You are on page 1of 18

The Role of Private Events in the Interpretation of Complex Behavior

Author(s): David C. Palmer

Source: Behavior and Philosophy, Vol. 37 (2009), pp. 3-19
Published by: Cambridge Center for Behavioral Studies (CCBS)
Stable URL: .
Accessed: 21/06/2014 17:45

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact

Cambridge Center for Behavioral Studies (CCBS) is collaborating with JSTOR to digitize, preserve and extend
access to Behavior and Philosophy.

This content downloaded from on Sat, 21 Jun 2014 17:45:04 PM

All use subject to JSTOR Terms and Conditions

The Role of Private Events in the Interpretation

of Complex Behavior

David C. Palmer

ABSTRACT: Like most othersciences, behavioranalysis adopts an assumptionof

uniformity,namelythatprinciples discoveredundercontrolled conditionsapplyoutsidethe
laboratoryas well. Since theboundary between publicand privatedependson thevantage
pointof the observer,observability is not an inherent propertyof behavior.Fromthis
perspective,privateeventsare assumedto enterintothesame orderlyrelationsas public
behavior,and thedistinction betweenpublicand privateeventsis merelya practicalone.
Privateeventsplayno rolein theexperimental analysisof behavior,buttheypermitus to
makesenseof manycommonplacephenomenawherecontrolled observationis impossible
but unsystematic data are available. Such interpretive exercises serve both to guide
researchandto displaceoccultexplanations.
Key words: cognition, definition of behavior, interpretation, private events,

The term "private events" was used by Skinner to refer to behavioral

phenomena that are observable, if at all, only to the behaving organism,events
such as subvocal speech, joint pain, muscle fatigue, mnemonic behavior, and
visualization. Because interobserveragreementfor such events is impossible by
definition,they pose a challenge for an experimental science, but they are
neverthelessof interest.Not only are they are commonly salient to the behaving
organism,they appear to contributeto the proximal control of other behavioral
events,and forthisreason a considerationof theirstatusis morethanan idle parlor
game. Privateevents are partof the subject matterof a behavioral science, and its
conceptual scheme must find a place for them. It is the purpose of this paper to
offera concise statementof this scheme. Despite Skinner's extensivediscussions
of the topic (e.g., Skinner 1945/1999, 1953, 1963, 1974), behavior analysts
disagree about both the statusof privateevents and the value of consideringthem
is in order.
(Palmer et. al., 2004). A restatement
The behavioral approach to complex human behavior is a kind of atomic
theory.Phenomena commonlycalled "cognitive,"such as recall, problemsolving,

AUTHOR'S NOTE: This articleis a modifiedversionof a paperpublishedin K. A. Lattal

& P. N. Chase (Eds.), Behaviortheoryand philosophy(pp. 167-185).New York:Kluwer
Academic Press and is reprintedby permissionof the publisher.Please address
correspondence to theauthorat:

This content downloaded from on Sat, 21 Jun 2014 17:45:04 PM

All use subject to JSTOR Terms and Conditions

composition, planning, and imagining, are typically not functional units

themselves but compounds of elementaryor atomic operants. For example, we
solve a long division problem by a series of one-digit multiplication and
subtractioncalculations along with various orderingoperations. The compound
usually serves some adaptive purpose, and over repeated instances can itself
emerge as a kind of behavioral molecule- we can dash off the solution to a
problem we have seen before but more commonlysuch compounds are unique;
we seldom repeatourselves exactly when we solve problems,recall an episode, or
plan our day. In any case, it is the firstinstance of a phenomenon that poses a
special challenge to science. From a behavioral perspective,such phenomena are
best analyzed at thelevel of theelementaryoperant,appealing only to principlesof
behavior thathave emerged fromexperimentalscience. Our task is to show how
such behavioral atoms can combine to produce complex human behavior.
Although some examples, such as solving long division problems, may be
formulaic,others,such as recallingwhat you ate fordinnerSunday night,are not.
The challenge is formidable.The experimentalanalysis of even a single operant
requiresconsiderableeffort,and the studyof the relationsbetween two competing
operantshas kept researchersbusy fordecades. If most human behavior is indeed
an interactingmixtureof many elementaryresponses, an experimentalanalysis
will be a formidablechallenge indeed.
Nevertheless,the topic is of centralimportanceto an understandingof human
behavior,and howeverincompleteour account,behavioranalysis mustadvance its
case, for alternative approaches are influential. I begin by identifyingan
assumptionthatdistinguishesthe behavioral approach fromcompetingparadigms.
I then submit several examples of human behavior that seem to call for special
treatment:at the level of our observations,behaviordoes not seem to be relatedin
an orderlyway to environmentalantecedentsor consequences but seems to emerge
fromwithinthe individual.I suggestthatthislack of orderinevitablyarises when
thereare gaps in our data, and thatforcognitivebehaviorsuch gaps are common. I
thenpresentan example of a behavioralinterpretation in which orderis restoredto
the data by referringto plausible unobservedvariables (i.e., privateevents) thatfill
the gaps. Next I discuss the role of such tentativeinterpretations in science and
claim that they are not peripheralbut central to our understandingof nature. I
conclude by characterizing a behavioral program for analyzing cognitive
phenomena and argue that,in contrastto the fruitsof otherparadigms,the results
of such a programoffera genuineexplanationforsuch phenomena.

