Jurnal

Toward an Epistemology of Physics
Author(s): Andrea A. diSessa

Source: Cognition and Instruction, Vol. 10, No. 2/3 (1993), pp. 105-225
Published by: Taylor & Francis, Ltd.
Stable URL: http://www.jstor.org/stable/3233725
Accessed: 05-11-2015 04:03 UTC
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp
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 support@jstor.org.
Taylor & Francis, Ltd. is collaborating with JSTOR to digitize, preserve and extend access to Cognition and Instruction.
http://www.jstor.org
This content downloaded from 134.117.10.200 on Thu, 05 Nov 2015 04:03:56 UTC
All use subject to JSTOR Terms and Conditions
COGNITION AND INSTRUCTION, 1993, 10(2 & 3), 105-225
Copyright@ 1993, LawrenceErlbaumAssociates,Inc.
Toward of Physics
an Epistemology
AndreaA. diSessa
School of Education
Universityof California, Berkeley
The aim of this work is twofold: to understandthe intuitive sense of mechanism

that accounts for commonsense predictions, expectations, explanations,andjudg-
ments of plausibilityconcerning mechanicallycausal situationsand to understand
how those intuitive ideas contributeto and develop into school physics. To facili-
tate this, I provide a frameworkfor describing and correlatingcharacteristicsof
weakly organized knowledge systems. The frameworkis aimed at answering, at
a coarse level of detail, a set of questions central to a full theory of knowledge:
What are the elements of knowledge; how do they arise; what level and kind of
systematicityexists; how does the system as a whole evolve; and what can be said
aboutthe underlyingcognitive mechanismsthatare responsiblefor the normaloper-
ation of the system and its evolution?
The empirical base is a set of clinical interviews of undergraduatephysics stu-
dents trying to solve a set of specially designed problems. Observationsfrom this
core and from the existing literatureare extended with informal data and synthe-
sized using the general framework.
Major claims are that the intuitive sense of mechanism involves many simple
elementswhose originsare relativelyunproblematic,as minimalabstractionsof com-
mon events. The system as a whole is only weakly organized, and it is subjectto
a numberof constraintsincluding a relative lack of depth in justificatorystructure
and the inabilityto resolve conflicts on the basis of knowledge within the system.
Despite weak organization,the system exhibits some broadly characteristictraits,
a numberof which are identified. They include a prominentcausal schematization
in terms of agents, patients, and interventions("causalsyntax");a tendency to fo-
cus on staticcharacterizations of dynamicevents, includingthe globalform of trajec-
tories; and a relatively rich phenomenology of balancing and equilibrium.
The intuitivesense of mechanismcontributessubstantiallyto understanding school
physics. This developmentrequiresa denser knowledgeorganization,addingdepth
and breadththat allow more confidentapplicationof fewer fundamentalexplanato-
ry elements.
Requestsfor reprintsshouldbe sent to AndreaA. diSessa,GraduateSchool of Education,University

of California, Berkeley, EMST Division, 4533 Tolman Hall, Berkeley, CA 94720.
106 DISESSA
Thelatersectionsof themonograph havethreegoals:(a)to abstractandclarify

the theoreticalframeworkintoa formsuitablefor applicationto knowledgesys-
temsotherthanthephysicalsenseof mechanism; (b) to summarizeandindepend-
ently motivatemajorclaims;and to the
(c) by contrast, highlight implications of
thebasicclaimsof themonograph in comparison withotherproposedframeworks
for understanding commonsensephysicsknowledge.
INTRODUCTION
The Physical Sense of Mechanism

In dealing with the physical world, humansgraduallyacquirean elaboratesense
of mechanism-a sense of how things work, what sorts of events are necessary,
likely, possible, or impossible. When my older son was 6 years old, he was puz-
zled when the moon appearedto follow us in the car. He said that he thought
it was an illusion, althoughhe readily admittedhe did not know how it worked.
Thus, he must have had indirect reasons for believing the moon did not move
in that way. He also evidently knew about illusions and the contrastof appear-
ance and reality. He had enough feel for what an explanationwould be that he
knew he did not have one: Asserting that the moving moon was an illusion was
not deemed sufficient. All in all, such a simple remark is full of implications
concerning how much he knew about what things happen, what things do not
happen, and how one explains it all. This monographis about that naturallyac-
quired sense of mechanismconcerning the physical world and how it develops
towardexpertscientificunderstandingof physics. The ultimategoal for this work,
which is approachedhere only very approximately,is a computationallyexplicit
genetic epistemology thatexplains how experience feeds into knowledge like my
son showed, which in turn, feeds later into learning school physics.
A sense of mechanismis the knowledgethatprovidesus with the capabilityto:
1. Assess the likelihoodof variousevents based on generalizationsaboutwhat

does and does not happen.
2. Makepredictionsand "postdictions."Thatis, one can traceentailmentsfor-
wardor backwardin time, explainingwhatwill happenon the basis of what
is the case, and explaining what must have been the case in order for the
present circumstancesto exist.
3. Give causal descriptionsand explanations.Thatis, one can look at a phys-
ical event and assign credit or blame for what happens to certain aspects
of the circumstances and to general facts about the world.
Control of the physical world is one function for the sense of mechanism-
being capable of taking actions so that they have felicitous consequences. But
TOWARD AN EPISTEMOLOGYOF PHYSICS 107
there must be many such aspects of muscularcontrol that are inarticulate.I am

