifestation of run -
ning? . . . The sheep, repeatedly flexing the foreleg in response to the conditioned stimulus, is really trying to run away from a situation about to becomedangerous" (12, pp. 57 ff.). This view is supportedby the late appearanceof single-limb movements in our results. Locomotion clearly dominated the repertory in the early
stagesof training. Schlosberg(17, 18, 8) and Wolf and Kellogg (22) also
120
E. J. Gibson
found that preciseflexion camelate, if at all. The real puzzle to be solved, then, is why locomotion is curtailed . When the animal is restrained in a stock or frame, as were the animals in all previous experiments of this sort ,
the problem is not so evident. But even in the present free situation locomotion was reduced. Liddell, in discussingdevelopment of neurotic manifestations, describedthis curtailment as "self-imposed restraint" (11). But what does this phrasemeanin terms of learning theory? Differentiation and individuation of responseare generally observed in studies of motor learning or development. Perhaps development of motor responsewill tpnd
to QO in the direction
of elimination
of gross
movements
even
when
'-"
the segmentalresponsesare not specifically adaptive- particularly when the gross movements are not themselvesadaptive. This is not generally true of the rodents, as has been pointed out; in their case, freezing or a tense quiescence seems to follow reduction of running or struggling . But in
animalswith , presumably, a somewhatmore differentiatedcortex, locomotion will break up into segmentalresponsesif locomotion is inadequate; if a given segmentalresponsedoes not answer, trial and error persists, yielding either a variety of segmentalresponsesor a combination of thesewith regression to locomotion .
Such, in general, was the picture when the animals received inevitable shock. Even as late as the twenty-fifth day of training, variability persisted in these animals; segmentalresponsesoccurredbut were not consistently repeated. Becauseuniformity did develop in animals permitted to avoid the shock, we concludethat the hypothesis of a dual role for unavoidable shock is supported. If it follows a motor response, this response tends eventually to be suppressedand supplantedby another. The result can be described as an interminable process of trial and error , by which the
problem is never solved. This conclusion is in general agreement with Estes' (4) contention that shockcausessuppressionof precedingresponses , although it does not accomplish complete extinction. In the course of the long trial and error with unavoidableshock observedhere, the animals often regressedbriefly to responsespreviously tried and dropped. The other role of shock- strengthening of a conditioned emotional reaction to the signal- we find consistent with the behavior of both avoidance
and
nonavoidance
animals , for , as we have
tried
to show , the
resulting action is appropriate to a state of internal " emergency ." There is, thus, support for Schlosberg (19) and, later, Mowrer (15) in assuming two
processesof modification. One dependssolely on paired stimulation with CS and US (the conditioned
fear ), whereas a second process follows
the law
of effect (fixation or suppression of the motor reaction to fear). The results tend to support Mowrer 's contention that motor reactions (at least other than very
diffuse
emotional
ones ) are modified
in accordance
with
the
secondprocess. It is impossibleto agreewith Culler that "the CR begins as
Role of Shock in Reinforcement
121
a copyofURandthengrows intosomething different. Inthefirststages, itmaybeindistinguishable fromUR;indeed withdecorticate preparations
it remainsindistinguishable throughout. Normally, however,CRdiffer-
entiates intoa specific preparation fortheoncoming Us,,(3,p. 142).It
appears,rather,thatthe greatestresemblance betweenthe CRandthe UR
comes inthelatestages oftraining whensingle-limb flexion develops. The early reactions to the signal take the form of rapid backing or running and are not copies of the UR. It willbeaskedby thesupporters ofGuthriewhetherthereactions to the
shock itself didnotshowequally widevariation, withthepossibility that
eachreactionto the signalmightvaryin accordance withthereactionto
thelastshock. Thedataasheresummarized donotofferproofto the contrary, butit wastheexperience ofalltheobservers intheexperiment thatnosuchcorrelation existed. Variation ofresponse to thesignal was
enormously greaterthanthatto shock, notonlyinnumber ofshiftsbutin varietyof patterns.In otherwords,anticipation of noxiousstimulation seemsto havemorewaysof displaying itselfthanreactionto thenoxious
stimulation itself.
Summary and Conclusions
I. Inlinewithrecenttwo-factor learning theories it issuggested that
conditioned responses developed withinevitable shockwillexhibit characteristics derivedfromtwo functions of the shock;the shock
reinforces a conditioned emotional statewhichinstigates a motor response ofa defensive character, andit alsosuppresses themotor
response,causingit eventuallyto be supplantedby anotheraction
belonging totheanimals natural escape-defense repertory. Kidswere trained ina situation permitting freelocomotion under threearrangements ofshock: (a)shock inevitable, (b)shock avoidable bya given consistent action, and (c) random shock. 2. At least ten differentreactionsto the CS were observed.The dominantreactionin earlytrainingwas locomotionbackward. All
forms oflocomotion became lessfrequent thansegmental movements,
suchasflexion, in laterstagesof training.
3. Inhibition ofresponse (standing still) occurred on24percent ofthe trialsin the animalsgiveninevitableshock.
4. Withanimals giveninevitable shock, therewerefrequent shifts fromonereaction to another, evenonthetwenty-fifth dayoftrain-
ing;but whenshockcouldbe avoided,therewas a definitetrend
toward uniformityof reaction.
5. Afteran animalhadachieved a uniform response whichavoided the shock,introduction of inevitable shockbroughtrenewedvan-
122
E. 1. Gibson
abilityof reaction. A different avoidance reactioncouldthenbe fixated.
6. Randomdeliveryof shockin 25percentof thetrialsdidnot reduce variabilityas comparedwith 100percentshock.
7. Inevitableshockgave a generalpictureof continuoustrialand error.Therewasno supportforthe Pavlovian viewthatshockactsto reinforcea withdrawalmovement,in the sense of increasingthe
probability ofrecurrence ofthesamemotorreaction; it had,instead, a tendency to suppress a preceding action,withtheresultthatanother took its place. Notes
1.Thisinvestigation wassupported(inpart)by a researchgrantfromtheNationalInstitute of Mental Health, U.S. Public Health Service.
2.Grateful appreciation isexpressed to Dr.Howard Liddell, Dr.A.UlricMoore,Mr.James Block,and Mrs.MiriamSalpeter, colleagues whosharedin the observations to be reported. Dr.Moorewasresponsible fordesign andconstruction ofapparatus.
3.It maybe notedthatJames(7)foundit impossible to condition legflexion in the opossum, although itcould beconditioned toattack ortoplaypossum; Liddell andhis
co-workers (12)foundit impossible to developa preciseflexionin the rabbit,which continued to reactwithstruggling movements evenafterprolonged training; Brogden (1)
notedthattheguinea pigsits tight orfreezes before delivery ofinevitable shock; inthis laboratory a youngramrecently exhibited, instead oflegflexion, pawing andbutting responses to the CS.
4. Fourof theanimals wereaccompanied in theexperimental roomby theirmothers. This feature was irrelevant for the present purpose.
References
1.Brogden, W.1.Theeffectoffrequency ofreinforcement uponthelevelofconditioning. 1. exp. Psycho!.,1939, 24, 419431.
2.Brogden, W.J.,Lipman, E.A.,andCuller, E.Theroleofincentive inconditioning and extinction. Amer.]. Psycho!., 1938, 51, 109117.
3. Culler,E.A. Recentadvancesin someconceptsof conditioning. Psychol.Rev.,1938,45, 134153.
4. Estes,W.K.Anexperimental studyofpunishment. Psycho!. Monogr., 1944,57,No.263.
5. Hilgard, E.R.,andMarquis, D. G. Conditioning andlearning. NewYork:AppletonCentury,
1940.
6. Hull,C. L. Princip!es of behavior.New York:Appleton Century,
1943.
7.James, W.T. Anexperimental studyof thedefense mechanism in theopossum, with emphasis onnatural behavior anditsrelation to modeoflife.J.genet. Psycho!., 1937, 51, 95100.
S.Kappauf, W.E.,andSchlosberg, H.Conditioned responses inthewhiterat:III.Conditioningasa function ofthelengthoftheperiodofdelay.]. genet. Psycho!., 1937,50,2745 9. Kellogg, W. N. Evidence for bothstimulus-substitution andoriginalanticipatory responsesin theconditioning of dogs.].exp.Psycho!., 1938,22,186192.
Role of Shock in Reinforcement
123
10 . Liddell , H. S.The nervous system asawhole : theconditioned reflex . In]. F. Fulton , Physiology ofthenervous system (2ndEd .). New York : Oxford Univ . Press , 1943 . Pp .491 -522 . 11 . Liddell , H. S.Conditioned reflex method and experimental neurosis .Ch . 12in]. McV . Hunt (Ed . ), Personality and the behavior disorders . New York : Ronald Press , 1944 . Pp .389 -412 . 12 . Liddell , H. 5., James , W.T., and Anderson , O. D. The comparative physiology ofthe conditioned motor reflex . Camp .Psycho /.Monogr ., 1934 , II,No.51. 13 .Maier ,N.R Schneirla ,T.C.Mechanisms inconditioning .Psycho /.Rev ., 1942 ,49 , 117 -134 ..F.,and 14 . Miller , N. E. Studies offear asanacquirable drive : I. Fear asmotivation and fear reduction as reinforcement in the learning of new responses .]. expo Psycho /., 1948 , 38 , 89-101 . 15 . Mowrer , O. H-solving . Onthe of.,learning :A reinterpretation of"conditioning " and "problem ."dual Harv .nature educ .Rev 1947 , 17 , 107 -148 . 16 . Mowrer theory and personality dynamics . New York : Ronald Press , 1950 . , o. H. Learning 17 . Schlosberg responses inthewhite rat.]. genet . Psychol ., 1934 , 45 , 303 -335 ., H. Conditioned 18 . Schlosberg , H. Conditioned responses inthewhite rat : II. Conditioned responses based upon shock totheforeleg .].genet .Psychol ., 1936 ,49 , 107 -138 . 19 .Schlosberg ,,H .The relationship success and thelaws ofconditioning . Psycho /. Rev ., 1937 44 ,379 -394 . between 20.Sheffleid ,F,.165 D.Avoidance and thecontiguity principle .].compo physiol .Psychol ., 1948 ,41 -177 . training 21.Skinner , B,.1938 F. The oforganisms : Anexperimental analysis . New York :Appleton Century . behavior 22.Wolf ,1.5.,and Kellogg ,W.N.Changes ingeneral behavior during flexion conditioning and their importance forthelearning process .Amer .].Psycho /., 1940 ,53 , 384 -396 .
TheEffectofProlonged Exposure to Visually PresentedPatternson Learning to Discriminate Them
EleanorJ. Gibson,RichardD. Walk Introduction to Chapters7 and 8
Rearing experiments witheither impoverished orenriched environments were very fashionable inthefifties. Interest began withhistories ofindividuals deprived of normal environmental rearing conditions, themostextreme cases being persons bornblind,whohadlaterhadsightrestored bycataract removal (vonSenden
1932,1960), andpersons thought tobereared inthewild,likethewildboyof Aveyron (Itard 1894, 1962). These cases hadalways stirred theimagination of
philosophers, ofcourse. ButDonald Hebbsbook(Hebb1949), as wellas the
dark-rearing experiments ofLashley, (Lashley andRussell 1934), Hebb(1937)
and Riesen(1947),broughtquestions of whether and howthe environment
exercised a roleindevelopment totheforefront oflaboratory investigation. Hebb thought thatschemasweredeveloped fromquitespecific experiences thataccounted forperceptual organization, withgroups ofcortical neurons exciting and reexciting oneanother. Hebbhadbeeninfluenced bySendens andRiesens pub-
lications, remarking that BothSendens dataandRiesens saidthatthereis no
pattern perception without experience (Hebb 1980,p. 295).(Thisdespite the contraryresultsofhisownexperiments.)
Hebbs book andcontemporary research ontheneural development andorganization ofthevisual system (Hubel andWiesel 1959) inspired a spateofrearing experiments, involving deprivation oflight,orpatterned light(seeE.Gibson 1969,
ch.12,forasummary). Hebbs ownearlier experiments were designed tostudy the
effect oflightdeprivation onvisualdiscrimination inadultrats.Butlateronhe
designed an earlyenvironment program in which younger members ofhis department at McGill participated (Hymovitch 1952,Thompson andHeron 1954, Meizack andScott1957). Thisprogram emphasized positive contributions oftheenvironment. Hebbsaidin hisautobiography, Thisprogram wasless dramatic thansensory-deprivation workbutperhaps moreimportant. It wasa Journal ofComparative andPhysiological Psychology, 1956,49,239242.
126
E. J. Gibson&: R. D. Walk
major inj1uence in persuadingpsychologists that the IQ is not built-in at birth, andso a factor in suchthingsas theHeadStart program" (Hebb 1980, 300). This work, interestingand valuableas it was, seemedto have rather little bearingon Hebb's theoryof visualcorticaldevelopment , with its emphasis on cell assemblies createdin the beginningby eyemovements following linesand angles in quite specificpresentations . As a theoryof perceptuallearning, it was of real concernto me, and I proposedto Richard Walk that we rear infant rats in a controlledrat environmentwith the additionof trianglesand circleson the cage walls. Though thesewere hardly typical of a rat's normal environment , the opportunityto view them daily throughoutearly development might result in appropriatecell assemblies developing , and thus facilitate later learningof a discriminationcuedby them. The circlesand triangleswere cut out of metal, paintedblack, and hungon the wire meshof the cages . When the animalswere testedlater in a discriminationbox with the sameblackfigurespaintedon the doors, they learnedsignificantlyfaster than their litter-matesrearedwithout similar exposure . We might haveleft it there, with a triumphant "Well, well!" But we didn't. We pushedon to further experiments (ninein all), varyingsuchfactorsas timeof exposure(was therea "critical period" early on?), variationsin the patternsat testing, the effectsof differentialreinforcement , and, finally, a comparisonwith dark-rearedrats ( Walk, Gibson, Pick, and Tighe1958; Gibson, Walk, Pick, and Tighe1958; Gibson, Walk, and Tighe1959). Resultsin thesereplications , alas, wereseldomas clear-cut as they had beenin theoriginalexperiment . In fact, rats deprivedof all light during theirfirst ninetydaysof rearinglearnedthe trianglecirclediscriminationas readilyas rats rearedwith an opportunityto view these patternsfor theequivalenttime. We had to conclude that we hadnofirm evidence that "perceptual discriminationof a form is an achievement resultingfrom an integrationprocess of theneuralelements involved" (Gibson, Walk, and Tighe1959). What might accountfor thepositiveresultsof thefirst experiment , and oneor two later ones? Examinationof all the experimentsrevealedone factor that appeared to favor a facilitationof laterdiscrimination learning - that waswhether thepatternshungon the cagewall werecut out, to providedepthat an edge , or werepaintedon a surface , with no edgedepth, like pictures . Our last experiment , consequently , was a test of cutoutsvs. painted patterns . The resultsof this experiment werepositive, in thesensethat thecutoutsprovidedthe only evidence for facilitationby prior experience with thepatterns . But theeffectwas very weak. Locationof an objectas somethingseparateand unitary in the layout is without doubt importantin organizingperceptionandgiving it meaning . But the notion that perceived form of an objectdepends on constructing a schema from repeated exposure to elementary pieceslike linesandanglesthatfall upontheretinaandare assembled in a phasesequence was unsupported . From my presentapproach that perception doesnot start with somethingpictoriallike a retinalimage, but is rathera searchfor informationaboutthingsin theworld- that is not surprising.
ProlongedExposureto VisualPatterns
127
Wegottheresults weshouldhave,andtheymakegoodsense, I think.Passive
exposure totwo-dimensional displays doesnotresultinperceptual learning, even when reinforced, anddeprivation ofsuchexposure hasnoilleffects at allonlater pattern discrimination learning. Exposure ofsimilar patterns isquite aseffective as exposure ofthesameones,whenanykindofeffectoccurs, as it didwithcutouts.
Perceiving anobject segregated fromitssurrounding layoutmayhaveattentional valuethatcarries overtoanother situation, buttheobject neednotbetheidentical one.Weneeded a theoryof perceptual learning, but not theonethat ledus to performtheserearingexperiments.
Recentliteratureon the developmentof discrimination has shownan
increasing trendtowardacceptance ofempiricistic explanations (2,9).That
ability todiscriminate visually presented patterns develops withtheexpe-
rienceandenvironmental reinforcement ofthegrowing animal maybethe case,buttheevidence forthisviewis stillinconclusive. Earlystudies by
LashleyandRussell(11)and by Hebb(8)on the rat favoreda nativistic
interpretation ofthedifferentiation ofvisual qualities, butlatercomparable studieswiththechimpanzee andpigeon(13,14)apparently favored an
empiricistic explanation. Recentexperiments by studentsofHebb(5,6, 10)
haveemployed anenrichment technique, withresults which appear to favor alearning hypothesis. These studies attempted toprovide agenerally rich environment andusedascriteria testsofa rathergeneral type.If
opportunity to viewa variedandpatterned environment is important in thedifferentiation ofvisual qualities, wedonotknowhowgeneral orhow
specificthe relevantexperience mustbe.
Theexperiment tobereported proposed to investigate thedependence
ofvisualformdiscrimination in adultratsona specific variation in visual stimulation duringgrowth. Tothisend,anexperimental groupofanimals
was raisedfrom birth in cageswhichexhibitedon the wallscirclesand trianglesidenticalin formwithoneslaterto be discriminated. Thecontrol
groupwasraisedunderthesamestandard conditions butwithout opportunityto seetheseformsbeforethediscrimination learning began.If the
opportunity to viewspecific formsfavorsdevelopment of the abilityto
differentiate themin a laterdiscrimination learning problem, theexperimental animals shouldlearnfasterandshowa higherproportion of Ss reaching thecriterion thanthecontrolgroup. Method
Rearing
TheSswerealbinoratsrearedfrombirthin identical i-in.wire-mesh cages
measuring 15by 13by 9 in.Thecageswereplacednextto eachotherina small,
128
E. J. Gibson& R. D. Walk
softly
lighted
three
empty
sides
fourth
side
the
, 4
room
wall
metal
of
the
total
the
of
four
, litter first
day
.
The
each
young so
begun
C2
7
white
blank
ft . from
the
the
cage
cardboard
wall
of
mesh
mates
walls
the
room
was
, a
in
two
. on
a
on
on
the
water
were
circles
side
.
the
ceiling
bottle
of
on
one
left
on
the
, two
experimental
four
circles
were
patterns to
were
fastened
The
. These
relationship
used
( experimental
litters
, and
litters
food
and
sides
of
were at
animals
. E2
1 ),
10
)
when
black 3
were
changed
water
. All
the
cages
in
.
in in
during
of
(n
=
(n
pups
age
,
in
separate 90
control
8 )
and
C2
the
approximately
two
experi
. These
( experimental
apart
and
of
were
E2
E1
days
=
weeks
females were
five (n
split
four
and
litter
2 ) . Litters
born
split
and
,
, litter
were
, not
males
the
animals
and
random
( control
weaned
that
when
was
litters
were
divided
top only
a
experimental
3t a
: E1
These
other
the
patterns
follows
between of
by
and
-
.
1 ) , and
born
of
assure
groups
as
interval
the were
were
to
animals
numbered
( control
surrounded mesh
triangles
stimulus
group
A
. At
cages
triangles
experiment
be
was wire
cage
equilateral
occasionally
mental
mesh the
the
two
and
position
the
the
.
,
diameter
cage
from
within
walls
forms
the
ft . from
food
the
. Each
inches
. Visible
, and On
room
, several
and
,
=
at
eight The
2 ),
=
of
2 )
C1
were
the
born
long
within
two
a
days
weeks
.
old
(n
or
will
litter
because
one
cages
days
C1
9 ) were
were
and
,
old
sexes
.
were
experiment
was
.
Apparatus The
apparatus
was
discrimination as
described
side
at
in the
stimulus of
holders
. The
glass
. A
The
stimulus
were for
25
- in
, and four
and
. - wide slid
end into
. bulb
on
the
of
were the 3
25 first
painted
background a
side food
stimulus
holders box
flat
a
by , one
*
, and
flat
white
, and
circle
was
pushing
open the
4i
in
circle
. by
4i
of
section
a
was
. in
door and
one
and
diameter
- in
. square
in
the with
metal
the
side
stimulus by
illumination
circle
Ii
.
by
covered
only
black in
the
,
side - in
and
front
the
- shaped
were
- in
. in
had the
V
constructed
partition
2i
holders
Two
floor
patterns
each furnished
with
( 7 ).
false
. center
floor
stimulus
a
. The - in
the
. The . The
Ii
grooves black
. above
Grice
stimulus
chambers
into a
in
two
the
fitted
in . on
discrimination
and
choice
painted
obtained
by
together
between
white
animal
separate
both
doors was
described
( 1 ) . The
grooves
mounted
holders
one
joined
Lawrence
. Masonite
triangle
each
of were
apparatus -w
painted
equilateral them
10
apparatus
were
modification
Baker
holders
the
a
compartments
.
triangle and
the
doors
center the
in
. There triangle
,
.
TrainingProcedure Pretraining . Animalswereplacedon a 24-hr. feedingcyclefor approximately one weekprior to the start of experimentation . They weregiven threeto four days' trainingin obtaininga smallquantityof wet mashfrom the stimulusholdersby pushingopenthe door in the center.The stimulusholderswerepaintedflat black for this pretraining . Thedooron only onesideof thediscrimination box wasraised at a time. As soonasthe animalobtainedthe food from the food cup, E lowered the door in front of the stimulusholder. The doorbetweenthe two discrimination boxeswasthenopenedfor the next trial, andthe animalsecuredfood by pushing
ProlongedExposureto VisualPatterns
129
itsnoseagainsttheblackstimulus holderat theopposite end.Theanimal ateten timesfromthecupintheholderinthefollowing order:RLLRRLLRRL.