The Assumption of Uniformity

Because of the complexityof human behavior and the difficulty of collecting

reliable data, any account of cognition must be tentative and cautious, a
circumstancethathas encouraged a profusionof competingmodels and theories.
To regardthemas equally plausible would lead to paralysisor aimlessness.When I
was a boy, I sometimes wondered if therewere a black void behind my back. I
would tryto test this proposal by turningaround suddenly,but I was never fast

This content downloaded from on Sat, 21 Jun 2014 17:45:04 PM

All use subject to JSTOR Terms and Conditions
Prvate Events and Complex Behavior

enough; the world would be set perfectlyin place just as if it had always been
there,while behind my head the formerscene had been replaced instantlyby the
black void. I was never able to refutethiscurious hypothesis,but its shortcomings
soon became obvious. Not only did it requireconsiderable machineryforwhich I
had no evidence, I realized thattherewere innumerablealternativehypothesesof
equal merit,and no evidence could ever weigh in favorof one over another,forthe
domain was, by definition,unobservable: Perhaps the world behind me was
populated by unicorns,griffins,and otherfabulous creatures,or by a phalanx of
angry elementaryschool teachers. The conventional assumption that the world
behindmyback was of the same stuffas the world in frontof myeyes was simple,
adequate, and requiredno imaginarystage managers.
The behavioral approach to cognition assumes an analogous kind of
uniformity:Behavioral phenomena do not obey one set of principles when we
observe themand a second set when we do not. The topic of cognitionembraces
phenomenathatare challengingto studybecause some of themcannot ordinarily
be directlymeasured or even observed, and it is temptingto populate this hidden
domain with conceptual unicorns and griffinsthat have no counterpartamong
observable phenomena. However, the behaviorist doggedly assumes that this
landscape is composed merely of behavioral phenomena governed by the same
principles that govern observable behavior. Thus cognitive phenomena are
interpretedas complexes of elementarydiscriminatedoperants, some of which
may be covert. This assumptionof uniformity sets the behavioral approach apart
fromall other paradigms in the domain of cognition. Like every assumption in
science, it might be wrong, in which case all that follows from it will be of
uncertainvalue. However, theassumptionis bothparsimoniousand practical:Until
it is shown to be inadequate,it forestallsthe need to entertaincountlessalternative
mechanismsthatone mightinvent.
A glance at the historyof science suggests that the assumption has been
invariably adopted. Our understandingof all nature, from the interactionsof
subatomicparticlesto theoriginof the universe,restsupon the assumptionthatthe
principles that emerge in the laboratory can be extended to domains where
experimental control is impossible. Geologists, following Hutton and Lyell,
enshrined the assumption in a technical term, uniformitarianism,since our
interpretation of most geological phenomena requires it. Our understandingof
evolutionarybiology and cosmology would be negligible withoutit. Inspired by
such distinguishedexamples, we may ask where the assumptionleads us in the
domain of cognitivephenomena.

The Problem

When, in the presence of certain stimuli,a particularbehavior is commonly

followed by a reinforcer,we speak of a three-termcontingency.Investigationof
the relationshipsamong the termshas revealed general principles that formthe
nucleus of operanttheory.An analysis of such contingenciesfacilitatesa plausible
and straightforwardinterpretationof much human behavior. When a child

This content downloaded from on Sat, 21 Jun 2014 17:45:04 PM

All use subject to JSTOR Terms and Conditions

responds "144" to the question,"What is 12 times 12?" we naturallyinferthatthe

question has been encounteredin school and thattheappropriateresponse has been
reinforced. There are two reasons why we are unlikely to cavil with this
assumption.First,the inferredhistoryis plausible; a parentor teachermay even be
able to point to a relevant flashcard or worksheet.Second, the performanceis
analogous to paradigmaticdiscriminationexperiments,such as one in which a
pigeon pecks a key under differentialcontrol of a green light. A great range of
other adaptive human behavior yields to comparable interpretations.Naming
objects, identifyingfaces, navigatingthroughfamiliarterrain,following a recipe,
operatingmachinery,recitingpoetry,swattingpesky insects, and countless other
commonplace behaviors can be assumed to have arisen over the course of an
appropriatehistoryof exposure to three-term contingencies.Furthermore, one can
plausibly assume that of
programs graduallychangingcontingencies have shaped
the highly differentiatedbehavior of the skilled athlete, artist,craftsman,and
technician.The three-term contingencyis a powerfulinterpretive tool because such
contingencieshave been thoroughlyanalyzed experimentally,theyare ubiquitous,
and their scope of application is apparentlylimitless. Moreover, the three-term
contingencyis a fundamentalunitof analysis: Orderemerges,even over lapses of
space and time, without our invoking mediating events such as physiological
processes and anatomicalstructures.
But not all human behavior is interpretedso simply. Consider the following

1) Appropriatebehavior sometimes occurs on occasions in which critical

featuresor relationshipshave never been encountered.For example, it is
unlikely that an educated adult has encounteredthe particularquestion,
"What is 542 plus 20?" yetfew would fail to reply"562." Althoughsuch a
performanceis commonplace, we cannot explain it by simplypointingto
the reinforcement of thatresponse to the question on an earlieroccasion.
An act of "mentalarithmetic"seems to be required.
2) Sometimesonly one responseis reinforcedin a particularsetting,but when
that settingrecurs,an entirelydifferentresponse is emitted.Suppose, for
example, we ask a friendon two consecutivedays, "What is today's date?"
Even if we were to lavishly reinforcethe firstresponse, "June 30th,"and
took care to hold constanteverydetail of theimmediatesetting,along with
all relevantmotivationalvariables,we would be surprisedif our frienddid
not emit a different response, "July 1st,"on the second day. Must we not
appeal to a "mentalcalendar,"ratherthana historyof reinforcement?
3) Some behavior seems to be undercontrolof futureevents or hypothetical
events that are not representedin the currentsetting.For example, we
mightpass up a rich dessert,citingits expected effecton our waistline,or
do calisthenics,pointingto its expected effecton our blood pressure.All
examples of self-controlillustratethe apparent control of behavior by
futureevents. More generally,we resolve all ethical dilemmas apparently
by comparingthe projectedoutcomes of our various alternatives.How can

This content downloaded from on Sat, 21 Jun 2014 17:45:04 PM

All use subject to JSTOR Terms and Conditions
Prvate Events and Complex Behavior

the future affect present behavior, unless we invoke expectations,

predictions,plans, or intentions?
4) Some behavior seems to occur in a stream,oftenhidden fromview and
seeminglyindependentof context,a streamthat cannot be easily broken
intodiscretethree-term units.We dip intothatstreamwhen we ask, "What
are you thinkingabout?" We usually accept the answerto such a question,
although we may sometimes suspect the respondentof lying. In neither
case, however, do we doubt that there really is such a streamof hidden
behaviorthatis the propersubject of our question.
5) Some behavior seems to be controlled,not by antecedentevents, but by
stored "memories." When we ask, "What color is Steve's new van?" the
answer appears to be a response to a retrievedimage, or perhapshas been
directlyretrievedfromsome storagevault where information about objects
in our experienceis kept.
6) Verbal behavior is richly structured,but the variables controllingthis
structureare obscure. We ask, "Where are you going?" but seldom "Where
you are going?" One can characterizethisstructurewithgrammaticalrules
thatdo not apply to nonverbalbehavior,and it is temptingto suppose that
humans are equipped with a special "faculty" that organizes our verbal
behavioraccordingto theserules.