primarilyinterestedin the pieces of our ability to control the world thatcan have
impacton articulatereasoningand problemsolving, pieces thatconstituteor feed
into learning science.
An alternate,simple descriptionof the sense of mechanismwould be causali-
ty. Which events follow which othersregularly,and why do they do so? I deliber-
ately use the term sense of mechanismto emphasize that the picture I paint of
humancausality is dramaticallydifferent from many other characterizations.It
involves diverse and diffuse judgmentsand impressionsmore than it consists of
some small set of sharply defined and necessary principles.
I wish to chart the structureof this sense of mechanismas a knowledge sys-
tem. What are the elements of the system and to what extent are they isolable?
Are some of these the commonsense equivalent of physical laws? How are the
elements organized? Do they cluster, form hierarchies, or settle into layers of
importance?Is there a core notion of causality, even if it is not entirely localized
or theory-like, or is this a profoundly distributedsystem? How does sense of
mechanismdevelop, and, more important,because the motivationfor this work
is educational, how could it be made to develop?
To begin the exploration, consider an expert's understandingof the physical
world. There are two partsof a physicist'sexplanation.The first, circumstantial,
describes the conditions that, although they need not be the case, conspire to
produce the result. It happensthat one billiard ball strikes another. The second,
explanatorypart, is more central to our concerns. It is the physicist's sense of
mechanism and involves those facts about the world that are true in more and
more general circumstancesas the physicist considers more and more universal
and fundamentaldescriptions,ending with basic laws or principlesthathe or she
holds alwaysto be true. Balls bounce, but, more fundamental,energyandmomen-
tum are always conserved, and these determinethe detailed outcome of the bil-
liard collision. Althoughthe most fundamentallayer, here representedby energy
and momentumconservation,is the most readily recognizedas physics, not even
physicists have time to reduce every complex event to basic principles. Instead,
they accumulate a phenomenology of events, like bouncing. These events or
phenomena,althoughnot fundamentalin the sense of containinglaws, still have
special statusto the extent thatthey are known to happenwithoutdetailedjustifi-
cation at each occurrence. Light (i.e., illumination,shadows), the spectrum,ra-
dio waves, and even mechanicalforces such as pushes and pulls by which means
objects are moved aroundare all consequencesof Maxwell'sequations.Yet these
phenomena are, in some degree, independentlyunderstood.
In the less fundamental,more "phenomenological"layers of understanding,
the distinctionbetween circumstancesand explanatoryprinciples becomes con-
fused. This is because special circumstanceswarrantthe encoding of particular
phenomenafor rapid explanationand prediction, independentof a deeper sense
of mechanismthat might exist. Some specialized phenomenamay sometimes be
108 DISESSA
treatedas self-explanatory.Everyoneknows thata book can rest on a tablewithout

consideringthe compression of the electron clouds of the book and the table and
without doing the calculation that shows, with ordinarydensities of matter, it
is possible but very unlikely thatthe book could fall throughthe table. Everyone
knows a hot filamentcan be seen to glow without checking from first principles
thattungstenin a nearvacuumcan producevisible radiation(i.e., thatobtainable
temperaturesput the peak of the black body radiationcurve in the region of hu-
man visual sensitivity to electromagneticradiation).
The natureof the continuumfrom common events (which are simply known
to happen in familiar circumstances)to fundamentallaws (which are supposed
to explain all events) is a crucial concern in this monograph.Two of my central
claims relate to it. The first is that the naively developed sense of mechanism
does not come close to the expert'sin depthand systematicity.Instead, it is both
less focused and less integrated.Although physics-naivepeople make important
distinctions between superficial phenomenology and deeper mechanisms, these
distinctionsare very unlike those made by experts. For experts, phenomenology
must uniformlybe reducible, via an analysis of circumstances,to a few core the-
oretical ideas. This is not the case in the naive sense of mechanism. My second
centralclaim is thatlearningphysics can be viewed in significantdegreeas building
this gradient between phenomenologicaland fundamentalby reorganizingand
prioritizing existing phenomenology. This is an epistemological claim that the
development of scientific knowledge about the physical world is possible only
through reorganized intuitive knowledge.
IntuitivePhysics in Brief
Although there have been many attemptshistoricallyto understandthe develop-
ment of physical causalitythat relate in importantways to what is said here, one
recent trend is notable. Especially during the last 15 years, a significant body
of datahas accumulatedconcerningintuitivephysics, sometimes underthe label
of "preconceptions,""misconceptions,"or "alternativeconceptions"(Brown &
Clement, 1987, 1989; Clement, 1982, 1983, 1987; Confrey, 1990; diSessa, 1982,
1983; Eylon & Linn, 1988; McCloskey, 1983a, 1983b; Minstrell, 1982; Ronca-
to & Rumiati, 1986; Viennot, 1979; Vosniadou, 1989). Studies in the United
States have focused largely on studentsin high school and early college and, at
that level, tell a story of the failure of school physics to affect the fundamental
beliefs of studentsabout the workings of the physical world. When confronted
with qualitativeproblems,most of which scarcelyappeartrickyor out of the range
of basic understanding,studentsoffer descriptionsand solutions that are incon-
sistent with and often in direct contradictionto basic physics principles. Qualita-
tive is an importantqualification in that students may be capable of solving a
problemposed in explicitly quantitativeterms, yet they may thinkvery different-
ly when asked for a qualitativeanalysis of the same problem. These studies also
showthatintuitivephysicsis quiterobust-changingrelativelylittledespiteyears
of physicsinstruction.In fact, its robustnessmaybe its most strikingfeature.
Althoughtheparameters of its systematicity arenotwellcharted,it is notunusual
to findthatroughly50%of studentsagreeon the mostcommonanswerandthat
mostof theremainingstudentschoosefromamongrelativelyfew otheranswers.
A simpleandancientexampleof intuitivephysics,discussedin the dialogs
of Galileo,is the expectationthata cannonballreleasedfromthe top of a mast
of a movingshipfalls straightdown,directlytowardthe centerof the earth.Be-
causethe shipcontinuesto moveduringthefall, one mightreasonthatit moves
"outfromunder"the cannonball,whichthenlands"behind" the footof themast.
Thisis a misconception in thatthe cannonballactuallyfallsstraightdownin the
movingframeof the shipandmast.In doingso, it falls directlyto the baseof
themast(ignoringsmalleffectsof airfrictionandtherotationof theearth).Differ-
ently said, the cannonballstartsits fall alreadypossessingthe forwardmotion
of the ship,and,conservingthatmomentum,it thuskeepsupwiththe ship'sand
mast'sforwardmotion.
More subtly,one finds students'analysesthat are basedon correctliteral
descriptionsof phenomenabutthatimplyan incorrectunderstanding of the un-
derlying mechanism. A novice sees that a coin tossedin the air at
stops thepeak
of its trajectorybecauseof balancedforces. A physicistsees only one forceon
thecoin, theconstantdownwardforceof gravity.Theupwardmotionof thecoin
afterit leavesthe handperpetuates itselfwithoutexternalor internalforces.To
manynovices, satellites in
go circularorbitsbecausecentrifugalforce, which
by itselfwouldcausethe satelliteto fly away,is balancedby thepullof gravity,
whichby itselfwouldcausethe satelliteto fall to earth.A physicistagainsees
only one force, gravity,andmaintainsthatthe satelliteis falling(accelerating)
towardthe earth.Centrifugalforce simplydoes not exist.
Intuitivephysicsas describedearlierclearlyexists,butfundamental questions
aboundconcerningwhatone shouldmakeof it. Oneof thebestknowninterpre-
tationsis thatintuitivephysicsrepresentsa coherent,even theoretical,view of
the world.MichaelMcCloskey(McCloskey,1983a,1983b,1984;McCloskey,
Caramazza, & Green,1980)is a notableproponentof thisview. His ideasmake
an excellentpointof reference,to whichI referseveraltimes.The educational
implicationsof the view of intuitivephysicsas theoreticalincludethatmiscon-
ceptionscanandshouldbe confronted,overcome,andreplacedby validprinci-
ples (e.g., McCloskey,1983b).Forthisconclusionto be viable,misconceptions
needto be relativelyisolableandfew in number,theyneedto be falseor at least
unproductive so thatreplacement is in order,andthey needto be amenableto
with
"attack" dataandargument.Thismonograph questionsall of theseassump-
tions.Instead,I approach intuitivephysicsas anexpressionof anunderlying sense
of mechanism thatoccasionally exhibitsrelativelyuniformresultsbutonthewhole
lacks importantsystematicitiesof theoreticalscience. As such, it does not need
to be replaced so much as developed and refined.
110 DISESSA
Guideto the Monograph