Discrimination training.Duringdiscrimination training bothMasonite doorson
thechoice sideoftheapparatus were raised, exposing thetwostimulus patterns sidebyside. Bothstimulus holders werebaited. Assoonastheanimal pushed
against onestimulus door,theMasonite doorinfrontoftheopposite stimulus was closed. Ifthechoice wascorrect, theanimal wasallowed toeatthewetmashinthe
foodcup. After 60sec., thedoorbetween thetwocompartments wasopened and
theanimal proceeded to theopposite end,wherethenextchoice wasmade.Ifit
wasincorrect, a modified correction procedure wasfollowed. Bothdoorsin front
ofthestimulus holders wereclosed. After60sec.theanimal wasallowed tomake
a choice intheopposite discrimination box.Animals wereallowed upto three
errorspertrial.Following thethirderrorthedoorin frontof thecorrectstimulus
figure remained open, andtheanimal wasallowed toeatfrom it.Thisprocedure
meant thattheanimal ateequally oftenoneachsideoftheapparatus. Tentrials weregiveneachdaywitha maximum ofthreeerrorspertrial. Thepositive stimulus waspresented in thefollowing order:RLRRLLRLLR; LRLLRRLRRL; RRLLRRLRLL; LLRRLLRLRR. Theorderwasrepeated everyfour days.Forhalftheanimals ineachgroupthecircle wasthepositive stimulus, and
for halfthe triangle. Animals wererun untiltheyattaineda criterion of 18
outof20correct responses, withthelasttenconsecutive responses correct (one daysrun),oruntiltheywererunintheexperiment for15days(150trials). After theexperimental session animals wereallowed to eatfoodpellets for1 hr.The hunger drivewasa function ofapproximately 22 hr.deprivation. Each ofthetwo Esranone-halfof theexperimental andone-half of thecontrolanimals.
Results
Thenumber ofdaysofdiscrimination training andtheerrors(initial and repetitive) arepresented in Table7.1forbothgroupsof animals. In the
table areindicated thesexandlitterofeachanimal. Thesecond litters (LE 2 andLC ), 2 it willberemembered, weresplitatbirthandthusprovide a somewhat bettercontrolled population. It is obviousfromthe tablethat thereisa difference between experimental andcontrol groups. Outofthe control group,only1 animal reached thecriterion during15daysof
training. But15of the 18experimental groupanimals did.Bythechi-
squaretest, this difference is significant at betterthan the .001levelof
confidence. Ifwecalculate thechisquare foranimals ofthesplit-litter
groupsonly,usingFishers exacttest(4),thesignificance ofthedifference
isbetween .002and.001.Theerrors, bothinitial andrepetitive, reflect the
same trend.
Afurther check ondifferences inthepopulation studied ispossible by areexactly 50in100thatthereisanydifference between sexgroups.
testingmalesagainstfemales. Whenthiscomparison ismade,thechances
130 Table
E. J. Gibson& R. D. Walk 7 .1
Number J. " - U~' V ' -~ '-' of Experimental &
Days - - J ~ Trained - . - -- -- - and - - - Errors -
for
Group No
.lAnimal ~ ' ~~~' W"
the
Experimental ..
and
Control
Control
Group -
.
No
Days
Initial
Repetitive
Run &, ~ .
Errors - -- - - -
Errors
Animal
12 .
39
44
LCl
Groups
2 J
.
Days
Initial
Repetitive
Run
Errors
Errors
14 .
25
11
LEl
30
d
LEl
31
J
7.
22
34
LCl
~
15
68
52
LEi
32
J
11 .
28
19
LC2
4 ~
15
50
44
LEl
33
J
5.
11
9
LC2
6 ~
15
59
30
LE 1 35
~
15
32
21
LC2
7 ~
15
67
18
LEl
37
J
14 .
40
41
LC2
12
~
15
60
28
LEl
40
J
7.
24
29
LC2
15
~
15
80
80
LEl
41
d
8 -
24
24
LC2
11
0
15
66
72
LEz
44
~
7.
23
28
LC2
5 J
15
68
80
LEz
47
~
10 .
25
16
LC2
13
J
15
74
117
LEz
62
~
15
70
73
LC2
14
J
15
84
92
LEz
63
~
9.
23
16
LE2
64
~
12 .
39
50
LE2
43
J
15
57
51
LE2
45
J
13 .
28
12
LEz
46
J
10 .
25
16
LEz
60
J
9.
30
23
10 .
37
34
20
LEl 616 Mean 10 50 .00 14 .91 63 .73 J.Y.l"-UJ .' ~_.._ - 32 ---.0 --6 30 -.73 . . 56 . I" ~Indicates that animal reached criterion ;the criterion day 'strials are included innumber of days run . Figure 7.1 showsthe learningcurve for experimentaland control groups. Percentageof correct responsesis plotted against days of training. The animalsthat reachedthe criterion are included in the percentageson the assumption that they would continue at their final level of performance. The curves show that the groups begin to diverge by the third or fourth day of training and diverge increasingly thereafter until the tenth day, when a majority of the experimentalgroup had learned. Discussion The results presentedshow conclusively a differencein easeof learning a circle-triangle discrimination between the group reared with these forms exhibited on the cagewalls and the control group. Sincethe control group had the same conditions of training (and pretraining), the same living
Prolonged Exposureto Visual Patterns
131
100
EXPERIMENTAL
GROUP
9
8
7 "
p. . ,
,, "
6
, '
"
A
,
" ,
" " 0
' 0"
,
' d
p' , , ~. o- - - . o- - - - O- " "
"
"
'
, d"
"
...0- - - - 0"
50
,
"
"
p. ,
CONTROL
GROUP
;,; \
,
' t:(
0 1
2
5
6
7
8
9
10
11
12
13
14
15
DAY
Figure 7.1 Learning curves ofcorrect responses per day , forthe experimental and control groups ., inpercentage
S3SNOdS3 ~ l.J3~~OJ::JO39\tl.N3J~3d
conditions , and, in our secondlitters, the sameheredity, thedifference must be attributedto someadvantagearisingfrom the opportunityto look at theforms. Thisadvantage couldbesomethingspecificwhichhappensearly in the animals ' development , analogouswith "imprinting" (12) or with Hebb's postulateddevelopment of reverberating neuralcircuits(9). On the other hand, a learningtheoristwho favors "hypotheses " as a factor in learninga discriminationmight suggestthat seeingthe formson the cage walls favors formationof the correcthypothesis . Sincethe forms were left on the wallsduringthe learningperiod, it is not possibleto conclude that earlyexperience in viewingtheformsis the basisof the effect. Suitable controlsareat presentbeingrun to clarifythis point. Sinceresearch in discrimination learninghascenteredroundthe continuity hypothesisin recentyears,it might beaskedwhetherthepresentresults tend to confirmor deny this hypothesis . The animalsin the experimental groupprofited, in the discrimination task, from an opportunityto view the two formswithout any differentialreinforcement of them. Nondifferential reinforcement in viewing thesecouldhaveoccurred , sincethe animalsate anddrankin their presence . Spence 's 1936article(15) suggeststhat some degreeof positiveexcitatorypotential,irrespective of differentialreinforce ment, would be consistentwith fasterlearningwhendifferentialreinforce mentis introduced . On the otherhand, the valuesselectedfor his analysis
132 are
E. J. Gibson& R. D. Walk
purely
arbitrary
reinforcement Bitterman the
and
course
of
is
a
by
an
with
the
on or
of
test
the
,
.
occurs of
in
differential
further
finding
that
reinforcement
the
. transfer
The from
complications
intro
-
.
the
facilitation
hypothesis
positive
without
described for
the
lack
their
the
reinforcement
problem
despite
nondifferential
stimuli
time
resulting
, by
clearly
of
critical
the
stimuli
of
nondifferential of
differentiation
beclouded
demonstrate
application
effective
statement
perceptual
test is
effect
to
that his
that
retarding
research
specificity
concluded
viewing
optimal
concluded refutes
conclusion
seem
specific
Further
be or
experience this
general
in
duced
cannot
(3 )
sheer
results
.J experience
it
confirms
. But
Dresent
is
so
Elam
reinforcement
there
, either
will visual
of
investigate
whether
experience
,
discrimination
and
there
the
learning
relative
.
Summary
This
experiment
exposure
sought
to
adult
animal
raised
from
Animals
certain learns birth
of
cages
.
and
made
- illuminated
control
errors
with reinforcement
the
them
,
forms
to ,
facilitated
ease
groups by
mounted
control
circle
group
. It
discriminated the
discrimination
was
,
even
an were
cardboard walls
.
of
their
duration
of
days
- triangle
criterion
which
animals
the the
90
the
continued
with
white
on
approximately a
and
of
throughout
reached
be
early
the
Two
had
learned
group
of on
surrounded
birth
were
the
.
also from
groups
than
,
cages
animals
experimental
effect
visually
group triangles
and
fewer
differential
well
the
the
experience
discriminate
and
the
presented
to
When
experimental of
determine
,
experimental
circles
experiment ~
Animals
in
the
black
to
forms
old
,
the both
discrimination significantly
concluded
in learning
that
the
. faster
absence
visual
of
.
References 1. Baker , R. A., & Lawrence , D. H. The differentialeffectsof simultaneous and successive stimulipresentation on transposition . ]. compo physiol . Psychol ., 1951, 44, 378- 382. 2. Beach , F. A., & Jaynes , J. Effectsof earlyexperience uponthebehaviorof animals . Psycho /. Bull., 1954, 5I , 239- 263. 3. Bitterman , M. E., &: Elam, C. B. Discriminationfollowing varying amountsof nondifferentialreinforcement . Amer.J. Psycho I., 1954, 67, 133- 137. 4. Fisher , R. A. Statistical methods for research workers . (11thEd.) New York: Hafner, 1950. 5. Forgays , D. G., & Forgays , JanetW. The nature of the effect of free-environment experience in the rat. J. compo physiol . Psychol ., 1952, 45, 322- 328. 6. Forgus , R. H. The effectof early perceptualleaming on the behavioralorganizationof adultrats. J. compo physiol . Psychol ., 1954, 47, 331- 336. 7. Grice, G. R. Theacquisitionof a visualdiscrimination habitfollowingresponse to a single stimulus . J. expo Psychol ., 1948, 38, 633- 642. 8. Hebb, D. O. The innateorganizationof visualactivity. I. Perceptionof figuresby rats rearedin total darkness . J. genet . Psychol ., 1937, 51, 101- 126.
Prolonged Exposureto Visual Patterns
133
9.Hebb ,D.O.The organization o/behavior .New York :Wiley ,1949 . 10 .Hymovitch , B . The effects of experimental variations on problem solving inthe rat .]. compo physio /.Psycho /., 1952 ,45 ,313 -321 . 11 .Lashley ,K.S.,.& Russell ,J.T./.The mechanism ofvision test ofinnate organization ].genet .Psycho , 1934 ,45 , 136 -144 . .XI.Apreliminary 12 .Lorenz ,K.Der Kumpan inder Umwelt des Vogels .].Om .Lpz .,1935 ,83 , 137 -213 . 13 .Riesen , A . H . The development of visual perception in man and chimpanzees . Science , 1947 ,106 -108 . . 107 14 .Siegel , A..].I.compo Deprivation visual form definition in.the ring dove .I. Discriminatory learning physio /.of Psycho /., 1953 ,46 , 115 -119 15 .Spence ,K..W.The nature ofdiscrimination learning inanimals .Psycho /.Rev ., 1936 ,43 , 427 -449
8 The Effectiveness ofProlonged Exposure to Cutouts vs.Painted Patterns forFacilitation of Oiscrirnina tion Richard D. Walk , Eleanor ].Gibson , Herbert L.Pick ,Ir., Thomas I. Tighe Experiments by the authors (Gibson &Walk ,1956 ;Gibson ,Walk ,Pick ,& Tighe , 1958 ; Gibson , Walk , & Tighe , 1959 ; Walk , Gibson , Pick , & Tighe , 1958 ) have shown that prolonged exposure to visually presented patterns may ,under certain conditions ,facilitate discrimination of these patterns in a learning situation ! An auxiliary finding was that in experiments where the patterns exposed inthe cage were cutouts ,facilitation always occurred , though in other experiments where the same patterns were painted on a plain rectangular background ,there was no facilitation .et This effect has been previously commented on ( Gibson et al . , 1959 ; Walk al . , 1958 ) , but no single experiment has directly compared the effectiveness of cutouts with painted patterns .The present experiment was designed to do so . follows The experiment included four groups of albino rats , treated as : Group Iwas reared with cutout patterns (triangle and circle )on the walls of the living cages ; Group II was reared with the same patterns painted on metal rectangles on the cage walls ; Group III was reared with plain color rectangles ,like Group II,but without the painted pattern ;Group IV was reared with nothing mounted on the mesh walls of the living cages . From the results ofthe previous experiments following relationships among the four groups should hold . ,the 1.The cutout -pattern group (I)of should be superior tothe control group ( I V ) . This is a replication our first experiment ( Gibson & Walk ,1956 ). 2.The painted -rectangle pattern group (II()Ishould not be significantly different from the plain group II ) ( Gibson et al . , 1959 ; Walk etal ., 1958 ). 3. The cutout-pattern group (I) should be superior to the paintedpattern group (II) and the plain-rectangle group (III). This difference Journal of Comparative andPhysiological Psychology , 1959 , 52, 519- 521.
136
R. D. Walk, E. J. Gibson, H. L. Pick, Jr., & T. J. Tighe has not been tested directly, but was hypothesized in two papers (Gibson et al., 1959; Walk et al., 1958). 4. The position of the painted-pattern group (II) and the plainrectangle group (III) in relation to the control group (IV) cannot be predicted from previous research. If only the identical pattern, as a cutout, will produce facilitation, Groups II and III will be indistinguishablefrom Group IV . If the rectanglesfunction as cutout figures and are sufficiently similar to the test figures to yield transfer (Gibson et al., 1958), then Groups II and III should learn the discrimination faster than Group IV .
Method Subjectsand RearingProcedure The5s, 84 albinorats, wererearedin identicalt -in. wire-meshcagespaintedwhite andmeasuring16 by 13by 9 in. Thecageswereplacedin the centerof compartments(27 by 23 by 22 in.) whosefloorsand walls werecoveredwith masonite paintedflat white. Illuminationwas suppliedby fluorescentlight 7 ft. abovethe cages . Twelvepregnantratsfrom RocklandCountyFarmlitteredwithin a spanof ten days. Litterswere split amongthe four groupsso that eachgroup includedoffspringof all 12 mothers . The rectanglesor cutoutswere mountedon the cage wallsof theappropriategroupsbeforethe pups' eyesopenedandremainedon the wallsthroughoutthe experiment . The patternsusedfor GroupI weretwo sheetmetalcircles(3 in. in diameter ) andtwo sheet-metaltriangles(31 in. on eachside). paintedblack. One form wasmountedon eachwall. For GroupII, thesepatterns were paintedon white rectangles , 41 in. by 5 in. Group Ill's cagewalls held four 41-in. by 5-in. rectangles paintedwhite. Discrimination Training Trainingbeganwhenthe animalswere90 daysold. They wereplacedon a 24-hr. feedingscheduleand then habituatedto the apparatusandpretrained . The modified Grice-type discrimination box alreadydescribed(Gibson& Walk, 1956) was used.It containedtwo V-shapedcompartments joinedtogetherso that the animal alternatedfrom oneto the otherwithout handling.Stimulusholderspaintedblack wereusedfor pretraining . Animalslearnedto takefood from holderswith doors openedwide, andthedoorsweregraduallyclosed.Pretrainingcontinueduntil the animalpushedopencloseddoors, alternatingfrom onecompartment to the otherI at leasteight timesin 10 min. Pretrainingrequireda meanof 6.8 days. For discriminationtrainingthe blackstimulusholderswere replacedby white holderson which was paintedeither a black triangle(3 in. on eachside) or a blackcircle(2i in. in diameter ). Whenthe door separatingthe two compartments was raised , an animalfacedthe two stimulusholderssideby side. If the animal chosecorrectlyby pushingthe door of the positivestimulus , it waspermittedto
Pre-exposureand Discrimination
137
eatanda blackdoorwaslowered infrontoftheotherholder, butif theanimal
moved thedoor ofthenegative stimulus (which waslocked within.ofplay), the intheothercompartment. Three errors (oneinitial andtworepetitive) were
doorswerelowered infrontofbothstimuli andtheanimal madea second choice
allowed oneachtrial. When ananimal made a second repetitive error, thedoor
waslowered infrontofthenegative stimulus, andit wasallowed to eatfromthe positive one.Theright-left position ofthepositive stimulus wasrandomized. Ten trialsweregivena day,eachtrialfollowed byreinforcement, untilS attained 18
correct choices in20trials withthelast10correct. Oneanimal inGroup IVthat
hadnotlearned in300trials wasstopped. Tocontrol extraneous cues, eight
stimulus holders wereused.FourofthesewereusedforTrials 1to5,theotherfour onTrials 6 to 10.HalftheSshadthecircle asa positive stimulus, halfhadthe triangle. Each ofthefourEsrananimals fromeachgroup. Results
Themean trialsandmean initial errors foreachofthefourgroups are shown inTable 81.Thetestforreplication is thatbetween Group I (cutouts) andGroup IV(control). Thedifference isinthepredicted direction,buttheVsof1.42fortrials and1.54forerrors areonlysignificant at
the .10level(one-tailed). Actually, themeandifference of 26 trialsand
of12initial errors(15% facilitation fortrialsand20%forerrors) wasas largeasthatfoundinotherexperiments yielding significant differences (Gibson etal.,1958), butthevariance islarger. Thepredicted no-difference
between Group II(painted patterns) andGroup III(white rectangles) was upheld withVsof.39fortrials and.67forerrors (p> .50). Thecomparison
between thecutout group(I)andtherectangle groups (IIandIIIcombined) wasaspredicted withVsof 1.79fortrialsand1.73forerrors(p< .05,
one-tailed). Thequestion whether thestimuli forGroups IiandIIImight of.10fortrials and.40forerrors (p>.50)thatcompared these groups withthecontrols werenotsignificant. Thisexperiment, then,seems touphold thehypothesis from previous yieldsome facilitation canapparently beanswered negatively since theVs
research by theauthorsthatonlythecutoutsyieldfacilitation in the Table 8.1
Mean Trials toCriterion andMean Initial Errors fortheFourGroups Trials
Group
N
I (cutouts)
22
II (paintedpatterns)
21
III(white rectangles) 18 IV(control) 23
InitialErrors
Mean 144.7 175.2
168.4 170.5
SD
Mean
SD
62.9
49.0
24.2
54.9
49.2 56.4
57.0
61.4 61.5
17.6
21.4 28.2
138
R. D. Walk,E.J. Gibson,H. L.Pick,Jr.,& T. J. Tighe
discrimination learning task,butthestatistical levelof thedifferences betweenthecutoutgroupandtheothergroupsisnotverysatisfactory. Discussion
Including theexperiment reported here,wehaveperformed a series ofnine
separate experiments inwhicha totalof 422rodentsweretaughta triangle-circle discrimination, theexperimental variable beingalways sometype of exposure to visually presented patterns. In allexperiments wherecutouts wereusedas exposurestimuli,facilitation was observed(Gibson&
Walk,1956;Gibsonet al.,1958,Experiments I andII;Gibsonet al.,1959,
Experiment I;Walket al.,1958,Experiment I;thisexperiment), butthe
amountof facilitation variedgreatlyfromone experiment to the other.
Painted patterns asexposure stimuli neveryielded facilitation (Gibson et aL,
1959,Experiments IIandIII;Walketal.,1958,Experiment II;thisexperiment). Thereasonsuggested forthefacilitation obtained by thecutoutsis that the cutoutsare attention-getting.Thesesolidpatternoutlinesstandout in contrastto the background as objectsin depth.Whenlaterconfronted
bytwopainted patterns inthediscrimination box,theanimal ispresumably moreaptto differentiate thepatterns fromthetotalsurroundings andto connectthem as cues with the differentialreinforcement.This hypothesis
maybe relatedto the superiority of stereometricoverplanometric stimuliin discrimination taskswithmonkeysand children(Harlow& Warren, 1952;Stevenson& McBee,1958;Weinstein,1941). Summary
An experiment wasdesignedto test the effectiveness of metalcutouts compared withpaintedpatterns as exposure stimuli duringrearing. Four
groupswereused.GroupI hadblackmetaltriangles andcircles mounted onthecagewallsduringrearing. GroupIIhadblackpainted triangles and circleson whiterectangles. GroupIIIhad whiterectanglesin the living
cages.GroupIVhadnothingmounted onthecagewalls.OnlyGroupI
learneda triangle-circle discrimination fasterthantheothergroups,which wereindistinguishable fromeachother.Whilethe experiment tentatively confirmed hypotheses derivedfromprevious experimentation aboutthe
effectivenessof the cutoutsas exposurestimulito facilitatediscrimination
learning, thefirmness of thisconclusion is attenuated by the levelof significanceobtained in this experiment. Notes
1.Thisexperiment wassupported, inpart, byagrath from theNational Science Foundation
Pre-exposureand Discrimination
139
Behaviorof Light:-and Dark-RearedRat:s on a Visual Cliff
Richard D.Walk, Eleanor J. Gibson, Thomas J. Tighe Itwastruly serendipitous thatthetrouble required toreara large group ofratsin
thedark inspired ustoobserve theratsinasecond kind oftest, asortofartificial cliff. Theidea wasthattheratswould walk outonit(glass over a void) ifthey hadfailed todevelop depth discrimination during dark-rearing. T.J.Tighe, atthat
timeourresearch assistant, andI hastily puttogether a simulated cliffwithbits ofwallpaper, glass, androdsandclamps thathappened tobearound thelaboratory. Then we(Walk, Gibson, andTighe, ina party) putthelight-reared rats ona slightly raised center board, chosen asa starting place, andwatched them as,
onebyone, they gotdown ontheshallow side andcrept around, butuniformly avoided thedeepside.Then wetookthedark-reared animals outoftheir seclusion andtried them, onebyone.Their behavior wasindistinguishable from theirlight-reared peers.