These examples are commonplace,and theyshare the featurethatappropriate

behavior seems to come out of nowhere. At least when presentedas anecdotes,
thereis no orderor predictabilityat the level of observable behavior. A common
recourseis to assume thatorderarises fromwithinthe individualand will appear at
another level of analysis, the physiological level, or perhaps a hypothetical
cognitive level. Cognitive maps, schmas, lexicons, encoding and retrieval
mechanisms,intentionalfields,storageregisters,and otherhypotheticalconstructs
are invoked to impose order on the data. Unfortunatelyeach such concept
introduces a qualitatively new element that itself requires explanation or
justification.That is, each new term must ultimatelybe paid for in the coin of
physical,biological, or behavioral events. Like the improvidentdebtor who pays
offone creditcard by drawingdown the balance of a second, such devices provide
only temporarysatisfaction,forthe overall explanatoryburdenhas been increased,
In contrast,the behavioristinsiststhateach of the above examples is actually
a web of orderlybehavioral phenomena; the lack of order is an illusion arising
fromour observingonly a portionof the performance.As an analogy, suppose we
were to edit a videotape of a ping-pongmatchby excising or paintingout the ball.
The remainingperformance - two people waving paddles in the air, apparently
withouteffect would be bafflingto the naive observer,not to say ridiculous.We
are in much the same position when tryingto interpretcomplex behavior when
some of it is hidden from view, that is, when it is private or covert. We see
fragmentsof behavior thatoccur seeminglyindependentof controllingvariables.
The behavioristasserts that if all behavioral and contextual variables could be

This content downloaded from on Sat, 21 Jun 2014 17:45:04 PM

All use subject to JSTOR Terms and Conditions

evaluated, this illusion would disappear. However, this claim naturallyraises a

troublesome question: If relevant variables controlling human behavior are
commonlyunobserved,is not science helpless? Is thereno alternativeto invoking
hypotheticalconstructs?To answer this question, we need to digress and address
what it is thatwe can ask of science.

The Two Purposes of Science

Science serves two purposes. First,it underliesour masteryof nature:The

experimentalanalysis of natureleads to the discovery of principlesthat can be
exploited by technology; thus science can take credit for most of our medical,
material,and technical progress,and it promises much more to come. Second,
science helps us make sense of the world: It oftenprovides elegant and satisfying
explanations for the order we observe in naturalphenomena. Why are theretwo
tides a day, and not one? Why do all the planets orbitin a common plane? How
does a monarchbutterfly, with its insignificantbrain,navigate fromits birthplace
in Canada to a tiny winteringcolony in Mexico? We may have no interestin
controllingtides, planets,or butterflies,but we are fascinatedby such puzzles and
the answers that science provides. We would hate to be without the tangible
benefitsof science- our antibiotics,computers,and syntheticpolymers- but we
would hate even more to be plunged into the intellectualmidnightof the distant
past when naturewas seen as the slave of goblins and spirits.
The two functionsof science require differentlevels of control. As an
analogy, consider a card trickrequiringsleightof hand. To performthe trickmay
require extraordinaryfinesse and precision, but to understandhow it is done
requires none. Likewise, the masteryof naturerequires precise control of every
relevantvariable. If we are cloning a sheep, a few misplaced DNA bases or a few
extra hydrogen ions in the soup can spell ruin; if we are manufacturinga
microchip,slack of a few microns in our apparatus can make all the difference
between success and failure.But to resolve a puzzle about a naturalphenomenon
requires neithercontrol nor precision; we only need to discover one path that
naturemighthave taken to produce the phenomenonat hand, and if this account
invokes only familiar scientific principles and plausible events, we will be
satisfied. Hurricanes, volcanoes, tides, mountain formation,and the genesis of
planetsdo notlend themselvesto experimentalcontrol,but mostof us feel satisfied
by current scientific interpretationsof these things, however tentative and
Evolutionary biology supplies many such examples. Natural selection
explains an extraordinaryrange of cunning biological adaptations, and the
evolutionaryaccount is so elegant and so generalthatwe findit deeply satisfying.
But our interpretation of any particularadaptation is usually based on scanty
evidence. We explain the bombardierbeetle's noxious emissions by appealing to
millennia of differentialprdation in which the most unpleasant beetles were
avoided in favor of easier prey. We explain the extraordinarycamouflage of the
walkingstickin a similarway; those individualsmostdifficultto detectwere more