The three central sections of this monograph- Elements, Development, and
Systematicity-constitute a synthesized set of interpretationsof a single empiri-
cal base. Thatbase was obtainedover 3 years, mostly from interviewsof college
freshmenwhile they took a course in mechanicsat MIT. Each section has a differ-
ent style of analysis. The section on elements consists of a relatively long list
of descriptions of elements of the sense of mechanism, including descriptions
of the contexts of their use. The section on development includes a numberof
case studiesof the developmentof some conceptsof Newtonianmechanics,show-
ing how the naive sense of mechanismis involved and how it evolves. Because
these case studies generally cover a broadertime scale than the empirical base
per se, they are necessarily more interpretiveand integrating.They interpolate
and extrapolatefrom patternsactually observed, toward importantlonger time-
scale issues of conceptualdevelopment. The section on systematicityorganizes
the interpretationsmade to thatpoint into argumentsagainsthypothesesthatcom-
pete with those developedhere aboutthe level and kind of systematicityone finds
in the sense of mechanism.
The three central sections just described can stand on their own, especially
for readers interestedprimarilyin the empiricalbase. However, the section im-
mediately following this one, Theory Sketch, develops a brief theoreticalsketch
to orient the readertoward the interpretationsof following sections. The theory
section preparesthe groundfor more detailedtheoreticaldiscussionsto come later
and also serves as a summary and overview of this work.
The section following theory concerns empirical method. It attends to how
datawere used to identifyelements of the sense of mechanism.It may be skipped
by readersinterestedmainly in the theoreticalline or simply in results. That sec-
tion should be most interestingto readers who want to know, "How do I know
an element of intuitive knowledge when I see one?"
Following the three core sections, Cognitive Mechanismsketches some cen-
tral issues concerningthe computationalgenetic epistemologythat constitutesthe
ultimate aim of this work. It deals with issues of modeling and modeling lan-
guages, explains and justifies assumptionsmade earlier about cognitive mecha-
nism, and points the way toward future work by identifying open questions. It
will be of most interest to artificial-intelligence-orientedreaders.
The final section of this monographoffers an interpretivesummary, in part
by contrastingthe broadapproachto conceptualchange in science developedhere
with some other recent work.
Appendix A augmentsthe methodologicalcomments in the text proper with
two case studies for the existence and described characterof certain elements.
Appendix B gives a relatively complete listing of the intuitive knowledge ele-
ments discussed in this monographand their properties. It may be useful as a
review or to refresh readers'memories when necessary.
THEORYSKETCH
My programfor studyinghumans'senseof physicalmechanismcenterson iden-