Aswewatched them, webegan tobeconcerned thatit might bejustthatside
ofthecontraption thatthey didntlikedrafts orodors, who knows. Wequickly fetched more ofthewallpaper wehadused ontheshallow side, andputitdirectly under theglassofthedeep side.Every ratwasnowgiven a second chance, and theyallcrossed back andforth from sidetoside, darkandlight-reared alike. As wereturned thelastrattoitscage, Tighe said,I wouldnt have believed itifI
hadntseenit. It wasa memorable day.
In themanyexperiments withthevisualcliffthatfollowed thisone,our
first,weconcentrated oncomparative studies withprecocial andlessprecocial animals (Gibson andWalk 1960; Walk andGibson 1961). Itwasseveral years later when weperformed another dark-rearing study, withkittens assubjects. The results wereaninteresting contrast tothestudywithrats.TA/hen thekittens were
brought outofthedarkandplaced onthecliff, they wandered about everywhere,
notfavoring onesideortheother. Butit could notbeargued fromtheirbehavior
thatexperiences with dropoffs wasrequired foravoidance ofthem toappear. They
' AAAS.Science, 1957,126,8081.
142
R. D. Walk,E.J. Gibson,& T. J. Tighe
were putonthecliffdailyfollowing theirintroduction toa lighted environment
After twodaysinthelight, eighty percent ofthekittens avoided thedeep sideand aftersixdaysalldid.Experience should havetaught themthatit wasassafeas the other,if it taughtthemanything.
Darkrearing hasproblems asa method, buttaking theresults withthose of other experiments oneisstruck withthespecies differences, firstofall,inreadiness toactat birthand,equally, withtheamazing degree ofpreparedness toengage in
perceptually guided behavior when anaction system, suchaslocomotion, has matured toreadiness ina normal environment. A light-reared kittenavoids a cliff
when it isready towalk, anddetects theaffordance ofa supporting surface when it is visually specified. Normally maturing visionis essential for theproper outcome, but no learningofspecific S-Rbondsis involved .
Fromthe isth centuryto the present,the empiricistand the nativist theoriesof depthperception havebeenvigorously debated.Oneexperi-
mentaimedat resolving thedisputeis Lashley andRussells (1),in which rats rearedin darknessjumpedto a platformfrom a stand placedat a variabledistancefromit. Theforceof the jumpwasfoundto be gradedin
accordancewith the distanceof the platform.Thisis evidencefor nativism.
But,sincethetestswithgraduated distances werenotgivenuntiltherats thirddayinthelight,andafterpretraining, theconclusion wasnotindubitable.Confirmation by anothertechnique isdesirable andhasbeenprovided in the experimentdescribedin this report (2).
A technique of testingforvisualdepthperception whichinvolves no
pretraining at allthe visualcliffwas developed. It is basedonthe
assumption that,givena choice, an animalwillavoiddescending overa vertical edgeto a surface whichappears to befaraway(3).Theapparatus
(Fig9.1)wasconstructed of twothicknesses of glass(24in.by 32in.), parallel tothefloorandheldbymetalsupports 53in.aboveit.Aboard(4
in.wide,24in.long,and3 in.high)extended acrosstheglass,dividing it intotwoequalfields.On oneside(thenear side),patterned wallpaper wasinsertedbetweenthe twosheetsof glass.Throughthe clearglassof theotherside(thefar side)thesamepatternwasvisibleon thefloorand also on the walls below the glass surface.
Optically speaking, theedgeononesideoftheboarddropped awayfor
a distance of53in.(making thesimulated cliff), whileontheothersidethe
edgedropped away foronly3in.Thus, twovisual fields existed, bothfilled
withpatterned wallpaper, butthepatternofthefar fieldwasoptically much smallerand denser than that of the other and elicitedmore motion
parallax. (More binocular parallax wasalsopossible atoneedgethanatthe
other,but the rat is probablyinsensitive to this cue.)The fieldswere
matchedfor reflectedluminousintensity.The physicalspace,as distin-
Behavior ofLight-andDark-Reared Ratsona VisualCliff 143
Figure 9.1
(left) Apparatus fortheexperimental condition. Thelarger-checked field isthenearside,
optically; theclear glass field isthefar orcliffside. Fig9.2(right) Apparatus forthe
control condition.
guished from theoptical space, wasidentical onbothsides, since aglass surface waspresent ata distance of3 in.Theonlydifference between the twofields, therefore, wasadifference inoptical stimulation. Other possible
cuesforsafedescent (tactual, olfactory, auditory echolocation, aircurrents,
ortemperature differentials) wereequalized bytheglass. In addition to the experimental condition described here,a control
condition wasincluded, inorder tocheck onthepresence ofanyunknown
factors thatwould make forapreference foroneside. Apiece ofwallpaper
wasinserted between theglassonbothsides(Fig9.2);otherwise, the
apparatus wasidentical to thatfortheexperimental condition. Ifcontrols areadequate, animals should shownopreference foreither sideinthiscase. Subjects fortheexperimental condition were19dark-reared, hooded rats,90daysold,and29light-reared littermates. Twenty minutes after
coming intothelight, thedark-reared ratswere placed ontheapparatus. Ananimal wasplaced onthecenter board inabox, toavoid anyhandling
bias.Itwasthenobserved for5 minutes. Results aresummarized inTable 9.1.Thepercentage ofanimals thatdescended onthenearsidewasnot
significantly different forlightanddark-reared rats.Ofthelight-reared
rats,23descended onthenearside,threedescended onthefarside,and
threeremained ontheboard forall5 minutes. Ofthedark-reared rats,14 descended on the near side, three descended on the far side, and two remained on the board.
Buta comparison ofdescent behavior oftheexperimental animals with thecontrols, forwhom thevisual surface wasnearonbothsides, showed
adifference. Thecontrol group, alllight-reared litter mates oftheexperi-
144 Table
R. D. Walk, E. J. Gibson, & 1. J. Tighe 9.1
Comparison ofLightandDark-Reared Animals onVisu alCliff (Experim ental Group) and Comparsion of Bothwitha No-Cliff ControlGroup. Control group
Experimental group
Percentage descending
on near side
Light-reared
Dark-reared
Light-reared
(N = 29)
(N = 19)
(N = 10)
88.5
82.4
50.0
MeanNo.crossings
0.00
0.06
Percentage of time On near
76.0
57.9
On far On board
10.0 14.0
16.9 25.2
1.70 24.1
61.5 14.4
*Thecontrol grouphadnooptically far side.Reference isto thesamephysical sidethat was far
for the experimental
group.
mental group, showed nopreference indescending fromtheboard; five
wentto eachside.Thisgroupdifferssignificantly fromthe experimental group (p < 0.02).
Evenmoreinteresting isa comparison oftheexploratory behavior ofthe
animals.The light-rearedand dark-rearedrats of the experimental group
againbehaved similarly; mostofthemstayedonthesideof thecenter boardthattheyhadfirstchosen. Of the 43 experimental animals that descendedfromthe board,onlyone crossedto the other side.But the
controlanimals explored backandforth,oftencrossing theboardto the
other side severaltimes.The differencein crossingbehaviorbetween
experimental andcontrolgroupsis highlysignificant (p
without any specialtrainingor correction , seemsalsoto haveproduceda small reduction in variability of judgment . Since constant errors did not shift toward any common norm in this group , it seems likely that this
Estimation of Distance
215
reduction in individual variability, eventhoughsmall, accounted forthe similar smallreduction ingroupvariability.
Analysis ofvariance Inorder tocheck some ofthese trends, ananalysis ofvariance wasperformed onthedatayielded bytheCornell sample, Group E.Table 12.3presents theresults ofthese analyses forpretest and posttest. Theestimates havebeenconverted intologscores, asintheother sample. Ourprincipal interest, again, wasina comparison ofthepretest andposttest, between which training wasgiven. Thegroup variability (vari-
anceduetomen) wasreduced, asintheairman population. Andtheindividualvariability (variance duetomanx target) 5 wasreduced, likewise, from .0157 to .0056.Thisdifference is significant at betterthanthe.01level,
confirming thefinding thatwithin-S consistency isincreased bytraining. Theanalysis of variance permits, also,an examination of theeffect oforder ofpresentation oftarget. Order assuch doesnotproduce significant variance, andonthepretest, order-target interaction isbarely
significant atthe.05level. Butaftertraining, thevariance duetothissource isclearly apparent, perhaps because othersources ofvariance which masked it before havebeenreduced. Itmaybethat5,injudging, always decided
firstwhether thedistance being judged atpresent waslonger orshorter thanthejustprevious one,andperhaps evenwhatfraction ormultiple it was.Thesizeofthedistance stretch orstretches preceding a given judgment, therefore, would haveaneffect onanygiven judgment; infact,a different effect depending on the relative farness or nearness of the particulartargetcurrently to bejudged. Courseof Learning
Having established theeffectiveness oftraining forabsolute judgments ofdistance, thequestion ofthecourse ofthelearning arises. Whatkindof Table 12.3
Analysis ofVariance ofDistance Estimates (inLogs) Made byCornell Ss,Group F Pretest
Source Order
df 1
Mean SquareF .0306
Between Ssinsamegroup
18 1.4778
Targets Order x Target
17 .0216 17 .0260
S x Target
306
Total
359
= .05 .05 Men 34 .1558 1.4029 9.00 < .001 Residual 578 .0107 .0169 1.579 < .05 Total 629 .l v"uJ . ~- . . . . .. . These comparisons by theF testarepresented asa matterof interest , although the variances forthetwogroups areclearly unequal . Theprincipal conclusion issupported by nonparametric tests .
Judgmentsof Distanceover Ground
227
fourcases). Variance wasgreater forthecontrol groupinthreeofthefour cases, butnotsignificantly. Themeanerror, regardless ofsign,wassmaller fortheexperimental groupinallfourcases, significantly so(p< .02)in
three. Theevidence fortransfer oftraining tothese judgments istherefore
equivocal. Theabsence ofacontinuous ground stretch isprobably responsible forthelackofa clear difference between thetwogroups.
Comparison withtraining bycorrection Intheprevious experiment (3)in
which Ssweretrained bya correction procedure, training andtestjudg-
ments weremadeonthesame field. Inthisexperiment Sshadpreliminary
training bya scaling method, ona different field. Improvement inthetwo experiments canbecompared bycalculating thepercentage transfer ineach case. 2 Forthe previous experiment, therewas79%transfer basedon
constant errors; forthepresent one,62%. It seems, then,thattraining with a scale ofdistance isnearly aseffective ascorrection ofjudgments onthe
very same ground where estimation is tested. ExperimentII Method
TheSswereagaindivided intotwogroups, experimental andcontrol, andthe
experimental group wasgiven preliminary training identical withthatgiven the experimental group inExp. I.Both thecontrol andexperimental groups were then required tomake relative judgments forpairs ofdistances overa ground surface. Asingle distance wasthought ofasanimaginary lineextending radially away fromS.Thequestion was,howaccurately could S compare a givenstretch of
groundalongoneradiuswithanotheralonga different radius.
TheSswere taken tothesame athletic field used inExp. Iforrelative judgments. Theground wasmowed grass, andthefield wasempty except forthetwotargets. These wereattheendsoftwodistance stretches making anangle of120withSs
stationpoint,so thathehadto tumhisheadto compare thedistance to eachof
them. Thewideangle wasusedsothatSwould actually compare thedistance
stretches, notmerely theupanddown location oftwotargets located onthesame radius. Thetargets wereofdifferent shapes. Thestandard targetwasa white
triangle (altitude 94cm.andbase90cm.). Thevariable tagetwasa rectangle from Sbyaman concealed behind it.Itransmoothly onbicycle wheels andalways madecontactwiththe ground. Thepaths ofthetwotargets formed a V.withSatthejunction point andthe (91cm.wideand1S3cm.high). Thevariable target wasmoved toward oraway
standard andvariable located at somepointalongthearms.Therewerethree distances forthestandard target, 50,100,and200yd.(Themaximum distance of
thepathalong each armoftheVtothecomers ofthefield was275yd.)Thefield theselines. Theyweredriven flush withtheground, sothattheywereinvisible to
hadbeen surveyed byanengineer andstakes were laidat1-yd. intervals along
228
E. J. Gibson,R. Bergman,& J. Purdy
Ss.Thinwirewasstretchedalongeachlineso thatthe targetmancouldfollowa straightcourse.Thewirealsowasinvisibleto S.
The methodof judgmentwas a combination of a methodof limitsand a
constantmethod.The variabletarget startingat a distanceeithermuchgreater thanor muchlessthanthat of the standard,madea run whichtookit up to the
equal pointandthena considerable distance beyond it.During therun,it stopped 10timesatassigned positions toenable Stomakeajudgment offarther,equal,
or nearer. The total path traversedon one run by the variablestimuluswas
12yd.forthe50-yd. standard, 24forthe100-yd. standard, and48forthe200-yd. standard. Fora typicalrunwherethe standardtargetwas50 yd.,the variable targetwasstartedat 57yd.andstoppedforjudgments at 55,53,52,51,50,49, 48,47,and45.A typicalrunforthe200-yd. standard beganat 228yd.withstops at 220,212,208,204,200,196,192,188,and 180.Approachandwithdrawal runs werealternated,and the startingpointwas alwaysvaried.For eachstandard distancethe variablemade three approachand three withdrawalruns, so that S
made60 judgments foreachstandarddistance. Thetargetmankeptin constant touchwithEby radio, 3 so thatjudgmentsandstopswereco-ordinated. Forhalfthe Ss,the standardstimuluswason the left,andfor halfon the right. Sincethe illumination of the targetschangedfrommorningto afternoon,these
right-left positions wererotated soastobeequally divided between morning and
afternoon. Halfof theSsbegantheirjudgments withthe 50-yd.standarddistance andhalfwiththe200-yd.standarddistance. All60judgments fora givenstandard werecompleted beforegoingon to the next standard.One practicerun (10
judgments) wasgivento makesureSunderstood thetask.A restperiodofabout 10mm.wasgivenaftercompleting judgments foreachstandard. Thejudgments for all three standardsrequiredabout 2 hr.
FourSstookpartin theexperiment at onetime.Eachmanrecorded hisown judgments. TwoEswerealways present andwatched constantly to insurethatS wasrecording intherightblank.TheSsfacedstraight aheadaftereachjudgment, whilethe variabletargetwasmoved.TheE saidJudge whenthe targetwasin
place. TheSwasinformed whena newrunbegan, andwastoldthatforeachrun the directionof movementof the variabletargetwouldbe the same.He wasalso
toldthatthestartingpointwouldvaryforeachrun,andwasaskedto be sureto
compare thestretch ofground between himself andthetwotargets before making his judgment.
The four Ss at each session includedtwo membersof the experimentalgroup
andtwoofthecontrolgroup.Theystoodbehindoneanother,andpositions from front to rear were rotated for the two groups so that any advantageof station
pointwouldbe equalized. Allfourmencouldseethetargetseasily, however. Therewere32 Ss in eachgroup,allairmenin basictrainingat SampsonAir Force Base.
Results
Theeffect oftraining ondifferential sensitivity todistanceTwomeasures of sensitivity wereobtained,grouppsychometric functionsand individual
229
Judgmentsof Distance over Ground 100
U) ill U)
Z 0
7
a. . U) ill
cr:
LL
5
STANDARD
200
CONTROL
GROUP
0 I-Z
DESCENDING
ill U CI:
..., ~ " E "
25
ill a. .
'
I
,
,
,
x,
...
... "' ...
'
,
.... x ;
~
,
. . . .
" "
"
. .
. . . . . . .
SERIES
...
; ~ ;
.
~ ,
,~ "
,
,
YDS
. . .
" "
'
. . .
"
' X- -
- -
-
0 2
DISTANCE
Figure
13
Group
a
function
approaching
function
DL
.
the
To
of
distance
tions
,
tive
yd
errors
(
.
The
The
variable
DL
' s
runs
for
(
10
by
respective
p
.
SD
the
' s
experimental
at
all
A
t
require
distance
of
according
.
and
plotted
as
and
lower
while
The
all
200
-
yd
the
point
. ,
condi
of
-
subjec
standard
distance
positive
constant
ascending
re
variable
at
of
gave
the
-
,
series
each
the
p
gave
differential
' s
those
case
105
)
is
was
' s
concluded
sensitivity
obtained
the
points
responses
,
PSE
None
.
the
' s
The
the
Wood
and
-
their
limens
for
group
this
the
t
that
to
the
significant
since
of
,
control
difference
,
of
on
by
mean
for
made
.
be
those
and
described
DL
variances
therefore
S
s
groups
no
.
'
method
than
in
,
to
DL
experimental
but
equal
similar
mean
lower
( 2
must
all
very
the
,
,
of
the
group
the
Individual
to
presents
of
not
yd
is
of
.
from
slightly
assumption
,
were
.
observations
did
200
percentage
typical
responses
obtained
distances
the
from
)
method
control
are
It
at
farther
.
were
3
.
and
control
series
individuals
,
Table
confidence
,
function
fairly
outward
standard
)
)
both
paired
the
of
of
.
for
standard
by
level
)
set
equal
a
the
is
descending
this
group
three
test
or
standard
420
plot
displaced
of
( PSE
each
,
standard
,
,
for
This
than
' s
the
nearer
as
function
nearer
obtained
with
plotted
distribution
was
farther
equality
six
the
,
DL
,
of
was
.
variable
the
subjective
shows
case
set
(
averaging
worth
236
.
functions
farther
approach
ones
by
and
expected
this
group
responses
psychometric
,
. 2
the
in
204
, YDS
.
approaching
showing
negative
scale
13
control
of
group
equal
variable
equality
at
. 05
the
,
Figure
the
the
percentage distance
nearer
.
with
of
The
variable
obtain
sponses
not
.
of
' s
VARIABLE
. 2
psychometric
variable
OF
' s
does
reaches
the
training
distance
.
test
with
.
Limens
a
230
E. J. Gibson, R. Bergman , &: J. Purdy
obtained from the group psychometric functions likewise showed no significant changeas a result of the training. The effect of illumination of the standardand the effect of order (beginning with the lOO-yd. standard vs. beginning with the 50-yd. standard) on the 0 L were calculatedseparatelyfor experimentaland control groups. For illumination, no significantdifferencein DL was found for either group. That is, it madeno differencein the limen whether the standardtarget was illuminated by direct sunlight or not. Order had no significant effect on the DL for the experimental group, but the control group had significantly lower DL's for two of the three standards(50 yd. and 100 yd.) when 5 began his judgments with the standardat 50 yd., rather than at 200. Constanterror as a function of training and luminanceof standard The PSE did not differ significantly betweenexperimentaland control groups. However, there was a small constant error for both groups when ascending and descendingtrials were averaged(seeTable 13.3). The CE, small as it is, tends to be positive- that is, the variable looked the samedistanceaway as the standardwhen it was slightly farther. It might be askedwhether bright sunlight shining directly on the standard target would make it appear nearer than when it is not directly illuminated. Manv- Ss commentedon the apparentdifferencein brightness dependingon the sun's location with respectto the targets. The PSE's were therefore compared for the three standard distanceswhen the standard target was in a position to be directly illuminated and when it was not (when it "looked brighter" or "looked shaded"). None of these differed significantly from the standarddistance. In other words, no CE was producedby direct or indirect illumination of the standardtarget.
Table ._ 13 3 -Mean -Distan Indivi DL ' s ( , i Yard ) . n and PSE ' s ( i Ya ) n at Th St D Diffe Lime Po of Su E Cont Exp Co .-200 A ~ _ Mean SD Mea SD Me SO M S 50 yd 1 . 3 . 5 1 . 2 1 . 5 9 50 . 6 4 1 . 4 2 5 . 2 1 . 7 0 100 2 7 1 3 6 0 8 2 10 3 3 0 9 1 4 8 3 5 2 4sign 4 572 1 6the 20 9 7dis 6,2 7as5-positi ~ J .two -CEo These value diffe from sta sh
OL asa functionof distance As Table 13.3 shows, the DL in yards increases with distance. The SO likewise increases , and is notably large. There was considerablevariability in DL's for different individuals. The increment
Judgmentsof Distanceover Ground
231
required toproduce achange injudgment canbeexpressed asapercentage of the standard (AD x
100
Forthethreestandard distances, themeanDLisconstant at about2.5%. Overtherangeofdistances used, Webers Lawappears tohold. Effect oftarget separation onSD Thejudgments ofdistance obtained inthis
study required thatScompare twostretches ofground receding radially from him, thetworadii separated byanangle of120.Bycomparing the present results withthoseofTeichner, Kobrick, andWehrkamp (8),the effect ofincreasing angular separation ofthestandard andvariable target canberoughly determined, since theirtargets wereseparated byonly3 mm. ofarc.They used asestimates oflinear threshold theSDofSssettings around theCE.Comparable measures werecomputed forourdata.Foreach Sforeach run,weobtained thefirstvariable distance judged equaltothe
standard distance. TheSDofthisdistribution is givenbelowforeach
standard distance (infeet,ratherthanyards,to conform withtheother
study):
Distance
SD
l5Oft. 300 ft.
10.83 ft. 21.48 ft.
600 ft.
40.80 ft.