This content downloaded from on Sat, 21 Jun 2014 17:45:04 PM

All use subject to JSTOR Terms and Conditions
Prvate Events and Complex Behavior

likely to pass on theirtraitsthan othermembersof the group. Such accounts are

commonly speculative. For supportingevidence, we seldom have more than
scatteredfossil remains, some field observations, a few suggestive anatomical
details, perhaps some genetic data, in addition to only very general information
about changes in prevailingcontingenciesof survival over evolutionarytime. In
some cases we can offerno more thana plausible scenario,withno directevidence
at all. Nevertheless,evolutionaryexplanationsare satisfying, because theyshow us
one path thatnaturemighthave taken to produce the remarkableinterrelationship
between an organism and its environment,and the explanations appeal only to
easily demonstrated contingencies such as variability among offspringand
differentialreproductioniteratedover generations.No termsare introducedthat
cannotbe validatedunderideal conditionsin a laboratory.
It is importantto note that any particularevolutionaryaccount might be
utterlywrong; most such accounts are fleshed out with plausible inferences,not
empirical observations. For the purpose of resolving a puzzle about nature,it
doesn't matter,for what is at stake is not the accuracy of our account but the
adequacy of our analyticaltools. As long as thetools can be shown to be adequate,
we can resist the temptationto invoke imaginaryor occult influences. Perhaps
thereare threedifferent ways in which a magician mightpull a rabbitout of a hat.
We only need to know of one way to be satisfied that the performanceis not
outside the compass of our understandingof how the world works. Similarly,
evolutionary interpretations,however tentative, serve to displace appeals to
unknownor mysteriousforces.
Scientificinterpretationsfromthe fields of geology, cosmology, evolutionary
biology, and meteorologyall share the virtue of displacing occult theories by
showingthatestablishedscientificprinciplesare sufficientto account forpuzzling
phenomena. In the face of incomplete data, such accounts are tentative,but the
greaterpart of what passes for scientificunderstandingof the world is of this
tentativesort.Only a handfulof naturalphenomenahas been submitted,or is ever
likelyto be submitted,to controlledexperimentation. Experimentalanalysis has an
exalted statusamong scientists,and deservedlyso, forit underliesour masteryof
nature;but perhapsits most importantservice is thatit providesthenecessarytools
foreffectivescientificinterpretation. It not our laboratorydemonstrationsbut our
interpretations of naturethatare all encompassing,or nearlyso, and it is theythat
We distinguish,then, between the experimentalanalysis of nature and the
interpretation of nature.Experimentalanalysis underliesour masteryof natureand
requires the observation,measurement,and controlof all relevantvariables. Our
discovery and understandingof general principles of nature arise entirelyfrom
experimental analyses. Interpretationis the extension of these principles to
domains where observationand experimentalcontrolof all importantvariables is
impossibleor impracticalbut whereincompletedata are available.
Our understandingof complex human behavior is of this tentativesort. No
one, neitherbehaviorist,cognitive scientist,nor neuroscientist,is in a position to
offera definitiveaccount, except in veryrestricteddomains, for such phenomena

This content downloaded from on Sat, 21 Jun 2014 17:45:04 PM

All use subject to JSTOR Terms and Conditions

typically exemplify the interactionof many variables, some are unobserved,

outcomes may depend upon unknownhistories,and attemptsto measure relevant
variables are likelyto distortthe behaviorunderstudy.Perhaps the best we can do
is to liftthe shroudof mysteryfromsuch phenomena.Our purpose,then,is to offer
plausible interpretations of such complex behavior, interpretationsthat rest only
upon principlesthathave been establishedindependentlyof the phenomenato be
explained. It is importantto acknowledge thetentativenatureof thisenterprise,lest
we underestimateour successes.

An Example

Let us consider an example, chosen because everyone is likely to agree on

how appropriatebehavior actually emerges; it is only superficiallypuzzling (for
otherexamples of the presentapproach,in several cognitivedomains,see Donahoe
& Palmer, 1994; Palmer, 1991; 1998). Suppose we ask someone, "If each letteris
assigned a numberequal to its ordinal place in the alphabet, so that A=l, B=2,
C=3, and so on, what is the sum of D plus J?" Most adults answer this question
correctly,aftera considerable pause, but it is fair to assume thatthe question is
novel. If we restrictour considerationto just those responses thatcan be observed,
the behaviorof announcingthe answeris puzzling. The response is adaptive,in the
sense that it is scheduled for reinforcement,but it appears to be evoked by
variables (the context,the question) that have not been encounteredbefore. We
cannot simply point to a historyof reinforcementfor correctrespondingto the
question,as we can with"What is 12 x 12?" We explain thisapparentanomalyby
arguingthatthe subject has engaged covertlyin precurrent behavior,thatis, some
sort of related behavior thatpotentiatesthe targetresponse. Specifically,we note
that the question is similar enough in both intonation and wording to other
questions that our subject has encounteredthat the mere posing of the question
signals a negative reinforcement contingencyin which an aversive conditioncan
be escaped only by emittinga particulartargetresponse. Unfortunately, the target
response,"14," is not the prepotentresponse under prevailing conditions.We must
appeal to a history in which our subject has learned to to
respond questions for
which no immediate response is strong by emittingcollateral behavior that
produces supplementarystimuli that, together with contextual variables, are
sufficientto occasion the targetresponse. Our subject has undoubtedlylearned to
identifythe ordinalposition of an itemin an arrayby arrangingthe itemsin order
and countingoff items until the item is reached. In the presentcase, we suspect
thatthe subject has, among otherbehavior,emittedan intraverbalchain (reciting
the alphabet) while engaging in collateral tallyingbehavior. The emission of two
targetnumbersin the contextof "addition" evokes a final intraverbalresponse,
which is then emitted as the answer. This precurrentbehavior might be
conspicuous, but more commonlyit is below the thresholdof observabilityto an
onlooker(indeed, beforeannouncingthe answer,most subjectstwitchtheirfingers,
apparentlypacing a covert recital of the alphabet). The response "14" must be
assumed to be a low-probabilityresponse to the question, "What is the sum of D


This content downloaded from on Sat, 21 Jun 2014 17:45:04 PM

All use subject to JSTOR Terms and Conditions
Prvate Events and Complex Behavior

plus J?" but it is a high-probabilityresponse to the intraverbalchain, "4 and 10."

Thus the response is surprisingin relation to observable variables but is quite
unremarkableif we consider plausible mediating behavior. Readers who have
troubled to answer the question have undoubtedly found themselves doing
somethingof thissort,or somethingequivalent.
Our account is an example of an interpretation, analogous, for example, to
evolutionary accounts of bipedal locomotion in primates.It controlsno variables
and cites no experimentaldata. But, it shows how a superficiallypuzzling example
of behavior can be understoodas the productof a plausible historyand familiar
principles, such as generalization,reinforcement,and chaining. However, the
plausibilityof theaccount restson an appeal to covertmediatingbehaviorand their
stimulusproperties.Can we justifyresortingto variables thatwe cannotobserve or
measure? Is this not cousin to the discredited but popular habit of inventing
hypotheticalmechanisms to explain behavior? We must make a distinctionhere
between private behavioral events and the internal representationsand other
machinerycommonlyinvokedto explain complex behavior.