tifyingandanalyzingspecificelementsof knowledge.Theseanalysesareintend-
ed to accountforthe"preconceptions" of studentsin problemsolvingandrelated
activities.Ona largerscale,theyarecentralto a geneticepistemologyfor com-
monsenseknowledgeandto thepartiallearningtheoryforschoolphysicssketched
here.As I havedoneelsewhere(e.g., diSessa,1988),I referto thepresentview
of the sense of mechanismas "knowledgein pieces."
Thisviewof physicsunderstanding andphysicslearningis stronglyknowledge
based.It assumesonly a few very simplecognitivemechanisms,althoughthe
resultingknowledgesystemis conjecturedto be largeandcomplex.Methodo-
logically,I believeit importantto pursueat once a set of relatedissuesthatis
germaneto theorybuildingaboutany knowledgesystem:
1. Elements:Describethe size andcharacterof theknowledgestructures in-
volved.Relevantbutinsufficiently precisecategoriesareideas,categories,
concepts,models,andtheories.
2. Cognitivemechanism:Providean imageof the operationof the intuitive
knowledgesystem.Thepointis, in thefirstinstance,to havesomegrounds
forinterpreting problemsolvingandreasoning thatuseintuitiveknowledge.
Beyondthis, we want eventuallyto have a computationallyexplicitmodel
thatshowshow developmentemergesfromuse.
3. Development: Understand thegenesisanddevelopment of thesystem.We
wouldliketo understand howelementsandsystempropertieschange.Pat-
ternsin thosechangesshould,in turn,accountfor instructional difficul-
ties, like the persistenceof alternateconceptions,and shouldsuggest
instructional opportunitiesto enhancedevelopment.
4. Systematicity: Describethe level andkindof relatednessof the elements
in thesystem.Thisincludesdescriptions of decompositions intosubsystems
thatarebothrelativelyintegrated withinthemselvesandalso relativelyin-
dependentof othersubsystems.
Thecentralfocusin addressing theseissuesis a hypothetical
knowledgestruc-
tureI call a phenomenologicalprimitive,p-prim for short,havingtheseproper-
ties (followingthe previouslist):
1. Elements: P-primsarerathersmallknowledgestructures, typicallyinvolving

configurations of onlya few parts,thatactlargelyby beingrecognizedin a phys-
ical systemor in thesystem'sbehavioror hypothesized behavior.In somepartic-
ularlyimportant cases, p-primsarethemselvesbehavioral,or necessarilyentail
behavior, which allows them to serve importantroles in explaining physical
phenomena. P-prims of this sort may be self-explanatory- something happens
112 DISESSA
"becausethat'sthe way things are."In these cases, p-prims become the intuitive
equivalentof physical laws; they may explain otherphenomena,but they are not
themselves explained within the knowledge system.
The name,phenomenologicalprimitive,is meantto captureseveralof the most
importantcharacteristicsof these objects. They arephenomenologicalin the sense
thatthey often originatein nearly superficialinterpretationsof experiencedreali-
ty. They are also phenomenologicalin the sense that, once established, p-prims
constitutea rich vocabularythroughwhich people rememberand interprettheir
experience. They are ready schemata in terms of which one sees and explains
the world. There are also two senses of primitivenessinvolved. P-primsare often
self-explanatoryand are used as if they needed no justification. But also, primi-
tive is meant to imply that these objects are primitive elements of cognitive
mechanism-nearly minimalmemory elements, evoked as a whole, and they are
perhapsas atomicand isolateda mentalstructureas one can find. This latterspecu-
lation is not essential to the more general image of intuitiveknowledgepresented
here, but it helps specify a preliminarycomputationalmodel.
2. Cognitivemechanism:P-prims act largely by being recognized. Recogni-
tion does not literally mean being seen. Rather, it means being cued to an active
state on the basis of perceived configurations,which are themselves previously
activatedknowledgestructures.One can view this recognitionas occurringroughly
in layers. At the top are relatively conscious ideas and concepts that involve and
are cued by lower level elements, down to sensory schemataor other low-level
but less directly data-drivenaspects of internal state. In this very rough model,
p-primsoccupymidlevels.They belongneitherto the lowest, possibly"hard-wired"
and data-drivensensory elements, nor to the world of ideas, or named concepts
and categories. Learningshould provide that p-prims are activatedin appropri-
ate circumstances,and, in turn, they should help activateother elements accord-
ing to the contexts they specify.
For describing the operation and systematicity of p-prims, we need a more
refined model of cognitive mechanismthan simply recognition. This refinement
can be provided by a description of the local topology of the recognition net-
work. The topology is based on successive activation-which elementscue which
others. The way a particularp-prim'stransitionto an active state is affected by
other previously activated elements is called cuing priority. High or low cuing
priority indicates a stronger or weaker connection between structuresthat are
antecedentin the cuing sequence and the recognized one. A high cuing priority
means only a small additionalcontingentactivationis needed over the described
context to activatethe element in question. The contingentactivationis provided
by other parts of the network, and the context should be described in terms of
particular,relevant,and active elements. Suppressioncan be representedby nega-
tive cuing priority.
Reliabilitypriority describesprocessing initiatedby the activationof a p-prim
that can more or less directly affect that element's state at future times. In other
OF PHYSICS
TOWARDAN EPISTEMOLOGY 113
words,it describespotentialfeedbackthatcanreinforceor undothe initialacti-