TheseSDsareallhigherthanthoseobtained byTeichner et al.forthe
same distances. Here theyareabout 7%ofthestandard distance, compared toabout -% 2 intheother study. Thus judgment ofcomparative distances overground giveshigher threshold values thandoesjudgment ofthe coincidence oralignment oftwotargets at a distance. Butreasonably
sensitive judgments canclearly bemadeforcomparative distances oftwo
ground stretches. Discussion
Theresults ofExp. I showed clearly thatpreliminary training witha scale
ofdistance improves absolute estimation ofthedistance to anunfamiliar targetinanewlocation. Theground surface itselfprovides a stimulus basis
forSsjudgment inbothtraining andtestfields. TheSobserves along an imaginary linestretching from hisfeettothetarget. Certain optical propertiesofanysuchlineonanylevelterrainwouldremain constant in the stimulus array: farther stretches (relative tonearer) arecharacterized by decreased sizeandincreased density oftexture particles, byincreased
232
E. J. Gibson , R. Bergman , & J. Purdy
uncrossed disparity, and by increaseduncrossedparallactic motion (4). Vertical position of a target in the field of view also moves upward with increasing distance. The other classical"cues" for distance are probably '--'
irrelevant for the present situation . Familiar size and interposition
have
been eliminatedin the experiment, and the kinestheticcuesare probably of little value for the distancesemployed.
Thesevariables , we believe, are in a psychophysical correspondence with impressions of distancealongtheground. The conceptualscaleof yards would have to be related to the concomitant gradients of stimulation by meansof learning. The relationship, in fact, was vastly improved by the training provided in Exp. I. The S can bring with him a scalerelationship which will have a beneficial
effect on estimations
in a new location ; even
the absence of any objectively confirmed reference point on the new field seemed to be no handicap. And if the yard scale was related, in the training
process, to stimulation provided by a continuousstretch of ground, it is not surprising that transfer should be partial or absent when the intervening surfaceis interrupted by objects or hills or gulleys, aswas the casewith the " landmark " targets . The present experiment does not permit us to evaluate the three compo nents included in the scale training . The " unit " of 10 yd ., for instance, may
have played little role. That the fractionation into 25-yd . intervals had an effect is revealed by a comparison of the categories used for estimation by
control and experimentalgroups. Both tended to use responsecategories which are multiples of 25, but the experimentalgroup used many more564 to Group C's 459. The difference is significant (p < .01). The far boundary of the training scale(300 yd.) appearedto exert an effect on the judgments of distancesgreater than this value. Further experimentwith the componenetsgiven singly might throw more light on what 5 learns. ExperimentII was designedto discoverwhether the training receivedby members of the experimental group increasedtheir sensitivity to small differencesin distance magnitudes. Since the DL' s for the experimental group were not significantly lower than the control group's, it must be concluded
that differentiation
within
the distance
dimension
was not in -
creased.It should not be concludedthat the relative judgment of distance cannot be improved, however; the amount of practice may have been too little, for 5s' performancewas already quite good and was perhapsat an asymptote. Also, the judgment madein the training situation was not very similar to that made in the test of relative judgment. The training actually was concernedwith judgment of grosserdistancemagnitudesthan was the test .
The results of Exp. II do, however, strengthen our assumptionthat a
psychophysical scaleexistsin which stimulationderivingfrom a ground surfaceis correlatedwith perceptualjudgments of greateror lesserdistance.
Judgments of Distance over Ground
233
Highly typical psychophysicalfunctions were found (see Fig. 13.2). Since the targets as suchprovided little basisfor a judgment of their distanceand a judgment of alignment was impossible, it follows that for any object in contact with the ground the perception of distance of the object may
be determinedby the impressionof distanceof the background surfaceat the point of contact . For an object not in contact with the ground , the perception of distance
must dependlargely on stimulation deriving from the object itself yielding impressionsof edges, depth, etc. The classicallist of cues, as well as factors of knowledge and inference, would be relevant. An experimentby Bourdon 1; and 10, p. 670) obtained DL's for judgments of relative nearnessand famess of two luminous circles, placed at right angles to one another at distancesaveraging around 20 meters. The surroundings were dark. The DL's for this judgment were about 22% of the standarddistance, compared to 2! % in our study (DL defined as one-half the interval of uncertainty). The absence of stimulation from a ground surface in Bourdon 's experiment
is probably the principal reasonfor the differencein results. Convergence, retinal disparity, and motion parallax were all operative to some extent in Bourdon 's situation , since when he modified it by restricting the conditions
to monocular vision and a motionless head, he obtained completely equivocal judgments. For a pair of objects in contact with a surface, but located so that the stretch of distance to one coincides or nearly coincides with that to the other , the perception of distance as such will be accompanied or even
supplantedby a different type of tridimensionalperception, the impression of in front of or behind. For this judgment, there is a different basis in stimulation . The object behind will be optically above the one in front and
vice versa. We cannot agree with Teichner, Kobrick, and Wehrkamp (8) as to the prevailing importance of this type of stimulation for impressions of distance. When the two target objects are not optically adjacent, the observer must discriminate the respective stretches of distance in order to
decidewhich object is farther or nearer. This is a frequent type of judgment in daily life, and it is probably associatedwith locomotion of all kinds. Summary The purpose of the two experiments described was to determine the effect of
training on (a) absolute judgments of distance, in yards, and (b) relative (nearerfarther) judgments of distanceto variable targets. Training with a scaleof distance was given the experimentalgroup in both experiments. A 300-yd. stretch along a grasssurfacewas bounded for 5, and a la -yd. unit designatedfor his scale. Then 5 made corrected fractionation judgments until the ground was divided into 25-yd. intervals. Following training, Ss of Exp. I were taken to a different field where they
234
&: J. Purdy
madeabsoluteestimationsof the distance, in yards, to unfamiliar targets at varying distances(52 to 395 yd. away). A control group made the absolute estimations without previous training. In Exp. II, Ss were taken to a large field and asked to make relativejudgments of the distancefrom S to two targets separatedby 120 . The standardwas set at 50, 100, or 200 yd. from S. The method of judgment was a combination of a method of limits and a constant method. A control group made the samejudgments without pretraining. Resultsof Exp. I indicated that Sswho receivedthe pretraining were superior to control Ss in both constant and variable error. Absolute estimation was improved even though Ss were not tested in the same field where they were trained, the targets were unfamiliar, and the distancesvaried. It was proposed that S learneda scale relating responses , in yards, to gradients of stimulation deriving from the ground surface. Pretraining with a scaleof distancedid not, however, lower DL's for distancein Exp. II. Sensitivity of the psychophysicalrelationship is apparently not increased by this training, although the yard-responsesbecome more accurately tied to it . The DL for distancewas about 2! % of the standarddistanceand was constant for the three standard distancesemployed. The fact that consistent psychophysical functions of the expected kind were found with wide angular separation of the standard and comparison target strengthens the hypothesis that gradients of stimulation made availableby the ground surfaceitself are in correspondencewith impressionsof distanceof varying magnitudes. Notes 1. This research wassupportedin part by the UnitedStatesAir ForceunderContractNo. 33(038) 22373monitoredby the Perceptualand Motor SkillsLaboratory , HumanResources Research Center,LacklandAir ForceBase . Permission is grantedfor reproduction , translation , publication , use, anddisposalin wholeandin partby or for the UnitedStates Government . Theresearch wasassisted by Dr. LeighMinturn, who actedasexperimenter for the scaletraining, andDorothy Serrie , who rodethe bicycle. 2. Percentage transferwascalulatedby the formula
Group C-Group . Group C-Ex100
A transfer comparison wasmaderatherthandirectcomparison of thetwo experimental groups because of thedesirability ofcomparing eachGroupEwithitsowncontrolgroup 3. Portable radioequipment waskindlyloanedby the CornellDepartment of Military Training . References
1. Bourdon, B. La perceptionvisuellede l'espace . Paris: SchleicherFreres, 1902. 2. Dixon , W . J., & Massey, F. J. Introductionto statisticalanalysis. New York: McGraw-Hill , 3.
4.
1951. Gibson, E. J. & Bergman, R. The effect of training on absolute estimation of distance over the ground. ] . expoPsycho I., 1943, 48, 473- 482. Gisbon, J. J. Theperceptionof the visual world. Boston: Houghton Mifflin , 1950.
Judgmentsof Distance over Ground
235
5.Holway ,A.H .influencing ,Jameson ,the D .A .,Zigler ,M .range J.,Hurvich ,in L.free M .,sWarren ,A .B .,&vision Cook ,. .Cambridge E B . Factors magnitude of e rrors pace and telescop Division of Research ,Graduate School ofBusiness Administratio ,Harvard Univ ., , 1945 . 6.Horowitz ,15812 M .W .),Washington &Kappauf ,W ..E ..Aerial target range estimation .(D.S .R .D .,1945 ;Publ . Bd . , No . . : U S Dep . Commerce , 1946 . 7.Princeton Branch ,Fire Control Division ,Frankford Arsenal .Analysis of range estima data . Branch Memorandum No . 20 ,1943 . 8.Teichner ,on W .H .,discrimination Kobrick ,].L..,Quartermaster &Wehrkamp ,Research R .F.Effects of terrain and observ distance depth and Developme Comma , Environmental Protection Division , Report No . 228 , Natick , Mass . , 1954 . 9.Wilcoxon ,R .Some rapid approximate statistical procedures .Stamford ,Conn .:Americ Cyanimid Co . , 1949 . 10 .Woodworth ,R .S .&perimentaJ psychology .New York :Holt ,1938 .
14
DistanceJudgmentby the Method of Fractionation JeanPurdy, Eleanor] . Gibson
As this project progressed , we becamemore and more interestedin how accurate
peoplereallyare in judgingdistances to betraversedand how theydo it. Theyare seldomaskedto estimateverballydistancestretchesin yards or in any absolute scale . But they maketheir way around(in cars, for example ) with no problem. Cars at any distanceare perceivedas constant in size. Is size constancyperceived
peculiarlyfor objects , perhapsderivedfrom knowledge offamiliar size?] .] . Gibson had suggested(1950) that constancywas not a characteristicof object perception but rather was a characteristic of perception of the total layout within which objectswere situated. It should be discoverablein perceptionof stretchesof ground surface, not only objects located on thesesurfaces. It seemsnow that a scalefor such layout constancyshould also be relative in someway to the perceiver. It may begiven in the perceiver's eyeheightin relation to the horizon, which is a constant peculiar to every individual and provides automatically a neat unit of measure-
ment. We did not think of that in 1955, but the desirabilityof studying the consistency (or nonconsistency ) of people 's scalingof the layout and objectscontainedin it had become evident.Still in a psychophysical tradition, we turnedto the method of fractionation. If there were consistencyof perceivedscaleof layout, it should show up with this method. The subjectsmadefractionation judgments as they had in the previous experiment, but this time we were not so much concernedwith learning as with the consistencyand unity of the dimensions of their perceivedworld.
Theresultswerevery illuminatingfor our question . Thesubjectswere, on the whole, surprisingly accurate. More important, errors were not related to visual angle subtendedon the retina; the subjectsdid not make "retinal matches" but showedgood constancyfor stretchesover ground. As Gibson said (1950, p. 181), "Scale, not size is actually what remains constant in perception." But what gave
thescale ?Not correction , certainly, as it waspresented in this experiment . A large group of subjects that receivedcorrection was not superior in fractionation to a group of naive subjects. We had no answer at that time. What is the information
238
J. Purdy & E. J. Gibson
for the remarkableconsistency of perceiveddimensionsof the spatial layout? If learningis involved, what kind of learning? Certainlynot simplereinforcement of responses .
Two questions beggedfor an answer. What information is available for
perceivingconsistent , accuratedimensions of thegroundand its furnishingsin our environment ? What kind of a theory of perceptuallearningcan we look to if simple correctionor reinforcementis not the answer? -
-
r
~
The literature of space perception abounds in both experiments and theo -
retical discussionsof the perceivedsize and distanceof objects.! The continuum of distanceitself, however, has seldom been specificallystudied as
a psychological dimension . That distancemustbe perceivedaccurately for objects to remain constant in their dimensions is a generally accepted proposition . Errors of constancy , indeed, have been attributed to errors in the estimation of distance (3; 5, p. 151). It is time to inquire , therefore , what
kind of psychological scalefor a dimension of continuing distanceexists. Can an observer tell when far and near stretchesof distanceare equal, or when one is half or twice the other? Gibson predicted that constancy would hold for the distancesbetween objects (2, p. 165) on the hypothesis that constancy is a property not merely of perceived objects but of perceived tridimensional
space itself . Little
evidence is available
on this ques -
tion . Gilinsky (3, p. 473 ff .) had two Os bisect distances from 8 to 200 ft .
long, startingfrom o . Shepredictedthat physicallyequalunitsof distance would
appear shorter as their distance from 0 increased . This result appar -
ently was found in her data, but the small number of 0 ' s used make further
experimentsdesirable, especiallysince the results of recent size constancy experiments indicate that constancy or even " over -constancy " (4) holds for
objects at very long distances. It would appearreasonablein the light of such a result, that " constancy ," or better , objective accuracy, should char-
acterizethe judgments of all the physical dimensionsof a surfacelike the ground. In the presentexperimentthe nature of a psychologicalscalefor distance was investigated by a method of fractionation. The 0 was askedto bisect or trisect distancesalong an imaginary line stretching over the ground from 0 to an indicated
marker . A further group of Os was available whose
judgments were corrected as they were made. These fractionation judgments had servedas a training procedurein another experiment (1). Method
The experiment took place in a quadrangle350 yd. long. The ground was mowed grass, intersectedby three walks. Trees and buildings lined the sides, but the view
DistanceJudgmentby Method of Fractionation 239 Table 14.1
Planof Fractionation Judgments
Serial
Bicycle Approaching
Bicycle Withdrawing
No.of
Stretches
Stretches
Stretches
Stretches
Judgment Bisected (yd.) Trisected (yd.) Bisected (yd.) Trisected (yd.) 1 2 3
300(300to 0) 150(150to 0) 150(300to 150)
300(0to 300) 150(150to 300) 150(0to 150)
4
75(300to225)
5
75(225to 150)
50(25to 75)
75(150to 75)
50(100to 150)
75(75to 0)
50(175to 225)
6
50(200to 150)
7 8
75(75to 150)
50 (125 to 75)
9 10
75(0to 75)
75 (150 to 225)
50 (50 to 0)
75 (225 to 300)
totheendofthequadrangle wasunobstructed. A300-yd. stretch waschosen for
0s scaling operations. The0 stood atoneendofthequadrangle, accompanied byE.Abicycle andridermoved upordown thequadrangle, asneeded, asan indicator ofthedivision point which 0 wasasked tofind. Thebicycle rider hada prearranged schedule ofmovements tofollow andcould besignalled byawhistle
when 0 wanted himtostop. Thearea hadbeen surveyed andhadmarks at25-yd.
intervals to guidetherider,butthesemarkswereinvisible to 0.
Each0 made10fractionation judgments. Sixofthesewerebisections andfour
weretrisections. A !/bt
istheactofstopping thebicycle at halfthedistance
covered bythestretch; atrisection istheactofstopping thebicycle atone-third thedistance covered, thejudgment inthelattercasebeingthefirstthirdorthe nearest thirdofthestretch covered. Table 14.1shows theactual stretches divided, andtheorder inwhich thefractionations were made. Theplanwassuchthatthe truedivision points forthe10judgments made upa series at25-yd. intervals from 25through 250yd.asindicated intheleft-hand column ofTable 2.TheOswere
divided intotwoequal subgroups depending onwhether thebicycle rodeaway from 0 ortoward him. Theorder ofjudging, andwhether 0 halved ortrisected, varied between thetwosubgroups soastokeep within bounds themileage required of the bicycle rider. The first three bisections, however, were comparable forwithdrawal andapproach trials. Before 0 wasasked tomake agiven judgment, markers were setuptoindicate thetwoends ofthestretch thathewastodivide. Themarkers were plain white rectangles, unlabeled. Insome trials, 0s ownstation point wastheorigin ofthe distance stretch indicated, sothatonlyonemarker wasrequired. The0 wasbrought tohisstation point, given hisinstructions, andasked toturn hisback while themarkers were setup.When these were inplace andthebicycle atthestarting point ofthestretch tobedivided, Easked 0 toturnandjudge when themoving target hadreached thedivision point ofthespecified fraction. TheE
240
J. Purdy & E. J. Gibson
signalled
the
one
rider
further
when
interval
of
the
a
was
permitted
markers
priate
labeled
,
as
of
53
Airmen
the
the
40
same
of
with
as
,
the
half
with
judgments
his
distant
error
were
.
and
Air
,
the
rider
were
completed
Sampson
setting
during
.
extent
yards
first
way
bicycle
There
the
approach
at
his
other
withdrawing
and
bicycle
training
.
made
were
the
basic
O
bicycle
of
with
the
by
to
direction
number
satisfied
faced
corrected
judgments
watched
in
not
measured
the
were
the
was
then
invisible
with
the
as
whom
were
see
a
He
was
was
who
with
points
If
error
half
as
to
division
group
of
. "
.
that
and
group
Stop
The
point ~
approaching
a
yard
.
confered
separate
that
"
permitted
judgments
fixed
bicycle
The
said
was
between
aid
a
adjustment
set
and
at
were
it
Base
the
appro
67
watched
Force
except
permanent
up
There
14
,
,
as
-
in
this
withdraw
.
All
.
Results
The
data
Table
obtained
14
for
. 2
each
.
Group
of
error
from
too
10
large
;
a
0
the
segment
,
interval
the
error
14
.
Accuracy
1
fractionation
-
14
given
of
. 1
or
of
-
very
largest
CE
to
error
is
is
3
be
%
7
bisection
.
'
equal
given
positive
him
if
. )
be
was
that
the
angle
small
can
are
A
so
that
visual
too
s
.
stretch
noted
,
The
the
subject
the
nearer
length
found
of
by
It
size
the
third
the
consulting
"
three
target
or
third
2
,
distance
)
. 2
is
the
a
.
stretches
are
directly
related
error
. "
judgments
of
50
-
yd
the
.
,
the
two
means
t
of
yd
is
a
20
On
'
but
The
,
the
average
fractionated
the
the
300
-
yd
direction
subgroups
significantly
are
distance
for
by
s
even
for
be
affected
-
average
total
. 2
uncorrected
differ
CE
,
the
the
to
be
0
tendency
.
to
21
act
When
corre
)
fractionated
to
one
.
the
.
There
stretch
to
the
results
group
bisection
appear
When
Nine
these
fraction
fractionated
length
for
.
not
for
the
uncorrected
fractionated
the
3
of
first
total
of
,
.
were
Only
6
the
point
who
above
stretch
Table
Os
.
go
distance
half
( see
sign
as
the
a
of
judgments
CE
of
division
feature
the
mean
relation
actual
striking
of
of
the
group
most
distance
direction
of
The
the
to
main
many
the
the
to
from
and
.
the
distance
increasing
movement
in
of
large
ranges
point
for
zero
%
concerns
way
measured
total
when
,
The
. 2
regard
.
-
the
the
-
does
from
( without
increases
SO
in
nearer
the
question
the
for
one
with
of
large
-
data
20
or
only
. 1
fractionated
compared
half
different
CE
of
first
of
error
of
well
significantly
the
and
segment
be
much
presented
divided
divided
may
be
yards
the
points
errors
of
out
one
-
their
amount
indicates
the
that
It
terminal
these
fractionation
of
third
presents
small
sponds
so
in
stretches
a
(
would
The
way
knowledge
the
its
)
are
.
half
Figure
CE
the
extents
,
and
of
judged
.
optical
geometry
fractionated
Table
small
trisections
of
that
too
into
optical
stretch
indicates
was
stretch
(
points
divided
him
by
and
errors
division
negative
nearer
divided
bisections
constant
true
that
segment
is
mean
the
means
the
are
at
150
.
DistanceJudgmentby Method of Fractionation
0 '1 . H 0 'I1 . 0 H ~ 0 H '\~ "Q + ~ ~ r:-. 'r~ ":-. I 0 H C (')1 .-tF )V 0 q ~ ' " " 0 'H Q~ e 0 + .cQ :~- J ~ . ' " 'c:-0 ~~ ..:,u ('t3 -.c-0 :t)=I0 0 tF )U ~ '~ "
t.... HHt* *..... \Q *~ *~ *~ H a rf') ar) to -.: ~ ~ ~ I + + + + + HH I I
' 0
t....
to. . . . .
~
C( )
~
~
~
lr )
\
~
Q
~
~
\
~
~
Q
C( )
* LI ' )
I - t
I
+
0 \
*
~
LI ' )
I - t
~
r:-.. :
Or
~
~
t--:
~
N
rf
rf
)
+
t-...
rl
~ )
rf
'
~ )
rf
H
H
*
I - t
~
I
)
H
LI ' )
+
I - t
HLr
I - t
ON
~
~ ~ 0\
\
~
:-.. :
~
I
I
+
0
Lr )
~ )
*
~
Q
rf
~
~
~
\
N
)
Q
H
* 1 ~ --q ~ 0'--: 0 H H I + + +
~ ~ I
* ~ ~ ~ 0 ~ Lr) I I I
Lr ) rf") '0
~
cx: >
t....
0\
N
to .. ~ (t:) to ... 0 t't') ~ \Q ~ ~ ~ ~ ~ ~ H ~ ~ r~-.;j ..0\ Lr) f") "' 0 ~ 0' 0' t-t' + + I +
"a a "' 0 ~ t-t' ~ I I +
0\
N
CXJ
t.....
H
\ Q
lr )
f' t' )
0 lr )
lr ) t.....
0 ONlr HHH
lr )
0
l/) t'-... ~
0 ON NN
lr )
t-t t-t' I
)
241
242
J. Purdy & E. J. Gibson
+ 20
.
+ 10 .
0
0 .
.
0-
Q
0
-
.
.
0
.
.
0
0
.
0 0
- 10
UNCORRECTED
0
GROUP
. TARGET
APPROACHING
0 TARGET
WITHDRAWING
S FROM
S
- 20 0
25
50
75
100
TRUE
125
150
175
200
225
250
1 / 2 OR 1 / 3 , YDS .
Figure 14.1 Plotof thegroupmean constant errorof thejudged fraction (in yd. fromthetruepoint ) relative to thetruepointofdivision , arranged in acontinuum ofdistance .