The Threshold of Observability

In the laboratory,we definebehavior as any activityof the organismthatcan

be shown to vary in orderly ways with the manipulation of antecedent and
consequent events. The orderlinessof these relationshipsserves as a criterionfor
determiningour analyticalunits in the science of behavior (Skinner, 1935, 1938).
More generally,then, behavior must be any activityof the organism that does
change in an orderlyway with such variables, whetherwe are in a position to
demonstratethat order or not. The status of behavior is independent of the
observer: Some behavior is out of reach of our manipulations,but it must be
assumed thatthe principlesgoverningthe behavior do not change, simplybecause
it is out of reach.
Whether a particularresponse can be observed is not a propertyof the
response itself; rather,it depends upon the vantage point, the faculties,and the
tools of the observer. A perfunctory "Hello" mightbe perfectlyclear in a quiet
room, indistinct a noisy one, and wholly inaudible across a quadrangle. For the
hard of hearing,amplificationmightbe required even under the most favorable
conditions.For the stone deaf, the response could be detectedonly by transducing
it instrumentallyinto another modality such as a flickeringlight or a speech
The same considerations,suitablymodified,apply to the bellowing of a drill
sergeant, the whispering of a school child, and the indistinctmutteringof a
disgruntledemployee. Each response will be observable under some conditions
and unobservable under others. Some behavior, such as covert speech, is
observable only to the subject himself, while other behavior, such as minute
muscle twitchesand vascular contractions,may be observable,even to the subject
himself,only with instrumentaltransduction.Perhaps some behavior is beyond
detectionwithour currenttools but will yield to futuretechnology.


This content downloaded from on Sat, 21 Jun 2014 17:45:04 PM

All use subject to JSTOR Terms and Conditions

Indeed, the boundary between public and private is continually shifting.

Recently researchershave shown thatthe activityof collections of cells, or even
single cells, in the motor cortex of humans and monkeys can be recruitedby
contingencies of reinforcement(e.g., Hochberg, Serruya,Friehs, Saleh, Caplan,
Branner, Chen, Penn, & Donoghue, 2006; Lebedev, Carmena, O'Doherty,
Zacksenhouse, Henriquez,Principe,& Nicolelis, 2005; Moritz,Perlmutter, & Fetz,
2008). By our definition, such neuralevents can be considered behavioral eventsas
well; their status is not defined by their dimensions but by theirrelationshipto
We can define the thresholdof observabilityas that set of conditionsunder
which theresponse is just detectable.Which side of thatboundarytheresponselies
on a particularoccasion will determinewhetheror not we call it observable. It is
simply a fact that, for a given observer at a given time, some portion of the
behaviorof a subject lies below thisthresholdof observability.
We can consider observabilityto be a continuum,and since the boundary
between the overt and covert is an arbitraryand variable point along that
continuum,we can assume that the laws of observed behavior hold for the
unobserved.That is not to deny thatdifferent response systemsmay be subject to
different constraints;rather,my claim is that observabilityitselfdoes not usefully
definetwo different categoriesof behavior,because it is determinedby the vantage
point of the observerand is not a propertyof the behavior itself.It is possible of
course that some classes of covert behavior do have special properties - there
mightindeed be a black void behind my back- but such a claim cannot be put to
experimentaltest,and it has no more statusthanany otheruntestablespeculation.
Since experimentalanalysis requiresthemanipulationof observable variables,
speculations about covert behavior play no role in the experimentalanalysis of
behavior. Rather,they serve science's second function:They help us understand
the world. Complex human behavior, particularlyphenomena commonly called
cognitive- language, memory,planning, problem solving- is one of the most
challengingfrontiersof science, and our abilityto interpret it runsfarahead of our
ability to experimentallyanalyze and control it. Our account of the subject
countinglettersof the alphabet is just a plausible possibilitythatis consistentwith
our own experience,but of course thisdoes not meet the standardsof a controlled
experiment.However, we have shown how adaptive behaviormighthave arisenby
exploitingonly termsof a behavioral analysis. No new termshave been invented.
To the extentthat we can point out ways that the known principlesof behavior
might account for all cognitive phenomena, behavior analysis can offer a
parsimonious, consistent, and intellectually satisfying account of human

The Illusory Power of More Permissive Paradigms

To recapitulate,behavior analysis interpretscognition as behavior, as the

confluenceof observed and unobservedevents interactingaccordingto established
behavioral principles.This mightseem self-evidentexcept thatit standsin contrast


This content downloaded from on Sat, 21 Jun 2014 17:45:04 PM

All use subject to JSTOR Terms and Conditions
Prvate Events and Complex Behavior

to a more permissive paradigm. Normative cognitive science also endorses

experimentalanalysis and acknowledges thatthe principlesof behaviorplay some
role in humanbehavior; however,it appeals as well to otherthings:

1) Hypotheticalconstructssuch as intentions,expectations,beliefs, images,

and representations.
2) such as memorystoresand thelexicon.
3) Controlprocesses such as encoding,storage,retrievaland elaboration.