vation.A highreliability(withrespectto a specifiedcontext)meansit is unlikely
a p-primwill be turnedoff by subsequent processing;it is an assertionof likely
reinforcement andunlikelysuppressionthroughall activationpathsthatreturn
to the element.
An examplemaybe helpful.Supposeyou tendto be tenseandnervousin so-
cial situations.We couldmodelthisby havinga tense-and-nervous elementthat
is cuedby numerousindicatorsof socialencounters,suchas the perceptionof
a flock of individuals(see Figure1). Thetense-and-nervous elementis related
withhighcuingpriorityto flock-of-individuals-in-proximity. The activationof
tense-and-nervous, in turn,mayhavea highcuingpriorityin relationto panic-
and-flee.Thisdoes not meanyou necessarilyenactthe consequencesof panic-
and-fleewheneveryou aretenseandnervous,butit meansit will nottakemuch
moreto enactthem.Tense-and-nervous specifiesa contextin whichonlya small
contingentcuing activationfrom other elements(e.g., these-are-threatening-
people)will activatepanic-and-flee.
Supposethatwhenyoumightneedsoothingyouthinkto lookforyourmother.
In thiscase, my-mommy-is-near? will havesomecuingconnectionfromtense-
and-nervous. My-mommy-is-near?, if activated,will initiatesensorysensitivity
or activitiesrelatedto determining whethermy-mommy-is-near! needsturning
on. If mommy-is-near! is turnedon, it stronglysuppressestense-and-nervous.
Innet,mommy-is-near? andmommy-is-near! arein thereliabilityloopfromtense-
and-nervous; the activation of tense-and-nervous mightresult,via some inter-
my-therapist- these-
is-on- people-are-
vacation! threatening
flock-of- tense- panic-

individuals-in- and- - and-
proximity nervous flee
mommy- mommy-
is-near! is-near?
FIGURE 1 A structuredpriorities network.
114 DISESSA
mediaries, in turningtense-and-nervousoff. Note that if you knew your mother

was nearby, tense-and-nervouswould alreadyhave been suppressedwithoutthe
reliability loop. Alternatively, other reliability checks activated by tense-and-
nervous, such as my-therapist-is-on-vacation!,could result in positive feedback,
securelylocking on tense-and-nervous.Tense-and-nervouswill have a high relia-
bility prioritywith respectto my-therapist-is-on-vacation! To summarize,the con-
nections, weightings, and triggering levels in the network establish the details
of a model. Cuing and reliabilityare qualitativedescriptionsof relationsbetween
elements in the model.
Of course, it is not good to press such a model too far with respect to the
complex processing suggested in this example. However, the frameworkat least
providesrelativelyunambiguousspecificationof possibilitiesfor simplersystems.
Structuredpriorities refers to the pair, cuing and reliability. Structuredmeans
priorities are not global; they do not provide a general ranking. Instead, priori-
ties are structuredaccordingto context, the stateof "neighboring" knowledgeele-
ments. In circumstanceswhere there is little chance of confusion or when I wish
to make a general statementabout both kinds of priorities in a fairly broadcon-
text, I will speak loosely of high or low priorityp-primswithoutspecifying relia-
bility or cuing.1
3. Development:As mentionedbefore, p-primsoften originateas minimalab-
stractionsof commonphenomena.In this monograph,I mostly assumethatorigins
are relatively unproblematicand focus more on the "life history"of p-prims, es-
peciallyhow they mightbecome embeddedin more physics-sophisticated thinking.
The developmentfromnaiveto expertphysicalintuitionis hypothesizedto occurin
the following ways. First, the ratherlarge but relativelyunstructuredcollectionof
p-primspresentin naive individualsgets tunedtowarduse in instructedphysics.
Unstructuredmeans cuing and reliabilityare only establishedin small neighbor-
hoods within the network. Priority is local, and there may be no central and
dominateelements. There may even be no sense-of-mechanism-basedway to de-
cide which of two p-prims actually should apply in a case of conflict. During
"tuningtowardexpertise,"the priorityof some p-primsbecomes greatlyenhanced
or reduced,andcontextsof activationmay migrate,expand,or contract,depending
on the elements' new roles in the developing physics knowledge system.
Undoubtedlysome entirelynew p-primsare generatedas the learner'sdescrip-
tive apparatuschanges to focus on different features and configurationsin the
1Structured priorities has little value in describingelements connectedby long chains of thought.
In the extreme case of several hoursor even days of thought, even if the thoughtsequence is coherent
in some sense, factors such as whether or not you managedto find the appropriateliteraturerefer-
ence, interruptionsby other tasks, along with the intrinsic instability of human thinking patterns,
make cuing and reliabilityprioritiesdubiousas intrinsicmeasuresof relationsin the knowledge sys-
tem. They "averageover" too many contingencies. However, intuitive thought operates primarily,
I believe, over short intervals of coherence. Long time-scale coherence, presumablya product of
persistentand/or easily reactivatedelements (e.g., problem-solvingstrategiesor goal, plan, and in-
terestpatterns),will not concernus in greatdetail.This is clarifiedin the sectionon cognitivemechanism.
physical world. But, a more drastic revision in the intuitive knowledge system
is in the change in function of p-prims. They can no longer be self-explanatory
butmustdeferto muchmore complex knowledgestructures,such as physics laws,
for justification. P-prims come to serve weaker roles, as heuristic cues to more
formal knowledge structures,or they serve as analyses that do their work only
in contexts that are much more particularthan the range of applicationof the
general or universallaws of physics. I call this reuse and integrationof intuitive
knowledge structuresinto the functionalencoding of expertisedistributedencod-
ing. This name is intendedto imply thatthe encoding of, for example, a physical
law may be spread over many intuitive contributorsthat each play some small
role in "knowingthe law." Whereas in some cases knowledge may be packaged
in explicit bundlessuch as propositionsand formulae,invoking such bundlesand
unpackingtheir meaning in contexts of applicationmay requirea large number
of specialized structures, which may be p-prims.
Accompanyingthis change in the functionof p-primsfrom relatively isolated,
self-explanatoryentities to pieces of a larger system is a substantialstructural
change in the priority network. The depth, breadth, and integrationof the ex-
pert's priority network marks a major change from intuitive physics.
4. Systematicity:Systematicityis difficultto approachfor three reasons. First,
no initial assumptionsabout systematicityare built into the theory. Second, the
empirical methods used so far are better crafted to identifying elements rather
than systematicities.Finally, work so far suggests thatthe set of physicalp-prims
is, in fact, ratherlarge and loosely coupled. Nonetheless, because of its probable
importance,I begin with the following a priori list of kinds of systematicity,to
which I later attach appropriateexamples.
A. Mutual use: The mere use of p-prims in dynamic sequence, or simply in
relatively standardclusters, provides a kind of systematicitythat can ac-
count for sets of p-prims all being raised or lowered in priority simul-
taneously.
B. Commonattributes2(common'base vocabulary"):If some p-prims all in-
volve the use of a common base vocabulary of other prims, they clearly
enter into a particularrelationshipwith each other. A specific set of attri-
butes might provide for a kind of utility package that is frequently used
and determines overall characteristicsof the system, such as the salience
of whole classes of phenomena.
C. Top-downcoherence: Symbolic and verbal propositionsare prominentin
instruction.It is possible to view these as being learnedprior to the broad-
er coordinationsin intuitive knowledge that are eventually required. This
is like the way learning slogans may precede a deeper commitmentto a
political ideology. Learningby starting"atthe top" in this way is similar
21 use the term attributein a very general sense as a feature of a situation, without necessarily
implying strict attachmentto objects, a family of attributevalues, or other systematicities.
116 DISESSA
to Freudianviews that take a single memorableevent to organize a pano-