'SOA'.lN3W90nr
/ l ~O l / l ::10 3J N~ 3W
and225yd. Themeansfor approachandwithdrawalalsodifferon fractionationsrepresentedby points at 250 yd. and 25 yd. However, these lasttwo differences arenot necessarily dueto thedirectionof movementof thetargetalone.Otherdifferences confounded with directionof movement for all judgmentsafter the third were: (a) the judgmentswere madein differentorders,(b) Oswerenot alwaysmakingthe samekindsof fractiona tion of the samedistancestretch(seeTable 14.1). Thus the difference betweenapproachandwithdrawalCE's for fractionations at 25 and250yd. couldhavebeendueto any of thesevariationsin procedure . In general , the error appearsto be positiveon approachtrialsbut muchlesspositiveor negativeon withdrawaltrials. At the Iso-yd. divisionpoint, for instance , 88% of the approachgroupbut only 53% of the withdrawalgroupmade positiveerrors. No differencein the variability of fractionationwas producedby the directionof targetmovementandthe kind of accompanying fractionation exceptat 200yd. wherethe SDfor the approachconditionwashigher. A comparisonof the accuracyof bisectionsandtrisectionscanbe made by comparingthe meanCE's of halving 50-yd. stretcheswith the mean CE's of taking one-third of 75-yd. stretches . By sucha comparison , the sizeof the CE is no largerfor division into thirds than for divisioninto halves , thoughgroupvariabilityis largerfor the one-third judgments . Variability was also influencedby the distancefrom 0 of the stretch to be fractionated . Trisectionsof 75yd. canbe comparedfor four stretches
Distance Judgment by Methodof Fractionation 243 which begin at 0, 75, 150, and 225 yds. from O. As the distance of the stretch from 0 increases , the SD also increases .
Indicationsof distance"constancy " It might be supposedthat perceptual judgments of distancemagnitudesare basedwholly or in part on the visual angles subtendedat the retina by the stretchesof distance. If two physically equalstretches, one nearand one far, were comparedon this basis, the resulting relative subjective magnitudes would not be equal; the farther stretch would subtend a smaller visual angle and would appear less. In order to makeit appearequal, the further stretch would have to be actually greater (a negative error in our experiment ). Thus if 0 made fractionation
judgmentswholly or partly on the basisof suchinformation, a negative CE would be the result. Gilinsky (3) may have assumedsomething like this in stating that " Perceived distances are foreshortened . The perceived distance
d increaseswith the true distanceD but at a reducedand diminishing rate" (p. 462). The presentdata, however, do not substantiatesucha hypothesis. The errors in general are positive -
that is, a made the nearer segment too
large in comparisonwith the farther. And the high degree of accuracyin general indicates that distance judgment exhibits a kind of constancy analo-
UNCORRECTED
8
GROUP
8 TARGET
APPROACHING
cn 150
0 TARGET
WITHDRAWING
0 >-
x
S
X/
FROM
S / /
N~
/
" "
~
/
0
/
w
/
(, 9
/
0
~
/
w
U Z
/0
MEAN
/
/
8 / 0
75 /
r-
/
C/ )
/
0 Z 0
-
-
-
o0
- : ; -- ~
/
>-,
""
Z 0
0
~ ,
- - -~~~
/,
,
0
.
.,
-
--
-
-
O ..:::::~ ~ ""' ~
- -
.
- - ' - - ---:-: . -
-
- -
. . .,. " .. . ,
~
. ..
, .o ~ ,....." . ... . ....,_.,..._."........,....,....- o
...'
- -- ---
5 -
Z
I
- -~
" '"
-
~
-
-
-
- - -
-
- - =: - - 0 -
-
-
-
-
-
-
-
...-
0 -
,/
~ U
/ /
......-
- 10
/
a:
/
LL
I
/
LL
0 w
/
- 15
/
I
u
/
Z
,
~
I
~
UNCORRECTED GROUP
I
- 20
I
. TARGET
APPROACHING
0 TARGET
WITHDRAWING
S
I
I
FROM
S
I
I
-
- 25 0
25
50
I 75 TRUE
-
' 100 POINT
1_ 125 OF
1/ 2
...,... 1/ 3
JUDGMENT
_150
DIVISION
175 , YDS
200
225
250
.
Figure14.3 ThegroupmeanCEof (a) halvinga 50-yd. stretch(solidline) and(b) dividinga 75-yd. stretchintoone-thirdandtwo-thirds(dottedline) asa functionof thetruedivisionpointin yd. Thedashed curverepresents theexpected errorif 0 wereto dividethestretches sothat theresulting fractionsubtended appropriate fractions of visualangle .
Distance Judgment by Method of Fractionation
245
Theerrorforeachbisection is plottedat thetruehalf-way pointofthe stretchbisected. Forinstance, thepointat 125yd.represents a bisection
of a stretchextending from100yd.to 150yd.Thedottedlinesconnect errorvaluesthat resultedfromdividinga 75-yd.stretchinto intervals subjectivelyequalto one-thirdand two-thirds.
Thehorizontal linedrawn atzeroerrorindicates theexpected judgments
if constancy for distancestretcheswereperfect.Thedashedcurve,on
theotherhand,describes thejudgments expected froma hypothetical 0
(6 ft. tall)whofractionated the stretchesby makingthe two fractions subtend(a)equalvisualanglesin thecaseofhalving50yd.,or (Ii)visual
anglesequalto one-third andtwo-thirds ofthetotalanglesubtended by 75yd.Theobtained judgments arecloseto thelineofobjective accuracy anddo not followthe hypothetical visualanglejudgment curve.The judgments at thefarstretches showsometendency to benegative. Since theerrorissmall andsincethetwohypothetical curves areclosetogether at thefarstretches, it isnotpossible to conclude thattheplotted points follow eitheroneortheothercurveat 175,200,and250yd.Thepoints plottedat 25,50,100,and125yd.followthe constancy line.
Thisdemonstration suggeststhe interdependence of all spatialcon-
stancies. Allthe dimensions of a surface, bothfrontaldimensions (termed size)andthethirddimension (termed depthordistance) showconstancy of
magnitude. Gibsons moregeneral statement maythusbeapplicablethat Scale,notsize,isactually whatremains constant inperception (2,p.181). Theeffectof correction onfractionationSo far,onlythe resultsof uncor-
rectedOshavebeenconsidered. Anotherlargegroupof Oswhoseerrors werecorrected as thejudgments proceeded canbe compared withthem. Table14.2givesmeanCEsandSDs for bothgroups.Thecorrected grouphad12of20CEssignificantly greaterthanzero,whileonly9 ofthe uncorrected groupsweresignificantly greater.A t testcomparing the magnitude oftheCEsofthecorrected anduncorrected groupsshowsthat
fiveoftheCEsaresignificantly larger (p< .05)inthecorrected group(at 125and200yd.withbicycle withdrawing, andat 25,200,and250yd.
withbicycle approaching). ThemeanCEisinnocasesignificantly larger in theuncorrected group.Correction, therefore, doesnotappearto increase theaccuracy of fractionation, whichis,indeed, verygoodalready. The fractionation withcorrection did,however, serveasaneffective training
procedurefor laterabsoluteestimatesof distance(1). Summary
Thepresent studyisanapplication ofa psychological scaling operation (fractionation)toperceived distance. A300-yd. stretch ofgrasscovered flatground wasused
246
J
for
as
.
Purdy
&
judgments
judgment
.
by
Six
so
(
as
.
.
angle
3
no
sub
.
positive
the
related
as
the
bicycle
4
.
of
approaches
withdraws
lie
( up
Perceived
and
every
.
ground
25
and
300
yd
to
distance
follow
on
. )
yd
.
into
of
of
from
ap
-
.
a
the
halves
distance
hypothetical
curve
magnitude
of
the
visual
.
motion
,
to
magnitudes
stretches
direction
bicycle
distance
based
distance
the
would
his
point
.
judgments
being
different
indicated
division
intervals
magnitudes
the
0
:
.
their
and
correct
different
point
of
physical
for
to
of
either
the
target
less
,
the
positive
CE
or
tends
to
negative
be
when
.
Fractionation
was
with
from
by
is
.
,
the
points
yd
accuracy
tendency
tended
Error
-
stretch
with
conclusions
good
result
,
division
stretches
well
is
this
considered
made
25D
following
down
he
true
the
very
would
or
what
the
divide
correspond
There
of
through
the
with
which
at
were
one
)
can
to
up
rider
trisections
suggest
pear
moved
that
thirds
2
bicycle
point
Observers
or
Gibson
four
chosen
results
1
.
the
station
The
J
A
and
stretches
.
stopping
bisections
zero
E
of
variability
distances
was
reduced
not
improved
by
correcting
0
' s
errors
,
nor
.
Note
1
.
This
No
research
.
33
( 038
Resources
tion
States
was
) -
supported
22373
Research
,
translation
Government
in
monitored
Center
,
publication
part
by
,
Lackland
,
use
by
the
Air
,
and
the
United
Perceptual
Force
disposal
States
and
Base
in
.
whole
Air
Motor
Force
Permission
in
part
Contract
Laboratory
is
and
under
Skills
,
granted
by
for
or
for
Human
reproduc
the
-
United
.
References 1.Gibson , E. J. Bergman , R., &: Purdy , J. Theeffect ofpriortraining withascale of distance onabsolute and relative estimation ofdistance over ground .J. expo Psycho /., 1955 ,50,97-105 . 2.Gibson ,J.J.Perception ofthe visual world .Boston :Houghton Mifflin , 1950 . 3.Gilinsky , A. S. Perceived size anddistance invisual space . Psychol . Rev ., 1951 , 58, 460 -482 . 4.Gilinsky ,A.S.Perception ofsize ofobjects atvarious distances . USAF Personnel Trainin Res . Cent .Bull ., 1954 ,No.TR -54 -92. 5.Smith , W.M. Amethodological study ofsize -distance perception ,J. Psycho /., 1953 , 35 , 143 -153 .
15
Continuous
Perspective
Perception
of
James
J
.
Constancy
in
of
There
move
don
' t
rigid
of
had
"
the
effect
had
the
transformation
-
sisting
forward
of
for
closely
the
to
the
Wallach
rather
to
than
The
looking
shadow
innovative
these
It
was
displays
that
the
static
such
we
objects
and
pathways
here
and
as
-
approach
we
things
them
these
.
Johansson
in
,
;
constancies
' s
if
?
they
What
in
was
information
for
absolutely
shape
and
stimulus
to
for
the
screen
did
was
perfectly
Journalof Experimental Psychology , 1957, 54, 120- 138.
was
more
later
empha
and
and
stimulation
of
-
form
.
memory
over
role
-
brought
,
time
.
perspective
.
in
the
depth
events
rigidity
the
rigidity
was
per
here
experience
investigate
there
under
he
of
displays
convincing
information
of
of
to
so
is
although
past
,
invariance
-
perception
by
)
depth
underlying
rigidity
,
on
kinetic
1953
of
the
Wertheimer
monograph
exposing
and
transformations
designed
for
used
(
in
.
the
theory
perception
effect
)
and
' Connell
time
for
motion
1957
' s
for
on
of
(
,
the
that
depth
information
O
,
at
role
Gibson
Johansson
change
work
the
.
1950
But
location
principles
J
,
and
.
of
shape
.
in
Wallach
kinetic
of
J
1923
geometry
the
;
theory
perception
as
here
setup
as
appeared
effectiveness
,
explain
caster
.
the
projective
reprinted
as
The
by
' s
for
experiment
transformations
resemble
time
of
sought
,
a
fate
in
Wertheimer
importance
just
constancy
substance
understand
by
"
motion
things
first
related
sized
is
not
perceived
grow
we
to
common
discovered
of
that
properties
led
Perception
role
,
discussed
that
things
of
expounded
of
been
with
that
Event
recently
concern
do
was
suggested
of
or
How
surface
was
constancy
away
.
ground
it
layout
and
substance
law
in
the
and
shapes
move
these
and
suggested
and
,
?
like
shapes
Configurations
a
we
rigidity
them
things
surrounding
,
as
questions
the
Gibson
perceived
environment
for
was
entire
of
shrink
information
It
the
maintain
perception
of
.
paper
constancy
the
to
they
the
of
also
in
appear
J
previous
scale
is
around
is
the
the
.
Eleanor
dimensions
with
and
Motion
,
perceived
characterizes
are
Rigid
Gibson
connection
there
Transformations
most
not
the
experiment
skeptical
need
flat
to
,
was
witness
but
have
the
also
watching
depth
transformations
itself
or
248
J. J. Gibson& E. J. Gibson
were
inevitably
made
seen
movies
Francisco
of
,
address
where
to
Oxford
,
in
could
The
) ,
all
be
children
in
rotation
,
each
of
time
the
years
ago
of
,
' oeuvre
( off
discoveries
went
on
nonchange
at
he
the
she
never
rectangular
.
I
swinging
)
his
film
,
it
wire
the
to
a
if
.
transparent
paths
in
seminar
experiment
wire
of
movement
ourselves
surface
to
wondering
of
film
on
around
the
and
the
then
sudents
( made
San
presidential
travelled
show
the
We
to
and
demonstrating
made
a
The
.
along
,
still
surface
We
)
.
hear
separated
.
screen
also
that
is
,
experiment
it
rigidity
,
and
a
" blobs
"
fared
referred
at
a
as
theory
and
a
our
practised
also
two
path
of
An
own
-duality - - -- - -
the
stimuli
"
,
tive
, "
a
in
is
,
it
in
underwent
,
J . J .
Gibson
all
detected
need
and
irregular
use
,
of
1979
,
pp
in
when
.
in
182
ff
later
arriving
in
objects
.) ,
act
be
was
great
,
an
object
,
observer
-
shape
in
phenomenon
of
the
The
swinging
of
shadow
This
An
is
familiar
movement
of
of
a
six
combination
does
of
any
,
of
ways
exhaust
it
six
-
in
. !
ways
dark
motion
the
,
world
another
,
case
it
the
( 6
in
that
gave
on
a
plane
kind
was
light
of
a
three
mechanical
of
this
of
always
background
of
and
of
yields
other
perception
not
suggested
sequence
transpositions
possibilities
could
7 )
pattern
continuous
a
,
a
any
stimulus
a
,
which
world
shapes
three
transfonnations
but
The
underwent
the
motion
the
motion
space
not
grouping
,
continuous
in
plane
optical
motions
geometrical
motion
a
of
of
screen
in
one
.
,
.
Constancy
the
was
and
varieties
kind
picture
transformations
the
rigid
in
that
motion
moving
which
a
motion
texture
on
.
locomotion
perception
one
of
geometrical
forms
concept
objects
( see
of
the
that
impression
"
a
cast
although
shadow
object
are
of
that
change
,
the
.
memory
the
during
movement
for
hypothesis
an
and
about
survey
as
no
and
of
straightfonvard
the
space
object
types
casting
rotation
more
.." .
exploratory
serve
,
the
" ideal
perception
only
of
through
Furthermore
,
is
and
change
detects
or
which
path
moving
more
observer
that
through
its
saw
the
more
both
or
statically
screen
arc
geometrical
not
his
~ ~ ~~ - -
of
duality
visual
about
rperceptual - ' - - r
a
as
that
us
perceive
two
or
and
impress
to
perceive
the
the
never
well
information
but
.
subject
perceptual
of
obtains
'of"' J
as
to
of
event
the
to
behind
form
particular
since
even
to
system
completely
angle
perception
,
was
visual
demonstrates
constant
simultaneous
involved
that
the
and
object
the
the
,
presented
the
detects
both
experiment
of
time
The
of
this
ability
sees
shape
observer
of
the
same
.
or
of
was
simultaneously
If
year
to
arcs
d
gate
them
with
year
layers
a
.
One
If
sabbatical
more
different
moving
like
took
and
Association
that
four
chef
rather
,
connection
embarrassing
of
our
in
colloquia
or
in
was
a
little
film
,
reason
them
spend
( thirty
once
,
a
depth
that
Psychological
of
is
the
at
helped
to
number
me
in
for
American
about
a
visible
,
showed
it
moving
beautifully
as
to
display
moving
the
although
really
final
of
were
,
object
husband
3
fact
,
that
net
my
we
occasionally
an
experiment
Division
where
Europe
as
this
.
perspec
rigid
-
surface
rotations
movement
geometrical
.
kind
Perception of RigidMotion
249
(but only a few exampleswere presented) it arousedperceptions of nonrigid or elastic surfacemotions, of the kind exemplified in the movements of organisms. Of the six rigid phenomenalmotions, three (rotation around the line of sight, transposition up or down, and transposition right or left) are inducedby a stimuluswhich common sensewould call motion; one (transposition along the line of sight) by a stimulus which common sensewould call expansion or contraction; and only the other two (rotation around a horizontal or a vertical axis) by a stimulus which common sensewould call a transformation . Optics, however, demandsgeometrical terms. All six projectedmotions are different parametersof continuousperspectivetransformation, and they are mathematicallyakin. Common sensetells us that the first three optical motions shouldgive the perceptionsthey do (a motion yields a motion) and that the last three should not (how can a changeof size or shapeyield a motion?). The assumptionis that a visual experience hasto resemblevisually the optical stimulusthat producedit. But a better assumptionis that experiencesneedonly correlatewith their stimuli, not replicatethem, and the presenthypothesis says of - that any ~ continuousseQuence - - -. 1 ------- - perspectivetransformationsis the correlateof perceptuallyrigid motion. There was in the film some evidence to suggest that this hypothesis must be qualifiedif the perspectivetransformationsare those obtained with parallel projection instead of polar projection, that is, with the special case of transformations when the focus of projection is at infinity . The two apparentrotations around a horizontal or a vertical axis then seemed to becomesomewhat ambiguousas to rigidity or elasticity, and apparent reversals of direction of rotation appeared. The apparent approach or recessionalso fails of necessityin this casebecausethe changeof sizeof the stimulus disappearswith parallel projection. If the above observationsare verified, the hypothesis should specify perspective transformations with polar projection. Previous experimental work on the kinetic depth effect (14, 15) or on other appearancesof depth in moving fields (3, 12) does not supply evidencefor or againstthe amendedhypothesissincein generalthe changesof shapestudied in them were not polar projective. Theseexveri~ ments, moreover, are mainly concernedwith what canbe calledthe appearanceof internaldepthof an object, whereaswhat we are here talking about is the appearanceof slant depthof the face of an object. The distinction is made clear in the film. The apparentmotionin depthpreviously studied by Smith (13), however, is relevant to our hypothesis. One or two of Wallach's many experiments on the kinetic depth effect (14, p. 212 ff.) are relevant indirectly if the changes of such impoverished stimuli as line segmentsor anglesare restatedin terms of perspectivetransformations. This hypothesis is comparableto, but different from, the principle involved in Wertheimer's "law of common fate" (16) in several respects.
250
J. J. Gibson & E. J. Gibson
Both refer to some kind of motion in or of a grouping of spots or forms, but Wertheimer's law predicts the organization of a figure in the visual field, whereasthis predicts the quality of rigidity of a surfaceor surface-like experiencein space. Wertheimer's law seemsto imply that the various parts of the complex are united by sharing a common motion (such as moving in the samedirection with the samevelocity) but this hypothesis assertsthat any perspectivetransformationis a single motion mathematically, including the size and slant transformationswhere, analytically considered, every part moves in relation to every other. Wertheimer's law leadsto experiments, on "configurationsof motions" (9, 12) in which each part of the complex undergoescomponentsof translation or rotation but the part is not itself transformed; a geometricaltransformation, however, is somethingthat permeatesevery part as well as the whole of a texture, and the apparatusused in the presentexperiment satisfiesthis condition. It might be noted that the problem of how we discriminatethe rigidity of rotating solid objectsand of approachingor recedingsolid objects in the environment is closely connectedwith the traditional problems of shape constancyand size constancy. Langdon has recently shown that the shape constancy of an object is considerably increasedunder conditions highly unfavorable for it when the object is made to rotate (10). Likewise the question of why we see a rigid environment when we move among the solid surfacesaround us is closely connectedwith the traditional problem of spaceperception (8). Aim of the experiment The experiment to be reported sought to answer four questions. First, doesthe appropriateparameterof continuousperspective transformationswith polar projection always give the perception of the changing slant of a constant shape? Second, are the judgments of amount of change of slant away from the picture-plane in good psychophysical correspondencewith the "extent" or "length" of the transformation sequence ? Also, how variable are thesejudgments? Third, is the outcome dependent on or independent of the kind of shape or texture on which the transformation is imposed? Fourth, how accurate, if at all, is the judgment of slant away from the picture-plane when only the static end product of the transformation sequenceis presented to 0 but not the motion leading to it?
Method Apparatus and Stimuli The optical geometry of the apparatusused is shown in Fig. 15.1. The device can be termed a "shadow transformer." Essentially, it presentsto an eye an optic array
Perception of Rigid Motion
251
TRANSLUCENT SCREEN WITHSHADOW CHANGE CHANGE
OF
OF
OF
" V I RTUAL
SLANT OF CASTER
SLANT " OBJECT
)
SHADOW
~
/-
PO SO
/ / /
-
?- -
-
-
. . ~~~- - - - -
VIEW
OF APPARATUS
FROM
ABOVE
PRODUCING
A SLANT
TRANSFORMATION
Figure15.1 The shadowtransformer .
of limited scope within the boundaries of which either static patterns or continuous
perspectivetransformationscan occur. In this optic array, unlike those of everyday vision, the differential light intensities and their structure are under E's control; the pattern is the same for either eye, and the need for differential convergence and
accommodationis eliminated. All the "cuesfor depth," in short, tend to determine a fIat plane except those of form and motion, which are thus isolated for study. The source of this converging array is a window in a translucent screen.