That is to say, it is not constrainedby the assumption of uniformity.One

mightargue thatsince cognitivescience embraces principlesof behaviorand much
else besides, it is a superset of behavior analysis and is necessarily the more
powerfulparadigm.The forceof this conclusion depends entirelyon the meritsof
the additionaltermspermittedunderthe cognitiveparadigm.Give a man a 5/g-inch
socket wrenchand ask him to removethe sparkplug fromyourlawnmower.Make
the same request of a second man, but give him the key to a warehouse full of
tools. Which man will fetchthe plug first?Extra tools are a liabilityif theyare not
to the purpose. In practice, cognitive scientistshave devoted nearly all of their
effortto exploitingthe additionalterms.In all theprodigiousliteratureof cognitive
science, thebasic principlesof behaviorare scarcelymentionedat all.
One mightargue thatan appeal to covertmediatingbehavior is nothingshort
of an endorsement,half a centurylate, of the practices of mainstreamcognitive
science, thata supposed covertrecitationof the alphabet is no more scientifically
respectablethanan appeal to representations, encoding,storage,retrieval,schmas,
intentions,and so on; all are hypotheticalinterveningvariables. To argue so is to
miss an importantdistinction: Covert responses are not representations.The
interpretivetools of the behaviorist are constrained by an independent
experimentalanalysis; no explanatoryconcept can be invoked thathas not been
analyzed in the laboratoryunderexperimentalcontrol,and the termsmustinteract
according to empirical principles.Consequently,such termsare not free to carry
whatever burden the example requires. A covert response, for example, must
change in strengthin orderly ways, it must be plausible with respect to the
prevailingcontingencies,and theremustbe a plausible historythatwould predict
such a response. As a unit of behavior, it must have the dimensions of behavior
and cannot have special ad hoc properties. Mediating behavior that is not
constrainedin these ways can carryno explanatoryburden at all. It is true that
most cognitive metaphors can be interpretedas having some behavioral
dimensions; a partial translationis usually possible. But it is a hallmarkof the
cognitive science paradigm that it regards itself as liberated from the
methodological and conceptual constraintsof behaviorism. From the present
perspective,thatis an odd qualityto celebrate.


This content downloaded from on Sat, 21 Jun 2014 17:45:04 PM

All use subject to JSTOR Terms and Conditions

The Limitations of Interpretation

I have argued that the distinction between experimental analysis and

interpretationclarifies what science can offer in the domain of cognition.
Furthermore,I have claimed that interpretation is not merely a poor cousin to
experimentalanalysis; rathereach enterprisecontributescrucially to one of two
purposes of science, and with respect to our understanding of cognitive
phenomena,interpretation plays a dominantrole. But if it is to carrya burdenthis
heavy, we must be clear about what we mean by interpretation.Under what
conditions should we engage in interpretation ratherthan experimentalanalysis,
and what are therisksof doing so?
Skinnerdefinedinterpretation as

. . .theuse of scientific
termsand principlesin talkingaboutfactsaboutwhich
too littleis knownto makeprediction andcontrolpossible.. . .Platetectonicsis
an example.It is notphilosophy butan interpretationof thestateof thecrustof
theearth,usingphysicalprinciplesgoverning the behaviorof materialunder
high temperatures and pressuresestablishedunder the conditionsof the
laboratory, whereprediction

As Skinner's example makes clear, an interpretation can be considereda kind

of hypothesis,in the sense of a provisional explanation of a phenomenonin the
lightof a scientifictheory,but the termhypothesisis more commonlyapplied in
science to a tentativeprediction about the effect of some variable than to a
plausible explanationof factsat hand. However, scientificinterpretations share an
importantpropertywithhypotheses:They help guide and organize research.As I
will argue later,experimentalanalysis and interpretation are mutuallysupporting
enterprises,particularly in domains such as complex human behavior where
establishinggood experimental control is difficult.
An interpretation is only one possible scenario thatmightaccount fora state
of affairs,not a necessary one. A particularevolutionaryaccount of a biological
adaptationmay be just one of several equally plausible alternatives.We may have
no reason to preferone interpretation to another,but theyall serve the purpose of
showing thata phenomenon can be attributed to naturalcauses.
Interpretation differsfrom mere speculationin thatthe latteris unconstrained
by experimentalanalysis. Speculation may play a helpful role, so long as it is
clearly identifiedas such. It may be a usefulexercise to suggest,forexample, that
children learn language so reliably because they are endowed with a language
acquisition device. Discussion of such a proposal can help clarifywhat is known
and what remains to be shown. However, to the extent that such speculations
appeal to principlesor phenomenathatdo not have empiricalfoundations,theydo
not solve thepuzzles in question.
Although scientificinterpretation has been exalted in this essay, we must
acknowledge its limitations.Interpretations do not tell us how natureworks, but
how it mightwork; they are just plausible scenarios, not facts about the world.
Interpretations are only the extensionof establishedprinciplesto domains outside


This content downloaded from on Sat, 21 Jun 2014 17:45:04 PM

All use subject to JSTOR Terms and Conditions
Prvate Events and Complex Behavior

the laboratoryand cannot discover anythingnew. Consequentlytheyshould not be

advanced when empirical studyis possible. Interpretation should be reservedfor
only those phenomena of which experimental controlis impractical,unethical,or
impossible. Since interpretationcan be done from one's armchair, and
experimentationis hard work, the appeal of interpretation may seduce potential
researchers and actually interferewith the progress of science. There is an
additional and perhaps more serious danger: A plausible interpretationof a
puzzling phenomenon tends to satisfy our scientificcuriosityeven when it is
wrong. In such a case, it dulls our tendency to investigatethe matterfurther.
Unfortunately, the plausibilityof an interpretation can only be evaluated by an
expert in the basic science: Does it in factfollow from empiricalprinciples,and is
it consistent with available facts? Natural phenomena that resist experimental
control can invite many competing accounts. Because few are in a position to
evaluate them, a superficial account may be widely acclaimed and retard the
acceptance of a more cogent interpretation. Such riskscannot be avoided, and the
influence of a scientificinterpretation will depend in part on the vigor of the
paradigmin which the interpretation is embedded.