ply of reactions and strategies.
The subtletiesand reliabilityof top-downcoherence generationas a de-
velopmentalprinciple are importantto understand.Most schooling seems
to countheavily on explicit andliterallyrememberableelements. My work-
ing assumptionis that this only works well within subsystemsthat already
involve a sufficiently rich and reliable network. As a minimal specifica-
tion, I would like the theory sketch developed here to be capable of ex-
pressing the difficulties in top-down development.
D. Mutualplausibility: The results of particularepisodes of situation-specific
reasoning can accumulatea kind of integration.Importantelements rein-
force each other or generatenew elements via specific episodes of reason-
ing.3 Phenomenologicalsyllogisms are a class of mutualplausibilities:One
may simultaneouslynote that "x means (implies) y, and y means z," and,
therefore, encode that "x means z." For example, heavy things generally
move more slowly, and if somethingmoves more slowly, it generallytakes
more time to complete an act. So one may "conclude"and separatelyen-
code that heavy things take more time.
In the previous example, both premises make presumptions. Heavier
things move more slowly only if the force used to propel them is the same
as thatpropellingthe lighter object. The conclusion also, therefore, makes
these presumptions.The accidentalfailureor systematicinabilityto encode
presumptionsandprerequisitesshouldbe expected to typify low-reliability
subsystems. That is, an expectationor deductionmay be valid only by vir-
tue of situationspecifics that might not be encoded. Indeed, it seems em-
piricallytruethatmany misconceptionscome simply from using an element
outside its range of legitimate applicability.
The term syllogism suggests predicatelogic, but logic is intendedonly
as a familiar example of reasoning processes. Weaker and less conscious
reasoningpatternsmust be involved if mutualplausibilityis to account for
systematicitiesof lower priorityelements. Describingthese processes, then,
is central to the goal of understandingthis type of systematicity. I make
no presumptionthatthese are complex, generic processes thatwork across
a range of circumstances, say, analogical processes. Instead, they might
best be described in terms of configurationsof knowledge elements that
lead to specific changed relationsas a result of a particularpatternof use.
E. Completeness:A set of lower priority p-prims should be expected to be
linked to some contexts in order to fill out explanations for real-world
phenomenology not covered by more fundamentalintuitive expectations.
This class includes "excuses" for why p-prims do not work in certain
3This is really a developmentalprincipleratherthan a systematicityper se. However, the result-

ing systematicity might be most parsimoniouslydescribed by the processes that create it.
circumstancesor why unpredictedphenomenaactuallyoccur. Frictionand

magic may be invoked to explain otherwise inexplicable phenomena.
F. Abstraction:In the denser partsof the intuitiveknowledge network, sever-
al phenomenamay be relatedby having a common abstraction.If there is
a core to intuitive physics, it may lie in a broad or universal abstraction
commonto diversephenomenology.In general,commonabstractionsshould
have their own cuing networks, which may be expected to behave some-
thing like the disjunctionof the cuing patternsof the specializedelements.
METHOD
The primaryempiricalbase for this monographis a series of interviewsconduct-
ed over a 3-year period with studentstaking elementaryphysics at MIT. Each
of the approximately20 students was interviewed for roughly 1 hr each week
during the course of first-term physics (mechanics). The sessions were audio
recorded. Studentswere selected on the basis of (a) doing well in physics in high
school (and almost all subsequentlydid well in their course at MIT) and (b) be-
ing reasonablycompetentin thinkingaloud as judged in a preliminaryinterview.
To this base, I have added the experience of informally interviewing a signifi-
cantly greaternumberof subjects, from high school level to adult nonscientists.
This seems appropriatebecause the natureof the enterpriseis to uncover plausi-
ble structuresand mechanisms, not to prove the existence of any particularone
or to accumulatereliable statistics.
Data Analysis Overview