This optical stimulus is artificially produced by the diverging ray sheaffrom a point sourceof light, into which shadowsare introduced by opacities of one sort or another attachedto a transparentplane mount. Rotations or translationsof the mount (on bearings or tracks outside the ray sheaf) yield corresponding transformation sequences of the shadow . This experiment utilized rotation on a vertical axis. The stimuli were the mirror reversals of these moving shadows , visible on the other side of the translucent screen. If an apparent rotation of a " virtual object " is
induced by such a stimulus it should always be opposite to the rotation of the shadow caster, without ambiguity . ...,
-
The seated0 , in a dimly illuminated room facing a large white surface, saw a luminous
square window
36 cm . on a side at a distance
of 1.80 cm ., made
of
translucent plastic 1 in. thick. The light source was fixed at the same distance behind the window as the eye was in front . It was a 300 -w . Sylvania point source
carbonarc lamp, but any lamp with a single filament of small diameter(up to 1 mm. or more) will serve the purpose. The window was visibly flat. Binocular vision was permitted 0 after preliminary work failed to show any differencebetween the use of one or two eyes. The mounts were transparentrectangularsheetsof i -in. plastic,
252
J. J. Gibson& E. J. Gibson
of such size (30 x 100 cm.) that when they were centered and rotated on a
turntable placed midway between the point sourceand the window they could be turned 700 from the parallel plane without the edges being projected within the window . The turntable could be rotated back and forth through an arc of variable length by an adjustableeccentriclinkage, geared to a motor with a variable speed drive. A speedwhich gave 2-sec. cycles of semirotation was chosen, after exploration indicated that an optimum might be in this neighborhood, although the rate was not critical for the experiment . The quantitative variable of this experiment ,
then, was the "length" of the transformation sequence , as expressedin degreesof angular excursion of the turntable. We shall return to this point later. Five degrees of semirotation were presented: 15 , 30 , 45 , 60 , and 70 . Each cycle began with and returned to the parallel plane. Theforms transformed The variety of forms, patterns, and textures that can be projected with this device has been suggestedelsewhere(7). Four were used in the experiment: an amoeboid group of amoeboid dark shapesor spots (the irregular texture), a solid amoeboidcontour form (the irregular form), a squaregroup of dark squares(the regular texture), and a solid square(the regular form). Eachwas cut out of gummed paper and attached to the central area of a transparentmount so that its shadow was projected to the center of the squaretranslucentwindow . With the mount parallel, the regular shadows extended 20 cm. each way in the 36 cm. squarewindow , and the irregular shadows about the same. It may be noted that the "regular" stimuli are constituted of rectilinear contours and alignmentsand the "irregular" stimuli of randomly curved contours and alignments. There are also differencesin symmetry, and perhaps other geometrical properties. The "forms" are bounded by a single closed contour and the "textures" by many closed contours; the total contour length is much greater in the latter stimuli. A texture might be described as a "form of forms," as distinguished from a form as such. These textures were, however, very 'Icoarse"; there were 36 squaresin the "platoon" and 36 "amoebas" in the /,colony." Thevariableprotractor For recording judgments of changeof slant, 0 had before him a sort of protractor with its baselineparallel to the plane of the screen. It bore an adjustablepointer which could be moved to indicate an angle of semirotation. The top side was blank but the bottom side carried another pointer and a scale which could be read accurately by E after each trial . Instructions
and Procedure
The experimental group Before receiving any formal instructions for the experiment , each a was seated and told : " You see in front of you a screen with
a window
in it which will be illuminated during the experiment. I will first show you a moving pattern of dark lines filling the window . If what you seeis a movement of some kind of object , describe it ."
A network (woven wire fencing of a common type) was then placed on the turntable and turned through various excursions. Although the question was intended to suggest neither a deformation in the plane nor a rigid rotation out of
Perception of Rigid Motion
253
the plane, alIOs reported seeing the latter, and spontaneouslyreported different amounts of rotation. The suggestions in the following instructions were hence consideredpermissible: "During the experiment proper, a dark form or pattern will appearin the middle of the window . It will seem to rotate back and forth on a vertical axis- to turn away from the plane of the screenand return. Your task is to judge how far it rotates, or the maximum angle it makeswith the screen. Use the circular model in front of you to make this -judgment." ~ One of the four patterns was then presented at one of the five degrees of transformation for 20 sec., which permitted 10 cycles of stimulation. The 0 had no difficulty in making his judgment during that interval. Twenty such trials (five for each pattern) in an order counterbalancedfor the group were made, and then another 20 trials in reverse order to determine whether a practice effect would appear. The 0 was not told his errors. There were 20 Os in the group. Thecontrolgroup A separategroup of 30 Os was treated as similarly as possible except that the four stimulus patterns were motionless. Only the end product of each transformation sequencewas presented, and only one degree of transformation was used- that with the mount at 60 . For the preliminary exposure, 0 was shown a motionless pattern filling the window, half the group seeingthe network of lines and the other half a less objective cloud-like pattern (this making no difference in the outcome) and he was asked if he saw an object of some sort. Then 0 was told that he would seea form or pattern in the middle of the window . It might be parallel or slantedaway from the screen. If he saw it slantedaway from the plane he was askedto judge the angle it made using the model in front of him. Four trials 1were given (one for each pattern) in an order counterbalancedfor the group. Results
Theexperimental group The first question is whether alIOs saw the changing slant of a rigid shape. As stated above, all did at the outset. During the 40 trials which followed, eachof considerableduration, many spontaneous descriptionswere offered, and 8 of the 20 Os observedat some stagethat the display could be seenas a compressionof a two-dimensionalpattern. They were all psychologists. Twelve did not so report, and stated at the end that they had never observedit . The two-dimensionalimpressionsdid not persist long enough to prevent the requestedjudgments of changing slant. There was no difference in this respect between the regular or irregular forms or textures. The secondquestion is whether the judgments of changeof slant are a function of the amount of changeof form. The "length " of the transformation sequenceis expressedas the inverse angular excursion of the shadow caster, and this variable is plotted on the horizontal axis of Fig. 15.2. The
254
(S33 'd930 ) S.lN3V'J90nr ;:jONOI .lVI /\ 300 'd~ ON~ .lS
J. J. Gibson & E. J. Gibson
CHANGEOF SLANT IN OPTICALSTIMULUS (DEGREES )
Figure 15.2 Judgmentsof change of slant as a function of the length of the transformation sequence.
9 .0
.~'.. '. :' 0 - ' : '0
8,
:
:
7
.::.'/ '
:.~ . ' .
! ' ::.
0
: .
: "
""0
" 0. , ....'"".; ' -- --- - -
.
" V
:: /
:
"""' "
" 0.
/ ,, "" . .
.
.." .'0.
\
\
0
\
,
"
:' j
\
"
".
5.
,
\
\ \\
'
\ ,
,
,
\\
\
1
-- ~
:',,/
I
40
.
\
!::./ ' ,1
.
,'. :'
\
\
,
.: , .' / !
6
- - -e
,
0
e./
D - - 0- - REGULARFORM
0
DD
/
00
3,
-
- 0 -
- REGULAR
PATTERN
Q - - -e--- IRREGULARFORM
0\\ .'."'X"'.'- I RREGULAR PATTERN 0
0
10 CHANGE
20
30 OF
SLANT
40
50
I N OPTICAL
60 STI
70 MULUS
Figure 15.3 Variability of judgments as a function of the length of the transformation sequence.
Perception
judgments
are
of
40
plotted
.
underestimate
,
also
for
can
is
the
the
be
Figure
and
slant
"
They
. 3
in
"
,
The
the
but
. 1
the
) ,
the
four
kinds
,
is
obvious
of
.
discover
4
of
and
approach
8
90
.
,
An
the
at
analysis
of
whether
either
errors
variance
multiolicitv
J
.
or
regularity
of
Multiplicity
at
the
or
only
the
weak
,
"
Table
5
"
%
patterns
significantly
"
level
. 2
The
is
form
. "
related
not
.
is
in
of
the
might
strengthened
of
Variance
of
Errors
of
~
Judgments
,
of
Change
of
excursion
turntable
( A
or
of
4
irregular
texture
Practice
( P
x
direction
by
the
( R
( T
)
significant
interac
-
F
1437
1
)
. 07
320
)
72
5
. 12
.
. 56
.
. 93
109
. 47
. 53
R
X
5
19
57
. 69
X
5
19
52
. 63
23
. 07
A
X
R
4
60
. 54
4
. 77
.
A
x
T
4
67
. 09
2
. 66
--
T
x
A
A
x
T
x
:
tests
R
x
R
Note
F
1
5
x
T
19
R
x
76
5
x
76
5
(
those
11
,
.
p
=
. 05
.
.
p
=
. 01
.
p
.
interactions
330
and
which
338
)
are
included
13
. 15
12
. 69
25
19
Only
299
appeared
to
in
.
. 90
453
1
76
5
13
. 68
1
19
5
X
is
theory
Slant
T
P
but
Gestalt
of
( 5 )
A
.
significant
MS
)
or
Form
is
from
'"
judgment
..,
df
Subjects
regularity
made
perhaps
-
Source
Regular
of
expected
be
,
errors
. 1
Analysis
Degree
to
effect
is
that
It
The
tendency
predictions
good
15
is
texturedness
considering
about
.
.
change
between
used
to
group
.
differences
pattern
necessary
the
about
to
edge
.
the
of
begins
an
and
transformation
within
between
slant
,
irregular
a
function
range
becomes
any
a
striking
transformed
are
as
to
more
that
judgments
maximum
Even
gets
mean
tendency
and
what
judgments
255
a
small
regular
suggests
errors
show
however
This
the
surface
.
the
of
the
a
angles
independently
variable
as
30
Motion
representing
for
for
.
Rigid
point
except
and
functions
of
The
not
with
15
s
each
,
15
variable
'
again
does
obtained
SD
,
linear
stimuli
,
phenomenal
graph
the
the
how
.
fall
a
)
such
stimulus
limit
Table
as
is
shows
the
rise
which
5
multiple
to
axis
is
of
question
15
to
similarity
single
related
vertical
function
4
responded
A
the
The
( about
however
(
on
reproductions
of
the
Table
be
of
.
some
. 24
. 13
interest
or
were
used
for
the
.
256
J. J. Gibson& E. J. Gibson
tion between angle and regularity which seemsto reflect the tendency, barely noticeable in the graphs, for the irregular forms ~o depar~ slightly more from linearity at ~he larger angles. All ~he forms in ~his experiment were apparently good enough to carry the transforma~ion, and i~ was ~his which mainly determined the judgments. This answersthe third question. The form of the changeseems to be what is important , not the form itself .
A conception of the various forms that optical changemay take is, however unfamiliar, probably necessaryfor an understandingof the perceptual process .
It may be noted from Table 15.1 that no significant practice effect appearedbetween the first and secondblocks of 20 trials. They have been pooled in Fig. 15.2 and 15.3. The two halves of the data independently warrant the same conclusions, when the curve of Fig. 15.2 is plotted separately for them . The control group
The outcome of the control experiment was radically
different inasmuchas the judgments of slant dependedon the regularity of the form or texture presented. The irregular stimuli, in fact, generally appearedin the planeof the screen(85% of 60 judgments) while the regular stimuli generally appearedat a slant from the screen (97% of 60 judgments). Even for the regular stimuli, however, the mean degree of slant perceived was only 240 (SO about 120) whereasfor the moving regular stimuli the mean had been 610 (SO about 60 ). This is gross underestimation
for the motionless and great accuracy for the moving stimuli . The under-
estimation of slant is consistent with previous researchon static optical forms and optical textures under similar conditions. A trapezoidalform can sometimes arouse an impression of slant, but an exact linkage between the
apparent shapeand the apparent slant (a "psychological invariant") is not obtained (2). A static optical texture with a compressionof texture on one meridian relative to the other induces a perception of surface slant, but
even when the texture is regular the slant is underestimated , and when the texture is lessregular the slant is more underestimated(4, p. 380). The irregular form and the irregular texture displayedin this experiment were evidently not of such a kind as to appear slanted when altered by a
slant transformation, since they generally still looked frontal. The Os, of course, had never seen them beforetransformation .3 A truer statement of the matter is that the family of perspective transformations of the amoeboid stimuli has no unique member with immediately identifiable properties . The
family of perspective transformations of the quadrilateral stimulidoeshave such a member - the square. At the outset there is one in this family which the other members can be transformations of, but none in the former family .
Hence, rectilinearcontour and alignment- a rectilinearstructure- provide
Perceptionof RigidMotion
257
a primarybasisforpresumptions of slantwhichnonrectilinear structure
does not.
Oneof thewritershasarguedthatthereexistsa betterbasisthan
contour forpresumptions orevenperceptions ofsurface slant, namely,
internal texture anditsdensity variables (4) 4 Itmight beexpected, there-
fore,thatthetexture ofamoeboid spotswouldinduce slantmoreoften thantheamoeboid form,andthetexture ofsquares moreslantthanthe square. Neither expectation wasfulfilled. Theexplanation maybethatthe
textures used(36elements, oronlyabout6 eachway)weretoocoarse to
make the density variablesdeterminate.
Evidently a configuration withonlythefeeblest stimulating power for
depthperception, ornoneat all,cannevertheless carrya transformation sequence whichyieldsaccurate depthperception. Thisanswers thefourth
question.
Individual resultsTime didnotpermit more thantwojudgments foreach condition per0 intheexperiment justreported. Itwasthought desirable to runa moreexhaustive seriesin orderto determine the extentof the variable erroroftheslantjudgments forsingle individuals. Accordingly, two0s wererunforfivedayseach,sothat10judgments percondition were available. 5 Table 15.2 shows themeans andSDsofthese judgments. Themeans arestrikingly likethoseforthegroup, especially for01,who underestimates thesmaller anglesandis moreaccurate or overestimates onthelarger. Thesecond 0 tendstounderestimate throughout. Allthe curves, when plotted, areclose tolinearity, andtheparticular pattern makes little difference. TheSDsareofthesame order asforthegroup. Discussion
Kinetic depth effect andmemory Wallach believes thatthekinetic depth ofpastexperience onpresent momentary experience (15,p. 364).He argues thatsince anysingle member ofthesequence looks flatinisolation, thepresent member hasdepth onlybecause thememory traces ofpast members enterintothepresent perception. Heassumes thatonlythe present member ofthesequence canbea stimulus fortheeye.Thisisa perfectly logical extension oftheclassical theory which strictly separates effect ofwhatwecalla transformation sequence mustbeduetosomeeffect
traces andstimuliasdeterminers ofperception (orbehavior). Butdoes itnotreduce thetheory toanabsurdity? Doesa stimulus lastfora second,
a millisecond, or a microsecond? Andwhataboutthedoctrine thata
stimulus isalways achange? Isitnottheoretically preferable tosuppose
thata transformation isa stimulus initsownright,justasa nontransforma-
I
!9 o( 3 ~ uosq !9 ofo(
uos
Perceptionof Rigid Motion
259
tionisa stimulus? Or,stillbetter, thatsequence, aswellaspattern, isa
variable ofstimulation? Isitnotbetter totaketheriskthattraces might vanish from psychological theory thantheriskthatthestimulus might vanish? Perhaps theaddiction ofGestalt theorists totheconcept oftraces iswhathasprevented someofthemfromstudying temporal forms as effectively as theyhavespatialforms. Wallach hasconvincingly demonstrated (andthepresent observations
confirm it)thatamotionless nonsense pattern ofacertain sortona shadow
screen which attheoutset usually looks flatwillusually lookdeepafter0
hasseen ifintransformation. This result may well beconsidered agenuine effect ofmemory, oratleast ofrecognition. Butitisfarfrom proving that
memory tracesarenecessary fortheperception ofrigidmotion. Ifmight provenot an effectof memoryon depthperception but an effectof
memory onthedepthinterpretation ofanambiguous staticpicture. It is possible thattheroleoflearning inspace perception isquitea different
affair, consisting notoftheenriching ofbare stimuli butthediscriminating
of rich stimuli.
The anchoring ofa transformation sequence andtheidentifying oftheobjectA motionless pattern ofsufficient irregularity appears intheplaneofthe
screenevenwhenits shadow-making patternis slantedto the screen.
What, then,would happen tothejudgment ofchange ofslantiftheshadow
casterwererotatedbetween,say,300and600insteadof between00and
60?Would 0 thenseethepattern asdeparting fromandreturning tothe frontal parallel plane? Istheneatcorrespondence between length oftransformation sequence andchange ofslantshown inFig.15.2destroyed in thesecircumstances? These questions leadintofundamental problems of psychophysical scaling andshape constancy; complete answers cannot be given, butenough evidence hasbeenobtained tobeworth reporting here. If the30 to 600cycleis presented to a naive0 witha rectilinear or
regular pattern, judgments aremade withsome accuracy. Theapparent surface doesnotbegin andendintheparallel plane. Ifthecycle ispresented withanirregular pattern, however, thenaive 0 reports a large
change ofslantwhich seems difficult toestimate, andtheobject does seem
to returnto theparallel plane.Thecrucial experience seemsto behisfirst
viewofthemotionless pattern. Heseesanobject buthisfirstimpression of
it isanobject thinner thantheshadow-casting pattern. Allobservations sofar suggestthatif he is thusledto identify at theoutsetan objectof the
wrongshape, thejudgments ofchange ofslantwillpersistently beoutof scale. Hisscale ofslantwillbestretched anddisplaced, asifwere, untilhe
is giventheopportunity ofanchoring if at 00and90 which,at thesame time,enables himto identify therigidshapeoftheobject.
260
J. J. Gibson & E. J. Gibson
Thesizeanddistanceof the virtualobjectareneverdetermined by the
optical stimulus in ourexperiment. Similarly, theslantandshapeofthe
virtualobjectmaybe misdetermined at the outsetby showingit, even
briefly, asa continuous nontransformationa staticfrontal pattern. This
establishes a falseshapeconstancy for the phenomenal object.Whenit rotates,anomalies of spaceperception willoccurof the sortdemonstrated
strikingly by Ameswiththerotatingtrapezoidal window(I).Ifcareis takenbyE,however, to avoidtheprocedure above,theevidence indicates thatan intermediate transformation cyclecanbe correctlyjudgedfromthe
outset,whenever thelengthofthetransformation sequence is sufficient. If a naive0 is firstshownan anchoredtransformation sequencefrom0 to
90, orevenifheisfirstshownthemotionless patternat 45 butistoldnot
to assumethat the objectis necessarily in the planeof the picture,thena 30 to 60 cycleis judgedapproximately as such.Thetentativeconclusion is that a motion consistingof a perspectivetransformationsequence
candetermineboth a definiterigidshapeand a definitechangeof slant
in perception, for a whollyunfamiliar object,withoutthe needof any presumption whatever abouttheprobable shapeof theobjectbasedon memory (2).
Itsoundsveryparadoxical toassertthata change offormofthestimulus canyielda constant formwithchange ofslantinthepercept. Theparadox probably arisesbecause twodifferent meanings of thewordform are employed (5),thefirstbeinggeometrical andthesecond substantial. Inany event,theassertion isinaccurate sincetheevidence nowsuggests thatthe formofthechange offormofthestimulus iswhatdetermines theperception of rigid motion.
Iftheimpression ofsurface rigidity invisual perception canbeaccounted for,theconstancy of shapeofobjects is explained at thesametime.The faceof a uniquesolidobjectis ordinarily givennot as a formbut as a unique family oftransformations to theeye.Thedifference between one solidobjectandanother, incontrast, isnotgivenasa relation ofperspective transformation, noristhedifference between anearlieranda laterstateof a physically changing object.Theperceptual problem oftherecognizing or identifying ofunchanging objects bytheirshapehasto beapproached in the light of these facts. Summary
Continuous perspective transformations ofvaryinglengthwerepresented in2-sec.
cycles toeach0 onthevisibly flatsurface ofa translucent screen. Judgments of the amountof changeof slantof the apparently rigidobjectwerein good correspondence withthelength ofthetransformation sequence, without dependingonthekindofpattern which carried thetransformation. Thepatterns differed
Perceptionof RigidMotion
261
withrespect toregularity vs.irregularity andform vs.texture. Regularity may
havehada small effect onthevariability ofjudgments buttexturedness didnot.
As a control,the samepatternswerepresented motionless at the endof a
transformation sequence. Ingeneral theirregular patterns appeared tobeinthe frontal plane butaltered inshape; theregular pattern appeared atsome degree of slant, butthejudgments were notaccurate. Evidently impressions ofchanging slantareprecise whereas corresponding impressions ofunchanging slantareambiguous orweak. Rectilinear contours andalignments seem toprovide some basis forimpressions ofunchanging slant. Asequence ofperspective transformations, on
theotherhand, seems toyieldanimpression ofchanging slantwhether ornotsuch
regularityis present.
Forirregular unfamiliar patterns there wasevidence tosuggest thattheperceivperceiving ofchange ofslantoftheobject. Misidentification oftheshape atthe ingof therigidshapeof thevirtualobjectis intimately connected withthe
outsetwasaccompanied byanomalies intheperception ofslant.
Theeyeappears tobeverysensitive toacontinuous perspective transformation
intheoptic array. Psychophysical experiments arepossible iftheparameters ofthis stimulus are isolated and controlled. Notes
1.Thisexperiment wasreported bythefirstauthor aspartofanaddress entitled Stimulation
andPerception delivered asretiring President oftheDivision ofExperimental Psychology,
APA,in September, 1955.Theworkwassupported in partby theOffice of Naval
Research under Contract NONR 401(14) withCornell University. Anearly form ofthe apparatus tobedescribed wasconstructed, andpreliminary experiments were performed byH.R.Cort.Thewriters arealsoobligated toDr.0. W.Smith forideas andassistance. 2.isThe 5% level is probably not acceptable here, since some inhomogeneity ofvariance evident in the data. 3.An0 whohadbecome familiar withthefrontal aspect oftheirregular formortexture (suchaseitherE)couldseeanother aspect asslanted. Soalsocouldan0 whohad
persistently observed these patterns undergoing continuous transformations. Presumably these Oshadlearned toidentify thesurface, i.e., torecognize apreviously unfamiliar object in nonfrontal aspects. This, we believe, is Wallachs memory effect in the perception of tridimensionalforms(15). 4.Inthestudy referred to,thecorrelate ofslant wassaidloosely tobea gradient of texture density. Thecorrelate might aswell beexpressed asanunequal density along two meridians, ortheratioofthese densities, oraspecial sortofcompression oftexture. These areallcomprehended inthegeometrical notion ofaperspective transformation.