A Strategy for Understanding Cognition

An appropriatestrategyforunderstandingcognitivephenomena includes two

parallel endeavors. First is the unabashed behavioral interpretation of the entire
landscape of humanbehavior,showing how everyaspect of it mightbe accounted
for by basic processes, thus laying claim to the domain; second is experimental
analysis guided by those interpretations. The two endeavors can be thoughtof as
analogous to an immense paint-by-numberspicture. In this case, the picture
includes the entirepanorama of cognitive phenomena. Interpretation provides the
numbers;thatis, it providesonly an outlineof whata completeaccount mightlook
like. Experimentalanalysis fills in the colors, a pixel here, a pixel there,wherever
controlled observation is possible. However, each blotch of color added by
experimental analysis is recognizable only in the context of the interpretive
As suggestedby thisanalogy,theexperimentalanalysis of cognitionwould be
guided and inspired by the interpretiveframework. In the absence of the
framework,the empirical work- a bit of color here and there would seem
pointless and random.For example, several studies have confirmedthe functional
independenceof verbal operants,such as tacts,mands, and intraverbals(Lamarre
& Holland, 1985; Lee, 1981; Sundberg, Endicott, & Eigenheer, 2000). This
empirical work is significantin the contextof Skinner's interpretation of verbal
behavior,which predictedsuch effects(Skinner,1957). Outside of thatcontext,the
studies would seem isolated and unrelatedto otherverbal phenomena.Thus, even
though a complete experimental analysis of cognition may be out of reach,
experimentaldetails will accumulate over time that,togetherwiththe interpretive
framework,will provide a mosaic of graduallyincreasingclarity.


This content downloaded from on Sat, 21 Jun 2014 17:45:04 PM

All use subject to JSTOR Terms and Conditions

One need not feel dispiritedby the modestyof this goal. The role of natural
selection in the origin of adaptive complexityin naturewas once appreciatedby
only a relative handful of ardent partisans,but it continued to gather support
incrementallyas the principlewas shown to be able to integratenearlyevery new
experimentalobservation and field datum that emerged over the next century.
Now, of course, the importanceof naturalselection is not seriouslyquestioned in
scientificcircles, even though its role has been demonstratedin only a small
proportionof biological adaptations.To the extentthatcomplex human behavior
really is the product of selection contingencies,behavioral interpretations will
accumulate supportin an analogous way.
We might evaluate this optimism by considering what has already been
accomplished, and to what effect.Skinnerdevoted much of the latterhalf of his
career to a comprehensive interpretation of complex behavior, much of which
would be embraced by the termcognition,and thereare few topics thathe has not
addressed, at least in passing. Of course Skinnerhas not been alone in offering
interpretationsof complexity; there is a substantial and growing literatureof
similar work. However, it is fair to say that the compass of Skinner's work is
unusually wide, and that he remains the most influentialsingle behaviorist.
Thereforehe serves as a useful testcase: Has all this interpretation
been fruitful?
Has it led to a growing appreciation of the power and parsimony of the
behaviorist's position analogous to the increasing influence of evolutionary
A cursoryglance at the field of cognitionwould suggest a resounding"No."
In the hundreds of empirical studies and conceptual analyses on cognitive
phenomena published every year in mainstreamjournals of psychology and
philosophy,Skinnerand behavior analysis are usually represented,if at all, only in
caricature.The patentfailureof Skinner's approach is oftencited, usually to help
justifyan alternativeapproach. Even authorswho findthemselvesdrawn to quite
compatible positions take pains to argue that there are fundamentaldifferences
(e.g., McClelland & Rumelhart,1986; Snow, 1996), apparentlylest the reader
dismiss theirclaims out of hand. Of course much of this animus can be dismissed
as merely recycled dogma: There is seldom evidence that the dismissal of
Skinner's work is based on familiaritywith it. A few prominentcritics (e.g.,
Chomsky, 1959, 1971; Dennett, 1978) tend to be cited as authoritativeto the
neglect of primary sources. Nevertheless, this state of affairs must count as
evidence against my suggestion of the potential influence of behavioral
If science proceeded by acclamation, the popularityof competingparadigms
would be discouragingindeed, but the circumstancesthatmake a positionpopular
need not ensure its survival; the earthwas once overrunwithtrilobites,but today
they are just fossil curiosities. A more cogent criterionis whether Skinner's
interpretive exercises have contributedto a self-perpetuating
and growingfield of
inquiry that is to some extent nourished by such interpretations,
and here the
evidence is unambiguously affirmative. For several decades, Skinner's
interpretations of cognitive phenomena were admired by a handfulof people but


This content downloaded from on Sat, 21 Jun 2014 17:45:04 PM

All use subject to JSTOR Terms and Conditions
Prvate Events and Complex Behavior

ignored by everyone else. However, they graduallybegan to guide experimental

work and to inspire refinedinterpretations, much of which would be difficultto
evaluate if not forthe frameworkthatSkinnerprovided. Althoughthispublication
steamis small relativeto thatof mainstreampsychology,in relationto its originsit
is enormous.
The behavioral approach to cognitionis progressingsteadily.At presentit is
by no means as influentialas traditionalapproaches,but in theevolutionof science
it has two importantadvantages: It is parsimonious,and it stands on a foundation
of independentempirical principles. It offersa genuine explanation of complex

What Does It Mean to Explain Something?

In thisessay I have attemptedto describe and justifythe behavioral approach

to understandingphenomena thatinclude privateevents. I have argued thatin the
face of such gaps in the data, all paradigmsmust resortto interpretation, and that
interpretationplays a centralrole in science. Behavior analysis is distinctivein that
its interpretations invoke only established principles that have arisen from an
experimentalscience, and it is thereforeespecially well placed to offercogent
interpretations.Moreover,by restricting termsin thisway, behavior
its interpretive
analysis offers actual explanations for cognitive phenomena. When we have
interpreted puzzling phenomenon in terms of principlesthathave an independent
justification,we have resolved the mysteryshroudingthat phenomenonand can
reasonably claim to have explained it. In contrast,the seductive metaphorsof
cognitivescience raise as many questions as theyanswer. As Machado, Lourenco,
and Silva (2000) have asked of one such metaphor:

If we say thata ratnavigatesa mazeefficiently
eye a storedrepresentationof themaze, a cognitivemap,and thenadmitthat
no mind'seye, literally
thereis literally actionof scanning,
no internal literally
no map,at least in thesensethatwe usually conceiveofeyes,scanningactions,
andmaps,thenhowdoes ouraccountexplaintherat'sbehavior?(p. 30)