This monographis theoretical.Its aim is to develop a frameworkfor understand-
ing the origins and developmentof commonsense knowledge aboutthe physical
world, particularlyas it influences the learningof school physics. Nonetheless,
dataplay three importantroles. First, they providethe basis for abstractingtheo-
retical ideas and a testing groundon which to refine them. The role of empirical
work in the developmentof theory is much less discussed and standardizedcom-
paredwith its use in testingpredictionsor developingfacts withinwell-established
frameworks. But the importantsuggestive and refining roles of data in theory
constructionare undeniable.Second, datagroundtheoryandinstantiateit in prepa-
rationfor furthertheoreticaland empiricalwork. The principlesthat identify in-
stances of theoreticalobjects are frequentlynontrivialand need articulationand
exemplification. That the notions developed here come with substantialempiri-
cal elaborationis, I believe, to the credit of the theory. Third, the particularem-
pirical results developed here should, if the theory sketch is correct, have value
in the applicationof the theory, for example, in instruction.If particularintuitive
conceptions play importantroles in the developmentof physics understanding,
it warrantsour effort to uncover and describe them.
118 DISESSA
The formaldatacollection phase of this work consisted of open-endedclinical

interviewsconcerningan evolving set of problematicsituations.I believe the data
themselves are relatively unproblematic.They are similarto that which has been
produced in many intuitive physics studies in the past 15 years. Indeed, I draw
directly on other people'sdatato build some of the argumentshere. I have added
detail to my own observationsroughly to the extent that I believe them to be sur-
prising and to the extent that they are not documentedelsewhere. Fundamental-
ly, however, this is not a data-impoverishedenterprise; what cleverness there
is lies in the synthetic analysis.
Interpretationof the data is not as straightforwardas its collection for two
main reasons. The first is thatthe datainitially served primarilyfor theory build-
ing. The second reason is that the theory sketch that emerged entails a number
of constraintsand difficulties in interpretingdata that are not easily overcome.
In particular,no uniformandpackageabletechniqueexists that, for example, iden-
tifies p-prims, and I do not thinkone shouldbe expected. The case for each must
be relatively extensive, involving diverse forms of argument.
Difficulties notwithstanding,I discuss interpretationtechniquesin orderto put
other researchersin a reasonableposition to evaluate the techniques'adequacy
and to help some who might choose to pursue similar analyses. I do this first
by explaining some of the in-principlelimitationsthat emerge from the theory.
Then, I provide a list of heuristic principles on the basis of which cases for the
existence and proposed characterof individualelements may be made. Identify-
ing elementsandtheir characteristicsis the core of the contributionthatdatamake
here. Extrapolationto system propertiesand developmentare made on the basis
of elements, with comparativelylittle direct empiricalinputto these otherissues.
Finally, I also provide case study argumentsabout the existence and character
of particularp-primsthat can be made on the basis of my list of heuristicprinci-
ples. These case studiesappearin AppendixA. Readerswill find them more com-
prehensibleafterthe examplesused in the appendixare presentedin the text proper.
Difficultiesin EmpiricalInvestigationof P-Prims
Micro-events. P-primsare, in general, rathersmallandparticularknowledge

elements among a large collection. Their applicationmay be fleeting, and a par-
ticular element may seldom apply. Even less often will a p-prim, in isolation,
direct the flow of problem solving, except locally. More likely, p-prims will be
used in clusters or in combinationwith other kinds of reasoning.
Nonbehavioral. Problemstrategies,goals, andplansorganizesubstantial

actionpatternsin ways thatare, comparativelyspeaking,transparent.Means-ends
analysis determinespatternsof search for operators,and it determinesconditions
of satisfaction. P-prims entail no such general patterns. Their content is not
specifiable in advance, for example, by abstracttask analysis, and reasoningthat

follows up a p-primactivationwill likely dependdelicately on the problemsitua-
tion and the subject'smental state beyond the activation of that single p-prim.
A p-prim that accounts for a subject'ssatisfaction with a part of his or her
problem-solvingstate is unlikely to be commentedon; in its firmest application,
a p-prim will be treated as self-evident. A p-prim that accounts for an unease
and motivates continuedconsiderationstill does not determine, of itself, much
about how improved understandingis sought.
Inarticulate. P-prims are not strongly related to dictionarylexicon. Much

less do they have explicit propositionalform. I presume that conscious access
to their applicationis very limited, mostly localized in satisfactionor dissatisfac-
tion with a currentstate of understanding.Subjectsmay make predictionson the
basis of a p-prim, but the prediction is not the p-prim.
Unfamiliarvocabulary. Ratherthanrelatingto linguisticallysanctionedat-

tributes,p-primsare likely to have been abstracted,for example, in termsof body
sensationsand internalsensorimotorterms. Not only does this contributeto sub-
jects' inarticulateness,but it also means that, as theorists, we have a muchtough-
er job finding adequatedescriptions.
Ultimately, we must describethese mentalobjects in terms defined by the rest
of the knowledgenetworkthatcues andfollows up on a p-prim'sactivation.Other-
wise, we will fail to discriminatethe contexts in which a p-prim will actually
be used from less relevantexternalcharacteristicsof the problemcontext or less
relevantcharacteristicsof the particularuse of the p-prim. This puts a heavy bur-
den on completeness and on the precision of the theory of operationof p-prim
networks. Interfacesto languageand other reasoningcapabilitiesmay eventually
need to be described.
it is an individual's
Bootstrappingproblem. Fundamentally, extended ex-
perience with the physical world that determineswhat particularp-prims exist.
These are not specified in advance by the theory. So validationthrough estab-
lishing systematicitiesamong elements and continuitiesin developmentdepends
on getting the elements right, not just their generic properties. There are few
strong "dataless"predictions to be made.
A sociological argument. The diversity of empirical claims relating to

intuitive physics in the existing literatureis impressive. There is no agreement
on terms of description, much less on particular elements. Intuitive physics
is described as a theory, a systematic but alternativeframework, or a series of
isolated misconceptions. Intuitive explanations may even be treated as noise
on the central problem of building expertise and, as such, ignored. See Smith,
diSessa, and Roschelle (in press) with regardto these differenttreatments.The
120 DISESSA
diversity of views on intuitiveconceptionscalibratesthejob before us. My judg-