5. Mr. John Hay kindlyobtainedthese data.
References
I.Ames, A.,Jr.Visual perception andtherotating trapezoidal window. Psychol. Monogr., 1951, 65, No. 7 (Whole No. 324). 2.Beck, J.,&Gibson, J.The relation ofapparent toapparent slant intheperception of objects.]. exp.J. Psycho!., 1955,50, 125133.shape
262
J. J. Gibson &: E. J. Gibson
3. Fisichelli , V. R. Effect of rotational axisanddimensional variations onthereversals of apparent movement inLissajous figures . Amer .]. Psychol ., 1946 , 59, 669- 675. 4. Gibson , J. J. Theperception ofvisual surfaces . Amer . ]. Psycho I., 1950 , 63, 367-384. 5. Gibson , J. J. Whatisaform ?Psycho I. Rev ., 1951 , 58, 403-412. 6. Gibson , J. J. Optical motions andtransformations asstimuli forvisual perception . (Motion picture film). State College , Pa.: Psychol . Cinema Register , 1955 . 7. Gibson , J. J. Optical motions andtransformations asstimuli forvisual perception . Psychol . Rev ., inpress . 8. Gibson , J. J., alum,P., & Rosenblatt , F. Parallax andperspective during aircraft landings . Amer . ]. Psychol ., 1955 , 68, 372- 385. 9. Johannson , G. Configurations inevent perception . Uppsala : Almqvist andWiksell , 1950 . 10.Langdon , J. Theperception ofachanging shape . Quart .J. expo Psychol ., 1951 ,3, 157 - 165 . 11.McNemar , Q. Psychological statistics . (2nded.) NewYork:Wiley , 1955 . 12.Metzger , W. Tiefenerscheinungen inoptischen Bewegungsfeldem . Psycho I. Forsch ., 1935 , 20, 195- 260. 13. Smi~h, W. A. Sensitivity to apparent movement in depthasa functionof "propertyof movement ." J. expo Psycho /., 1951 , 42, 143- 152. 14 Wallach . H., & O'Connell , D. N. Thekineticdeptheffect . J. expo Psycho /., 1953 , 45, 205- 217. 15. Wallach , H., O'Connell , D. N., & Neisser , U. Thememory effectof visualperception of three -dimensional form.J. expo Psycho /., 1953 , 45, 360- 368. 16. Wertheimer , M. Untersuchungen zurLehrevanderGestalt . II. Psycho /. Forsch ., 1923 , 4, 301- 350.
16
Motion
Parallax
as
a
Determinant
of
Perceived
Depth
Eleanor H
J .
.
_ 4
The
Gibson
-
paper
for
layout of us
to
specify from
{ if
]
.
.
Gibson
,
the
of
the
problem
O
.
W
.
Smith
,
or
. Helmholtz
projected
the
in
it
motion of
the
occurs were
and
.
previous
, the
some
of the
Note
the
The
indeed
to
be
questions
deeper the
-
.
experiment
the
exactly
informa
does
conditions
really
how
world
as another
movement
was
up
real
of
to
pattern
of
a
met
,
to
be
problem
of
information
following
eye
in
( p . 43
.
It
seems
that than asserted the
field
corresponding
transformation
of is
not
information
of
can paragraph
elements
register
pick
their the
will
a be
, and
immediately about
up
is
the
Gestalt
the
mutual
directions
to
.
changes
of
displacements difference directly in this
in
or
between related
space
.
may
be
evident objects
is
field the
or
absolute
that
a
As
responds
objects pattern
.
positions
eye
to
in
them to
than
rather have
the
to
eye
suppose
elements
the
can
, rather
,
between
eye the
better
two
specifying the
to
velocities
an
what
easier
sensitive
for
angular
pattern
might
distance
primarily
their
is
involve
is
,
change
velocities
is
this
solution
and
,
,
velocities
eye
in
from
of front
differential
exist
pick
role in
made that
which
how
elements
separation
velocities
to
can
positions
accordingly
tive
things
emphasized
of
angular
film }
awaited eye
understanding
have
separation
method
that
with of
difference
extent
the of
trouble
,
the
convinced
the
surface
:
difficulty
And
one
correspondences
information
theorists
be
of
for in
to . The
metric
arrangements introduction
The
investigation time
phenomenon
any But
further this
needed
surfaces
whether
kind
a ,
this we
separate
explored what
was
features
demonstration convinced
and
James
.. ..
following
tion
serve
,
Flock
what an
array
the to
Such
the rela
what the of
-
the best light
) .
Thisstatement soundsratherlike Marr's muchlaterpronouncements about computation (Marr 1982 ). Butthequestion herewaswhether differentialvelociJournalof Experimental Psychology , 1959, 58, 40- 51.
264
E. J. Gibson , J. J. Gibson , O. W. Smith , & H. Flock
ties carry information for separation in depthand whethertheycanspecify absolute , aswellasrelative , amounts of separation . Theshadow caster setupwasusedagain , thistimewith translucent surfaces in front of thesubject . Theactualsurfaces wereplexiglass andtheywererandomly textured bysprinklingthemwith a cleaning powder . Whentheyweremotionless , theappearance wasof onesurface . As soonasany motionwasintroduced , the surfaces sprangapart, butextentof separation wasambiguous . Notethesugges tion in thediscussion that the information for separation might bedisruption of adjacent orderbetween the two setsof texture . Thesuggestion foreshadows J. J. Gibson 's laterworkingoutofaccretion anddeletion at anedge asinformation for specification of separation . Whatwasemphasized in bothcases is change in opticalstructure , a discovery that hasbeenmadeby othersseveraltimessince then(Ullman1979 ). Thesecond experiment presented a gradientof velocitydifferences ratherthan just twoseparated surfaces , attempting to create a continuous flowfieldof velocity change , on thehypothesis that themetricambiguityof a difference in just two surfaces couldbeovercome by a smoothflow of changing velocityratios. There wasgoodindicationof a correspondence between differential velocityratiosand perceived slantin this case . Theseresultsare interesting in the light of recent research findingson thesamequestion (Anderson1989 ) that undersomecircumstances positiveresults(metriccorrespondence withoutsmoothcontinuityof change ) areobtained . It ispossible thatcontinuityof change becomes afactoronly whenno informationis availablefor an opaque surface . I founda comment writtenin themarginoff . J. Gibson 's reprintof thispaper , "Thetransparencies shouldhavebeen emphasized ." Thesetupwastoomuchlikeemptyspace , perhaps . I conducted mostof thisresearch duringa coupleof summermonthswhenI wastheonlyseniorinvestigator in thelaboratory , my husband andothershaving departed for Brussels wherean international congress wasgoingon. Duringa recentmove , I foundtwo letterswrittento my husband givingdetailsaboutthe earlyprogress of theexperiments : I wasverydiscouraged abouttheexperiment yesterday . 1ranHowardFlock andhecouldn 't discriminate anythingexcept thesmallest separation (1/2 in.) from all therest. WhenI plottedhisjudgments theywerescattered anywhereat all. Then1 friedhim wifh rating(1, 2, 3, 4, 5). Still - worse ~. But todayI havebeengiving him training , demonstrating with an arbitrary scaleandcorrecting him. After25corrected judgments hewasnearlyperfect (didnotconfuse 10different settings ). Buthehaddeliberately attended to the ratiosof thetwomotionspeeds of thepatterns andattached thescaleto the ratios.I amgoingto keeprunninghimandseeif thissticks , andwhether it evergetsconverted to a depthexperience . If it did, it wouldbea realcaseof learning todiscriminate a complex stimulusvariable andgettinga phenome nological change alongwith it (July24, 1957 ).
MotionParallaxin Perceived Depth
265
Laterreportsweremorecheerful:
Theexperiment isgoing fine, really. I have3kinds ofsubjects. Thesummer school people arenttooeager, soIonly askthem toDescribe what yousee. They allsaytwolayers orlevels.Then Iaskhowfarapart?One said 5ft today, andonesaid4 in.Thisiswiththelargest separation. Then I show thesmallest. Some cantseetheseparationthen 1increase ittillthey do.Thesecond class ofSsareasked tomake 20judgments, random order (takes about40mm.). Thethirdclass ofSsareFlock andHay,whoare
practising dailywithcorrection, for10trialsandthennextdaywithout (July 30, 1957).
Thelastmentioned subjects, senior graduate students Howard Flock andJohn Hay,were incorporated intoexperiment IIIofthepaper, a study oftheeffect of giving varied amounts ofinformation about theactual setup tothesubjects. If some sortofHelmholtzian inference principle were toexplain howvelocity differences functioned asa cue todepth, wemight have found aneffect ofconversion ofbidimensional impressions intoexperiences ofdepthreferred tosometimes as
percept-percept coupling (Epstein 1982). Butthisdidnothappen. Thetwomost
practiced subjects made inferences about depth, tobesure, onthebasis ofperceived velocity ratios. Butthey never reported aperceptual change inthedepth perceived. Thisexperiment discouraged anytendency I hadtogoalong witha Helmholtzian ora Brunswikian theory ofperceptual learning. However, itdidnotdiscourage my interest inperceptual learning, andmypursuit ofa viable, adequate theory of perceptual learning took offinnewdirections, ascontinued inpartIV . Motionparallaxis the opticalchangeof the visualfieldof an observer
which results froma change ofhisviewing position. 1 Itisoftendefined as
thesetofapparent motions ofstationary objects which ariseduring locomotion. Psychologists assert thatitisacue forperceiving thedepth oftheobjects, buttheoptical factofmotion parallax must bedistinguished from itscapacity toinduce perceptions. Ithasnotbeenexperimentally demonstrated thatmotions inthefieldofviewwillactually yieldcorresponding judgments ofdepth. Thisisapurely psychological problem. The optics of motion parallax, on the other hand, is a problem for geometry and ecology. Recently, thesuggestion hasbeenmade thata continuous gradient of motions in thefieldofviewwillinduce theperception ofslant-depth
(J.J.Gibson, Olum, &Rosenblatt, 1955) inasmuch astheperception of depthisintimately connected withtheperception ofsurfaces (J.J.Gibson, 1950). Thisstatement alsoneedsexperimental test.Thepurpose ofthe present studyistoinvestigate whatkinds ofmotion inthelightentering
266
E. J. Gibson , J. J. Gibson , O. W. Smith , &: H. Flock
an eye do in fact consistently arousecertain judgments of depth, and what do not .
The experiments must be carried out with artificial motions in a field of view
rather than those obtained
in a natural
environment
if we wish
to study the effect of motion parallaxin isolation from other cuesor stimuli for depth. The variables of size, density, linear perspective, differential blur, and binocular parallax should be eliminated or so reduced as to be ineffective in the array of light entering as eye. A method of achieving this result has been devised, and a suitablecontrol employed.
The experimentalmethod shouldalso precludeactualmovementor locomotion of o . If the cue of motion parallax is so defined as to require activehead movement or locomotion, proprioceptive and vestibular stimulation is also present. This definition is unjustified , since passive locomotion
in trains and airplanesshould be admitted as circumstanceswhen motion parallax occurs. Certain patterns of motion in the field of view of 0 do induceimpressionsof being moved through spaceif we acceptas evidence the illusions of locomotion obtained in viewing a panoramic motion pic-
ture, or in a training device for simulating aerial flight . Perception of absolutedistanceand of relativedepth The apparent displacements of the sensations of objects are said to be cues for perceptions of
their depth. What kind of depth? The question arises whether their distancesfrom the perceivercan be judged or whether they will only appear to be separatedin the third dimension. Helmholtz, in his description of motion parallax, assertedboth hypotheses. On one page he describedthe appearanceof objects IIgliding past us" as we walk through the countryside, and asserted that IIevidently under these circumstances the apparent
angular velocities of objects in the field of view will be inversely proportional to their real distancesaway; and consequentlysafe conclusionscan be drawn as to the real distance of the body from its apparent angular velocity " (Helmholtz , 1925, p. 295). On the next page he described the
appearanceof an indistinguishabletangle of foliage and branchesin a thick woods as a man stands motionless , but noted that " the moment he begins
to move forward everything disentanglesitself and immediatelyhe gets an apperception of the material contents of the woods and their relations to
eachother in space, just as if he were looking at a good stereoscopicview of it " (Helmholtz
, 1925 p . 296 ).
In the first quotation Helmholtz says that angular velocity is a cue for the perception of absolute distance. In the second, he suggests that a
differencein angular velocity is a cue for the perception of separationin depth, or of relative distanceonly. Thesetwo hypothesesare by no means the same, and they should be consideredseparatelyand tested separately.
We will be concerned hereprimarilywith the second .
MotionParallaxin PerceivedDepth
267
Two-velocity motion parallax andflow-velocity motion parallax Although
motion parallax hasbeensaidto applyto thewholearrayofobjects inan
environment anda largearrayofapparent motions inthefield, theexperi-
mentsperformed haveinthepastbeenconfined to twoobjects andtwo
velocities in a restricted field of view.
Bourdon (1902) reported experiments inwhich 0 looked withoneeye
at a pairofluminous spotsina darkcorridor. Thesources wereat different distances butthespotswereofthesameangular size.Whentheheadwas
fixed withabiting-board, 0 couldnotjudge atallaccurately which light
wasthenearer,butwiththeslightestmovement of theheadfromsideto sideit waseasyto judgetherelative depthofthetwo.Buttheabsolute
distanceof neitherlightwasdetectable.
Tschermak-Seysenegg (1939) improved onthisarrangement withwhat hecalled a parallactoscope, byanalogy withthestereoscope. Hedefined motion parallax as arising frommovement eitherofa groupof visible
objects ontheonehand, ortheposition of0s eyeontheother, emphasiz-
ingtherelativity of the situation. But,he studiedonlythe detection of
depthoftwoobjects withvoluntary headmovement. Hisapparatus wasa binocular depthperception. It permitted 0 to moveoneeyefromsideto side,witha sliding headrest, soasto obtain successive impressions equivalent tothesimultaneous impressions obtained withbotheyesopen. The modification ofthefamiliar two-pins setupusedto obtainthethreshold for
averageerrorof equatingthedistanceoftheverticalwireswassmallunder
these conditions, although notassmall astheerrorwithbotheyesopen. When onlyoneeyewithafixed headwasused, theerror wasverylarge. Graham, Baker, Hecht,andLloyd(1948; seealsoGraham, 1951)obtained thethreshold forseparation indepth oftwoneedles pointing toward
oneanother, as seenon a uniform fieldthrougha window. Theneedles
moved fromsideto sideona common carriage. Theyappeared to be
aligned at thecenteroftheirmotioncycleandoffsetat theextremes ofthe
cycle unless theadjustable needle hadbeensetintothesame frontal plane asthefixed needle. Graham thuseliminated forthefirsttimeinthistype ofexperiment theadditional sensory information produced byvoluntary head movement. Moreexactly, whatGrahamobtained wasthejustnoticeable difference
between twoangular velocities ina field ofview, under probably optimal conditions. Thethreshold wasextremely lowabout30sec.ofarcper second oftime. Itisnotable thatthereports ofwhatOsperceived, how-
ever,werenot unanimous. Somesawthe separation in depthas such; othersperceived eitherthe difference in velocityof the twoneedlesor
noticed thechange ofalignment oroffset oftheneedles. Although the
latterimpressions maybecuesfortheformer, theexperiment wasnotcon-
cerned withtheeffectiveness ofsuch cuesforproducing depth impressions.
268
F.J. Gibson,J.J. Gibson,0. W.Smith,& H.Flock
Somewhat: later,a caseof motionparallaxdifferent fromthetwo-velocity casewasdefinedmathematically by Gibson,Olum,and Rosenblatt(1955).
Thisisthearrayofangular velocities oftheoptical elements projected from a surface toa moving stationpoint.Thereisa flowofvelocities ratherthan
a pairofvelocities in suchanarray,andthephenomenon wasnamed motionperspective to distinguish it frommotion parallax asit hadbeen studied up to that time.
Forthestudyoftheperceptions induced byflow-velocity ina fieldof
view,including gradients ofvelocity, skewmotions, andtransformations, a different sortofapparatus isrequired fromthatpreviously employed. The two-velocity experimenters useda pairofrealobjectsat realdistances to produce theoptical motions. Morefreedom isachieved byusinga projectionscreenor someotheropticalmeansto producethem.Theexperiments to be describedusedshadowson a translucent: screen.Accommodation is thereby controlled.
Severalexploratory experiments havebeenpublished on flowingmotion.Theyareof varioustypes,andtheyhavebeenproduced in various
ways. 1.J.Gibson andCarel (1952) attempted toinduce theperception of a receding surface ina darkroom witha bankofluminous pointswhich
carried a gradient ofvelocities. Thisstimulus failedtoarousetheperception of a surface,however,and the depthjudgmentswereambiguous. 0. W. Smithand P. C. Smith(1957)investigatedthe perceptionof convexityor
curvatureof a texturedsurfacewithvariouscombinations of depthcues,
including theflow-velocity typeof motionparallax. Although motionin
the fieldcontributedto the judgmentof convexity,in no casedidmotion causea surfaceotherwise judgedas flatto be judgedas curved.Hochberg and0. W. Smith(1955)studiedthe perception of depthinducedby the
centrifugal flowof luminous patternelements in thedark,theexpansion phenomenon. J.J.Gibson andE.J.Gibson (1957) investigated theperceptionoftherigidrotationofanapparent surface elicited by thecontinuous perspective transformation ofregular andirregular patterns orforms. Theseexperiments differed in thestructure of theopticarrayusedto carry themotionin question, andtheyalsodiffered in thedegreeto whichperceptions ofspacewerearoused. Theyledto thechoice ofthe kindof randomtextureemployedin the presentexperiments, whichis
intendedto yieldthe experienceof a planesurface. What is now neededis an experimentalcomparisonof the judgments obtainedwith twovelocitiesin a fieldof viewand thoseobtainedwith
manyvelocities ina fieldofview.Although noclearlinecanbedrawn
betweenthem,the two-velocity type of motionparallaxappliesto the
problem ofperceiving a groupofobjects inotherwise empty space, while theflow-velocity typeof motionparallax applies to theperceiving of a background surface suchasa wall(orsubstratum). Thesearenotthesame
MotionParallax in Perceived Depth 269 problemfor perceptioneven though it may be difficult to distinguish sharplybetweentheir respectivekindsof stimulation . Opticalgeometryof motionparallax The environmentalsituationwhich leadsto an array of differentmotionsin a visualfield shouldbe defined more carefully. This is the optical geometryof motion parallax , as distinguishedfrom the visualappearance of motion parallax . Graham(1951, pp. 878ff.) hasgiventhe geometryof certainspecialcasesof this situation. J. J. Gibsonet al. (1955) haveanalysedthe caseof an extendedsurfacesuch asthe ground. What will be discussed hereis the caseof an environment of discreteobjects. Whenlight raysfrom permanent objectsof anenvironmentconvergeto a point, theyconstitutewhatmaybecalledanopticarray, andtheelements of this arrayconstitutea pattern. An eye or a cameraat the stationpoint can registerthis pattern of luminouselements . If the point moves, the patternis alteredin a way whichdependson both the displacement of the point andthelayoutof theobjects.How theeyerespondsto this alteration of patternis our problem. Thefirst questionis how to specifymathematically thechangeof pattern in a way thatis relevantfor vision. By choosinga coordinatesystemfor the array, one can specify the absoluteposition of eachelementand the displacement of eachelementper unit of time, that is, its absoluteangular velocity. It would then be true in a certainsense , as Helmholtz(1925, p. 295) said, that "the apparentangularvelocitiesof objectsin the field of view will be inverselyproportionalto their realdistances away." Butmore exactly, it wouldbetrueonly if thelinearvelocityof the stationpoint were constant(J. J. Gibsonet al., 1955). A given angularvelocity is a cuefor distance , or permitsa "safeconclusion " aboutdistance , only if the speed anddirectionof one's locomotionis known. Thetroublewith positionsandangularvelocitiesof elements in a fieldis the difficulty of understandinghow an eye can registerthem. As the Gestalttheoristshaveemphasized , what the eye seemsto pick up is the mutualseparationof elements , their pattern, ratherthan their positionor directions . And, accordingly , it is easierto supposethat the eye responds to changesof separation , or changeof pattern, rather than to absolute displacements or velocities . Helmholtzmight better haveassertedthat a differencebetweenthe angularvelocitiesof two elementsin the field will be directlyrelatedto the differencein distancebetweenthe corresponding objectsin space . Suchrelativevelocitiesinvolve a transformationof pattern, and this may be what the eye is primarily sensitiveto. It is not immediatelyevidentwhat the bestmethodis for specifyingthe information aboutobjectsin an arrayof light projectedto an eye.
270
E.J. Gibson,J.J. Gibson,0. W.Smith,& H.Flock
Butonefactshouldbe clear.Onlyif thereis an eyeat the pointof
projection andonlyif it is sensitive to themotions in theopticarray, relativeor absolute,doesa psychological questionarise.Willthe possessor
oftheeyeseemerely thechange ofpattern ofthearray? Orwillhesee moving objects inthefieldofview? Orwillheseestationary objects at differentdistances? In orderto showthatmotionparallaxis effective for
theperception ofdepthitmustbedemonstrated experimentally thatdifferentialmotionsinanarrayoflightto aneyewillyielddifferential judgments
ofdepth. Andthearrayshould besuchthatwhenthemotion iseliminated thejudgments ofdepthwillcease, foronlythenwillmotion parallax have been isolated from other cues for depth.