Not only should we like our explanations to rest on independentprinciples,

we should like them to be smoothly integratedwith the rest of biology. In
particular,we should like them to be compatible with what is known about
physiologyand evolutionarybiology, for otherwise,an account carries the extra
burden of accommodating any discrepancies. Here too a behavioral account
satisfies, for plausible neural mechanisms of behavioral processes have been
identified, and the adaptive significance of such processes is conspicuous
(Donahoe & Palmer, 1994). In contrast, metaphors do not have physical
foundationsor evolutionaryorigins. Only when they have been translatedinto
biological or behavioraltermswill it be possible to evaluate themby these criteria.
I have argued that the work commonly assigned to hypotheticalcognitive
constructs - mental arithmetic,problem solving, recall, planning ahead, and so
on- can be accomplished by mosaics of elementaryoperants,some of which are


This content downloaded from on Sat, 21 Jun 2014 17:45:04 PM

All use subject to JSTOR Terms and Conditions

commonlybelow the thresholdof observability.This claim cannot be proven,of

course, but it follows fromour assumptionof uniformity, thatwhatlies beyondour
field of view obeys the same principlesas what lies withinit. Since experimental
analysis requires the observation,measurement,and controlof relevantvariables,
the plausibility of the claim rests rather on our ability to offer cogent
interpretations of cognitive phenomena. Interpretations are not withoutrisks,but
they can a
provide hazy picture of the domain of and integratefragmentary
empirical work. Although such accounts are necessarilytentative,theycan serve
one of the main purposes of science to offerplausible explanationsof complex
phenomena in termsthat are rooted in experimentalanalysis and integratedwith
otherbiological sciences.


Chomsky, N. (1959). Reviewof VerbalBehaviorbyB. F. Skinner. Language, 35, 26-58.

Chomsky, (1971, December 30). The case against B. F. Skinner. TheNew YorkReview
of Books, 17, 18-24.
Dennett, D. (1978). Brainstorms. New York:Bradford Books.
Donahoe,J.W., & Palmer,D. C. (1994). Learningand complexbehavior , Boston:Allyn&
Hochberg, L. R., Serruya, M. D., Friehs,G. M., Mukand,J.A., Saleh,M., Caplan,A. H.,
Branner, A., Chen,D., Penn,R. D., & Donoghue,J.P. (2006). Neuronalensemble
controlofprosthetic devicesbya humanwithtetraplegia. Nature, 442, 164-171.
Lamarre,J.,& Holland,J.G. (1985). The functional independence ofmandsandtacts.
JournaloftheExperimental AnalysisofBehavior,43, 5-19.
Lebedev,M. A., Carmena,J.M., O'Doherty,J.E., Zacksenhouse, M., Henriquez,C. S.,
Principe, J.C., & Nicolelis, M. A. L. (2005). Cortical ensemble adaptation to
represent velocity of an artificialactuatorcontrolled by a brain-machine interface.
JournalofNeuroscience, 25, 4681-4693.
Lee, V. L. (1981). Prepositional phrasesspokenandheard.JournaloftheExperimental
AnalysisofBehavior,35, 227-242.
Machado,A., Lourenco,O., & Silva,F. J.(2000). Facts,concepts,andtheories: The shape
ofpsychology's epistemictriangle. Behaviorand Philosophy, 28, 1-40.
McClelland,J.L. Rumelhart, D. E., & thePDP ResearchGroup(Eds.) (1986). Parallel
distributed .
processingCambridge, MA: MIT Press.
Moritz,C. T., Perlmutter, S. I., & Fetz,E. E. (2008). Directcontrolofparalyzedmuscles
bycorticalneurons. Nature, (publishedon-line,October15,2008).
Palmer,D. C. (1991). A behavioralinterpretation ofmemory. In L. J.Hayes& P. N. Chase
(Eds.), Dialogueson verbalbehavior(pp. 261-279).Reno,NV: ContextPress.
Palmer,D. C. (1998). The speakeras listener: The interpretation ofstructural in
verbalbehavior.TheAnalysisofVerbalBehavior,15, 3-16.
Palmer,D. C., Eshleman,J.,Brandon,P., Layng,T. V. J.,McDonough,C., Michael,J.,
Schoneberger, T., Stemmer, N., Weitzman, R., & Normand, M. (2004). Dialogueon
private events. The Analysisof Verbal Behavior , 20, 111-128.
Skinner, B. F. (1935). The genericnatureoftheconceptsof stimulus andresponse.
JournalofGeneralPsychology, 12,40-65.


This content downloaded from on Sat, 21 Jun 2014 17:45:04 PM

All use subject to JSTOR Terms and Conditions
Prvate Events and Complex Behavior

B. F. (1938). Thebehavioroforganisms.
Skinner, New York:Appleton-Century-Crofts.
B. F. (1953). Scienceand humanbehavior.New York:Macmillan.
B. F. (1957). Verbalbehavior.New York:Appleton-Century-Crofts.
B. F. (1963). Behaviorism
Skinner, at fifty.
Science, 140, 951-958.
B. F. (1974). Aboutbehaviorism.
Skinner, New York:Knopf.
B. F. (1988). Replyto StalkerandZiff.In A. C. Catania& S. Harnad(Eds.), The
selectionofbehavior:Theoperantbehaviorism ofB. F. Skinner
: Comments and
Consequences(pp. 207-208).Cambridge:CambridgeUniversity Press.
B. F. (1999). The operational
Skinner, analysisofpsychological terms.In B. F. Skinner,
Cumulative record:Definitive
edition(pp. 416-430).Acton,MA: CopleyPublishing
Group.(Originalworkpublishedin 1945.)
Snow,C. E. (1996). Towarda rationalempiricism: Whyinteractionism is notbehaviorism
anymorethanbiologyis genetics.In M. L. Rice (Ed.), Towarda geneticsoflanguage
(pp. 377-396).Mahwah,NJ:Erlbaum.
Sundberg, M. L., Endicott,
K., & Eigenheer, P. (2000). Usingintraverbalprompts to
establishtactsforchildren withautism.TheAnalysisofVerbalBehavior, 77, 89-104.


This content downloaded from on Sat, 21 Jun 2014 17:45:04 PM

All use subject to JSTOR Terms and Conditions