Experiment I: MotionParallaxwithTwoVelocities Problem and Method
Thetwo-velocity experiments wererepeated with(a)twospotsina fieldto carry themotions, and(b)twosuperimposed textures filling thefieldto carrythem.In bothcasesthevelocity difference wastakento be theessential cueforpossible judgments of depth,nottheabsolute velocities. In thisexperiment, reportswere
obtained fora largevelocity difference, a small velocity difference, andnovelocity difference, that is,a motionless field.Thelastwasa control.
Apparatus andstimuliThelightentering 0s eyecamefromthetranslucent screen
ofapointsource shadow-projector (J.J.Gibson, 1951; J.J.Gibson &E.J.Gibson, 1957).Hesawonlya luminous rectangular fieldin whichdarkcircles or textures
could bemadetoappear andtomove. These wereactually theshadows ofopaque substances attachedto a transparent mountbehindthe screen.Thiswasa large
sheetofglassorplastic whose edgeswerenevervisible. Differential translatory velocity oftheshadows wasproduced withtwomounts, onebehind theother, whichcouldbe madeto moveparallel to thescreenon a common carriage. The
arrayoflightto theeyewassimply thereverse ofthearrayprojected to the screen, sincetheeyeandthepointsourceweresymmetrically locatedequidistant fromthescreen (Fig.16.1). Thewindow was32.2x 36 cm.at a distance of 126cm.fromtheeye,subtending anarray14.5highand16.20 wide.Thewindow
was viewedthroughan apertureby a seated0.
Thecarriage which borethetwomounts rolled silently ontracks andcould be
pulled from sidetosidethrough anexcursion of45cm.Itwasoperated byhand
toproduce amotion cycle inabout8sec.Asmall shutter closetothepointsource enabled Etoeliminate theshadow between trials,leaving thescreenilluminated by diffuse light.
Thetwoadjacent spotsin thefieldwereproduced by attaching smallpaper
circlesto eachmount,at differentelevationsso that theirshadowsdid not pass
through oneanother astheymoved across thefield. Thefaster spotwasabove the
slower spot.Thediameter ofbothwas5.2,onepapercircle beingcompensated in size to match the shadow of the other.
MotionParallax in Perceived Depth 271 VARIABLE
STANDARD
MOUNT
MOUNT
APERTURE
EYE
POINT
TRANSLUCENT
WINDOW
Figure 16.1
Theshadow projector viewed from above. Ina unitoftime, theshadow ofa spotatthe center ofthestandard mount sweeps through a certain angle andthatofa corresponding spotonthevariable mountsweeps through a lesserangle,asshown. Thetwomounts roll onthesame carriage. Iftheyareclose together, there isnodifference inangular velocity,
butasthevariable mountis positioned farther fromthepointsource andcloser to the screen, theangular velocity ofitsshadow decreases. Withthisapparatus, itcandecrease to aboutonehalfoftheangular velocity ofthestandard. Dytrigonometry, theratioofthe lesser(V)to thegreater(S)angular velocity is equalto theinverse of theratioof the distances oftheirrespective mounts from thepoint source. Inthediagram above, itisabout 0.7.
Thesuperimposed random textures were produced byatechnique ofsprinkling talcum powder overthesurfaces ofthetwotransparent mounts. Thisyields an
opticaltexturewithindefinite contours andindefinite elements. Whenthetwo weresuperimposed butmotionless, theyconstituted a singletexture withnocue
forsuperposition, andgavetheappearance ofa single surface, something likethat of a cloud. This apparent surface filled the whole window and appeared atan indefinite distance from 0. Asnoted, thetwoangular velocities assuchwerenotuniform, decreasing to
zeroat eitherendofa motion cycle, andchanging direction alternately. Minor variations invelocity alsooccurred asa consequence ofmoving thecarriage by
hand. Theindependent variable ofthisexperiment wasthedifference invelocity between thetwoshadows. It wasexpressed as theratioof theslower (the variable) velocity tothatofthefaster(thestandard) velocity, orV/S.
Procedure Each 0 wasseated attheapparatus, asked toapply hispreferred eyeto
theaperture, andinstructed simply to describewhathesawinthewindow.He wasfirstpresented witha motionless fieldforas longas he neededto makea
report, which wasrecorded. Hewasthenpresented withcontinuous cycles of motion atthemaximum velocity difference (V/S= .51)untilhisreportwascompleted. Finally hewasgiventheminimum velocity-difference (V/S= .97).TheE
made nocomment atanytime,since wholly spontaneous reports weredesired.
272
E.1.Gibson,J. J. Gibson,0. W.Smith,& H. Flock
Theorder ofpresentation wasintended tominimize theeffect ofsuggestion onthe perceiving of depth.
Agroup of26Oswentthrough thisprocedure withthespotfield andanother groupof46withthetextured field. Formal judgments andanswers toquestions wereobtained afterwards fromsomeOs,whichwillbedescribed whenrelevant. Theywererequested inthetermsusedspontaneously bythe0. Results
ThewordsusedbytheOsto describe whattheysawvariedwidely, and
the effortto identifythingswasreminiscent of descriptions of cognitive inference (Vernon, 1957).Butthe reportscouldlaterbe classified easily
withrespect to depthordistance. Themotionless textured fieldwasunanimously reported tobea single surface without anydifference indepth. Themotionless spots,however, werereported at different distances by4
of the26 Os.Thespots,therefore, didnotwhollysatisfytherequirement
thatimpressions ofdepth beabsent intheabsence ofmotion, although the combined textures did.
A largevelocity difference (.51)forthetextured fieldalways gavea perception oftwosurfaces separated indepth, asevidenced bythereports of all46 Os.Forthe spotfield,thereportswerenot unanimous, but22 out of 26 Os diddescribea difference in depthof the two objects.
Thesmallvelocitydifference (.97)wasevidently closeto thethreshold.
Noneof the Osreported twoseparated surfaces forthetextured field, andonly7 outof26reported different distances forthespots. The directionof the differencein depth reportedwas not unanimous
foreitherthespotsor thetextures. Insofar as two-velocity parallax is a
reliable andeffective indicator ofrelativedepth,thefastervelocityshould
correspond to thenearer object orsurface. But7 outof26Ossawthe slower spotasthenearer object instead ofthereverse, and10outof46Os sawtheslowertextureas thenearersurface. Somedegreeof ambiguity as
to thedepth-difference is alsoindicated by thefactthat7 Osreported spontaneous reversals ofthefront-back relationship between thetwosurfaces at one time or another.
Theamountofseparation indepthbetweenthetwoobjectsor thetwo surfaces wasestimated on formalrequestby a sampleof theseOs,after
themainprocedure. Theestimates werehighlyvariable. Forthe spots
theyranged fromzeroto 5 ft.Forthetextures theyranged from2 in.
to 3 ft. SomeOswereunwilling to judge,saying,It dependson what it is, or It couldbeinfinity.Evidently theimpression ofhowfarapart theseentitieswerewasindefinite, as alsowastheimpression of howfar away they were.
Figure 16.2represents anidealtheoretical possibility ofwhattheOs mighthaveseeninthisexperiment, butit cannot beasserted thatthis,
Motion Parallaxin PerceivedDepth -
-
-
-
-
-
0
-
-
-
-
-
-
-
-
273
-
-
-
-
-
1_-
1_- - -
- -- - - - , - - - - - - - - - - - -
is whattheydidsee . Thereportsindicated thattheyperceived two things of somekind in somekind of spacebehindthe screen , but neitherthe direction in the third dimension was definite . nor the amountof theirseparation
Discussion Thesignificant result of thistwo-velocity experiment isnotsomuch the effect ofmotion ondepth perception asitseffect inseparating onesurface intotwo.Withthetextures , alIOssawasingle frontal surface when there wasnodifferential motion , andalIOssawtwofrontal surfaces when there wasasufficient degree ofdifferential motion . Thisseparation isnotwhatis ordinarily meant by depth , since it wasnotalways clearwhichsurface appeared infrontandwhich behind . Thephenomenon is similar to theIIdisentanglement " of foliage and branches whichHelmholtz notedwhenhebegan to movein thethick woods . Butthisis notthesame ashisIIapperception of depth ." The separation isprobably related toWertheimer 's(1923 ) demonstration thata group ofspots interspersed among others willbeunified bywhathecalled
274
E.1.Gibson,J.1.Gibson,0. W.Smith,& H. Flock
their common fate if they movedtogether.Other conditionsfor the
seeingofonethinginfrontofanother, fortransparency andsuperposition,
havebeendiscussedby Koffka(1935).Althoughthe phenomenonmaynot
seemrelevantforsomekindsofspaceperception it iscertainly relevantfor object perception.
Howcanthisresultbe explained? Insteadof appealing to a processof
organization, or a lawof commonfate, onemightlookforits basis
in thegeometry of the opticalstimulus. Geometry distinguishes between
(a)perspective transformation offorms, (b)topological transformations of forms,and(c)disruptions. Thesecorrespond roughlywiththedistinctions in
physics between rigidmotions ofbodies, elastic motions ofbodies, andthe motions ofbreaking, tearing, orsplitting. Inthisexperiment, themotionof one set of textural elementsrelativeto the other was a disruption,geomet-
ricallyspeaking. Whensufficiently differentvelocities wereimposedon them,the adjacent orderof elements in the textureswas destroyed.More
exactly, therewasa permutation ofthisorder.It wasa particular sortof permutation, to be sure,foreachof two setsof elements retainedan adjacent order,butthedisruption oforderasbetween thesesetsbrokethe original continuity. Andthisproduced theperception ofdifferent surfaces withseparation between. Thedetection bytheeyeofcontinuity orsolidity as compared withdiscontinuity, disruption, or separation, is probably a fundamentalkindof perception.The continuityof a singlesurfacein two
dimensions maybegivenby a staticopticaltexture.Butthecontinuity of a solidobjectinthreedimensions probably depends onthekindofoptical motionpresented to theeye.Perhaps it wasthislackof solidcontinuity or rigidconnectedness betweenthenearerandfarthersurface in ourexperimentwhichprevented theidealpossibility represented inFig.16.2 from having been realized.
Theearlierinvestigators ofmotionparallax werewillingto assumethat
aneyewassensitive to thestimulus ofmotion, buttheydidnotseemto realizethat differentialmotionnecessarilyentailsa changeofpattern.In our
experiment thereweremotions oftheelements relative tothewindow but there were alsomotionsof one set of elementsrelativeto the other.For
example, whenbothsetsofelements weremovingto theleft,relative to
the window,and one moved fasterthan the other. the slowerwas moving
to theright,relativeto the faster.Spontaneous reportsof thisappearance
weregivenby severalOs.Whyshouldnot a differential velocity be perceived justas directlyas the twocomponent velocities? Whentwo moving elements arefarapartinthefield,onemightsuppose theslower and the fastervelocitymighthaveto be comparedin orderto detecta difference betweenthem.Butwhenthe elementsare adjacentin the field the difference is givenby the changeof pattern.Permutation of orderis
onetypeofchange ofpattern. Inorderto studythesensitivity oftheeye
275
Motion Parallaxin PerceivedDepth
toform ,to change of form ,and forms ofchange ofform ,ataxonom of these variables isdesirable . tothe Experiment
II
Problem
If
and
a
flow
two
: Motion
- velocity
field field
was
modified
velocities
of
bottom
of
the
was
If
this
or
pattern
in
producing
motionless was
,
the
so
a
Flow
texture then
,
as
is , a
of
Velocities
an
of research
angle
size
and
of
a
which
, the a
texture
to
fast
field
by
source 450
( Fig
yield
elements
slanted
surface
.3 ) .
By
even
the
principle away
texture
the , were
in
been
when
).
not
'" '" POINT \.
SOURCE
'" '"
-
TRANSLUCENT WINDOW
~
CARRIAGE
TRACKS
WITH
a
of from
a
grid
was But
sharp
~
\.
on
effective
MOUNT
EYE
.
texture
1955
-
the
mount
slant
spaced
Gibson
hor
motionless
. This
have
a
to
paint
.
(J. J. the
be
screen
would
demonstrated
the top
spattering
apparent
will
already
the
the
16
,
which
also
regularly
spacing
composed
in
and
an
depth
apparatus
from
could
produced
of
should
has
, the
point
eye
composed of
eye slow
were the
the
the
before
screen
at
gradients
- transitions
to
of
comparison
from
. As
the
at
impression
previous
present
gradient
perceptions
this
varied
screen
been
consistent make
between
the
had
to
to
on
gradient
the the
brightness
as
interposed
the
arouse
order
elements
shadows
,
,
not
In
the
toward
reversal screen .
s07
, that
sheet
slanted
mirror the
field
The
transparent
does
do
izontal
Apparatus
with
Method
- velocity
described
Parallax
as ,
it the
276
F.J.Gibson,J.J.Gibson,0. W.Smith,&H.Flock
elementshad indefinitesizesand shapes,and the staticgradients,if present,were
not effective in producing a perception of slant.Thistexturewasnevertheless sufficient to producethe perceptionof a continuoussurface.
Asbefore,0 lookedthroughanaperture whichprevented hisseeingeitherthe
edgesofthewindow oranyotherpartoftheapparatus. Thefieldwas82 wide by52 high.Theeyeandthepointsource wereeach75cm.fromthewindow. ProcedureAllOs werenaive.Eachwastoldthatwhenhe appliedhiseye to the
aperture hewould seea grayfieldofview, andwasasked toreportwhathesaw. Theyweredivided intotwogroups, oneofwhich waspresented withthemoving texturewithouteverseeingit static(GroupI),and the otherwiththe moving textureafterhavingseenit static(GroupII).Theirspontaneous descriptions were recordedandlaterclassified. Questionswereaskedonlyafterthesereports,andin the same terminology used by 0. Results
Nineteenout of 21 Os(GroupI)reporteda rigidmovingplanesurfaceof somekindslantingawayfromthemat thetopof thefield.Theinclination of the surfacewasestimatedwithoutdifficulty, and theyjudgedit to be receding upward. Ofthe2 remaining Os,onesreportwasuninterpretable, and the others was of a surfaceperpendicular to his lineof sight. With the statictexture,25 Os out of 28 (GroupII)reportedsomething whichcouldbe classedas surface-like, but whichwas in no caseslanted
backward intospace.Oftheremaining 3 Os,2 sawa surface whoselower
partwasperpendicular butwhoseupperpartwasslanted back,andone sawthe surfaceslantingbackintospace.Whenthese28 Os werelater
presented withthemoving field, 26sawitunambiguously tobea receding rigidsurfaceat a fixedinclination. Of the remaining two,one sawit perpendicular to thelineofsight;theothersreportwasuninterpretable. Estimatesof the slant of the apparentmovingsurfacewere obtained fromeach0 in termsof degreesof inclination backwardfromthe frontal
plane. ForGroup I,the19judgments varied from20 to60 withamedian of 40. ForGroupII,the 26judgments variedfrom12 to 55 witha medianat 37 . Thetheoretical valuebasedon thegradientof velocities alonewouldbe 45. The mediansshow a constanterror of underestima-
tion.Thesurfaceneverappearedto be at 90, thatis,it neverlookedlikea groundon which0 mightbe standing.
Estimates wererequested of howfarawaytheapparent movingsurface
seemedto be from0, but thesejudgments,in contrastto those of slant,
couldnot as easilybe made.Forthe totalof 49 Os,theyvariedfrom 3 in. to 5 miles.Someof these Os reportedthat it was possibleto see
themselves movingwithrespectto the surfaceinsteadof the surface movingwith respectto them.
Motion
Parallax
in
Perceived
Depth
277
Discussion
These
results
velocities receding of
a
in
depth
respects put
it
skew
of
in
a certain
rigid
surface
another
that way
pattern were first
the
kind
of
were
for
, ( a ) the
ever
, was
nected
recession for
the
Experiment III Under Various
In
and
Exp
.
I,
has
visual
been
velocities
would
be
the
there about
the
to
limits
about
length
the
greatest
ments
of
a
experience the
first
absent
in .
, but
distance
in
reluctance
II depth
of
in
the
"
two
and
, as
more so
kinds
third
a
this they
of
than
far spatial
a level
. This
ground
- Pairs
-
( b ) the , how
may
was
and
percep and
situation
distance
one
described
dimension . Neither
of
this
separation
of
Exp one asked
of
of
in
of
be
-
con
induced
Judgments
of , in
occur ,
or
-
.
of
Depth
between
, would
be
. This
three
degrees
a
not
alter
the
the
Helmholtz
,
interpret
to
perceive
suggestion with
that
instructions
perceptions
tested
in
a
this
not
learned
,
was
of did
, the
objects to
prediction
of of
between
spirit
theory
the
expected
the
because
this
perception
explanation
continuity
had
. On
the
. An
however
distance distance
depth
induce
absence
not
depth
Exp
and
. III , using
. I. of to
of
reproduce
between
the
an
nearer
described what
was for
both
and , was
see , "
( V IS that
of the
=
or
adjustable
, initially
you
difference
suggestion
on
, the V / S ratio with it .
apparatus separation
the
did
in
not
consistency
explanation
previously " Describe
did
great
terms
of
velocity
.97 ) . The
amount
velocity
difference
element
, as
, the =
Os
Velocity
perception
separation
request
( V IS
in
an
were
AlIOs
stimulus
of of
introduced
was of
situations
of
any
possible
given
the
ence
with
of velocity difference could be correlated
Procedure to
optical
spots
, and
The degree judgments
gradient property
was
with
dimension
Between
judgments
were
depth
1 m . in
tions
of
and as
third
perception
as a difference
always
textures
of
judgments
in
as
consistent
Naive
the
no
expected
velocity
the
lead
the
motions
was
in absolute
suggested
that
two
separation
distance
. Another
differential
like
elements
made
kind
naive
of
: Correspondence Instructions
difference
in
result two
the
was motion
experimental
by
that
with
. Judgments
were
the
Method
a
difference
fact
differential textural
one
two
judgments
of
with
Problem
than
. The
sufficient
is
combined
present
and
continuous . This
---
phenomenon
phenomenon
is
a slant
experiment
of was
more
parallax
sufficient
tion
. It second
variable . -
evidently
with of
of
elements
highly exoeriment
motion
direction . The
continuity
. is
parallax judgments
I permutation
of
1 There
motion
consistent
except
the
experiment were in
that
induce
continuous
most To
indicate
does
the
farther
the a
textures
and
surface
varied
to
. I . Each
the
scale
apparent
spontaneous
response
.5 I ) , and
Exp
information
unmarked
systematically
made in
verbal but
.
that
observa
the
-
motionless
least
velocity
then
made
spots
so
. The
20
differ
-
judg
-
variable
278
E. J. Gibson, J. J. Gibson, O. W . Smith, & H. Flock
transparentmount was so set as to produce 10 velocity ratios of .51, .54, .57, .61, .65, .70, .76, .83, .90, and .97. Eachratio was presentedtwice, in random order, in the texture seriesand the spot series. Half the Os began with one seriesand half with th_e other. An 0 was assignedto one of three groups, and then instructed as follows :
Group I (Leastinformation):
The 0 was shown the sliding scalebeside his chair
and was told that it could be used to indicate
the distance between
the nearer and
farther of the two (surfaces, spots). If 0 had used other terms instead of "nearer"
and "farther," these terms were employed. He was then told that he would be shown a number of different settings of the apparatus, and that each time he would
be asked to make a judgment (degree of separation, distance between, etc.). No othPT information was given. The 0 was encouragedto report as he went along ~----- -------but no commentswere made on his performance. There were 16 as. Group II (Maximum and minimum): After the adjustablescalehad been demon'-'
-
strated , these Os were given another demonstration of the greatest and the least velocity -difference and were told which one was the maximum and which the
minimum. The procedure thereafter was the same as for Group I. There were 17 as .
Group III (Most information):
After the preliminaries described, these as were
told : "These are shadows of two (surfaces, objects ) which are actually at different distances from you . One is farther away from you than the other . This is what they look like at their maximum separation , which is 18 in . (The 0 was shown 18 in . on the adjustable scale.) This is what they look like at their minimum separation ,
which is 1 in. (The 0 was shown ! in. on the scale.) Each time I will ask you to estimate how far apart the (surfaces , objects) making the shadows are." The procedurewas thereafter the sameas for the other groups. There were 20 Os. When each 0 had finished his judgments he was questionedby E as to how he had made them unless he had already made this clear. The questions were : How
did you make your judgments 7 Did you go by appearanceof depth, or did you try to use some other
cue ? If some other
cue , what
was it ? Did you ever see the
front and back (surfaces , spots) changeplacesor fluctuate? Did the two (surfaces, spots) ever appearto be connected, like parts of a rigid object?
Results For each 0 a rank order correlatiort was run between his 20 judgments of separationand the correspondingvelocity-ratios. Table 16.1 summarizes the median for the three kinds of instruction and for the two ---- --------- coefficients ~ kinds of apparent objects, surfacesand spots. They range from .51 to .84. Eighty-seven of the 106 correlations were significant at the 5% level (r , > .44). , The data thus demonstratesome correlation between amount of differential velocity and degree of depth judged. The spread of the individualcorrelationswas very great, however, ranging from - .52 to + .99. Group III, which was given the most information, had higher correlations (medians of .83 and .84) than the other two groups. If individual correlations are considered, 19 out of 20 were significant for both the
MotionParallaxin PerceivedDepth
279
Table 16.1
Medians andRanges ofCorrelations Between Velocity Ratios andJudgm entsofDepth Measure
GroupI
GroupH
GroupIII
Stimuli (N= 16)
(N= 17)
(N= 20)
Median r
Surfaces + .72
+ .67
+ .83
Spots
+51
+84
Range
Surfaces .23to + .95 + .34to + .85
.18to + .96
Spots
.45to + .99
+57
.52to + .94
Number rs significantSurfaces 12/16 at