ment is that, before any firm empirical agreementcan be fashioned (or simul-
taneous with it), we need an ontological agreement on the form of knowledge
that common sense and intuition represent.
All of this motivatestheory building as a response. It is only in this way that
the foundationfor cumulativeempirical work can be laid and that a breadthof
observations can be coordinatedto draw sharp and firm conclusions. Like the
discovery of electrons, I do not thinkconcepts such as p-primswill be confirmed
by fine methodsthat "observe"them. Rather, confirmationwill evolve if the de-
veloping theory finds broad applicationand success.
In summary, the sense of mechanism is, by hypothesis, rich, diverse, deli-
cate, inarticulate,and (in many ways) ratherunsystematic.It probablycontains
"meaningless"elements, by ordinarystandardsof meaning. Unreliableand sparse
knowledge systems (terms that are developed more systematicallylater) are re-
mote enough from everydayvocabularyandcommonsenseepistemologicalterms
thatcriticalexperimentsand refinedobservationaltechniquesare, by themselves,
extremely unlikely to do the job that needs to be done.
P-Prims
HeuristicPrinciplesfor Identifying
Despite the difficultiesenumerated,I believe thereare opportunitiesto buildrela-
tively firm empiricallybased accountsof p-prims. Again, the core task is to bring
data and argumentsof various sorts to bear on the natureof individualp-prims.
I use data in the verbal protocols to build a case for each p-prim. Subjects'spon-
taneouslyproposedpredictionsand explanationsare critical data. Evidence con-
cerning satisfactionwith a particulardescriptionof a situationand the predictive
implicationsof those descriptions(e.g., confidenceor, alternatively,ambivalence
or searchfor alternatives)is also particularlyimportantin establishingpriorities.
Many times I describe data as generalities about predictions and explanations.
These are comparableto typicalepidemiologicalresultsof misconceptionsstudies
that report correct and incorrectanswers and sometimes also report suggestive
explanationsproducedby subjects.Populationfrequencies,however, are not rele-
vant to the argumentsmade here. Sometimes I use particularcritical instances
or citations from protocols to make more particularpoints.
The cases that emerge concerning individualp-prims need criteria for judg-
ment, which are provided by the principles that follow. If a case is strong by
many or most of the criteria, the proposedp-prim should be consideredvalidat-
ed. If it is weak, the case may be improved by reformulatingit in view of its
particularweaknesses,proposingalternativeinterpretations,and evaluatingthose.
I will not belaborthe ways in which these principles reinforceeach other and
follow from the theory sketch. To help make connections with what follows,
however, I anticipatesome results in the context of the principle that gave rise
to those results. Most of the principlesare two-edged. They bothjudge proposed
p-prim descriptions and propose strategies to uncover other p-prims or better

descriptions.
Principleof obviousness. Thefamiliarity

andunproblematic
natureof some
physical events needs explanation.In the presentcontext, this usually meansthey
need a p-prim to attach to them. In general, p-prims establish abstractclasses
of unproblematichappenings. This is the opposite of misconceptions research
strategy, which never analyzes "correct"intuitions.The principleof obviousness
gains explanatorypower in conjunctionwith the principle of invariance(to fol-
low); having understoodp-prims underlying common events, we may be able
to understandsubjects'reactionsto uncommonevents using those same p-prims.
Principleof impenetrability.P-primsarerelativelyprimitivein anexplana-

tory sense. This is the primarysense of primitive in p-prim. If people are satis-
fied makingan explanationby assertinga description,this likely indicatesa p-prim.
In contrast, we must recognize that people can sometimes form complex expla-
nations, for example, of devices using mentalmodels in a more or less articulate
way. These explanationsmay hint at manyp-primsalong the way but are unlike-
ly to display p-prims directly in the presentedexplanation.The principle of im-
penetrabilityis limited by the fact thatp-primsmay frequentlybe only relatively
primitive. Thatis, some reliabilityloop may lead to finding otherexplanations-
other p-prims, combinationsof p-prims, or more macro-explanatorysystems.
Principleof diversity. I believetherearemanyp-prims.So, heuristically,

it makes sense to retaina skeptical stance towardunification. Attend to nuance.
The principle of diversity is limited in that taking each instance of felt-to-be-
explanatoryanalysis to entail a different p-prim underminesthe very function
of p-prims. Again, see the principle of invariance that follows.
Principleof coverage. Thebreadthof commonexperiencemustbe covered

by p-prims. Misconceptions approachesgenerally fare badly by this criterion.
It is easy to find situations that have nothing to do with misconceptions listed
in the literature,in which people have no trouble predictingor explaining. The
principleof coverage especially critiquesreductionistprogramsthat see physical
intuitionto reside in small sets of principles that cover only a schooled class of
problems, such as trajectoryproblems. In contrast,phenomenographicanalyses
(Marton, 1981) are more consonant with the principle of coverage.
The principle of coverage several times led me to discover new classes of p-
prims. For example, constraintp-prims (p. 133 and following) emerged in con-
siderationof obvious situations,such as why a book rests comfortablyon a table.
Subjectsalso gave accounts of other situations, such as why a ball follows a par-
ticular constrainedpath, that appearednot to involve already-catalogedp-prims
(or listed misconceptions). Similarly, problemssuch as an orbit arounda square
122 DISESSA
planet (p. 165) initiatedconsiderationof figural primitives as a class-those that

result from attending to overall spatial form, for example, of trajectories.
Principleof strong vocabulary. P-primsprobablyclusterin areasof strong

descriptive (representational)capability. Indeed, they may be classified by such
vocabulary. For example, figuralprimitives(p. 165 and following) contrastwith
interactiveones: p-primsof observationversus p-primsof participation.The fol-
lowing two principles suggest thatthinkingaboutthe origins of p-primscan help
identify them.
Principleof unproblematicgenesis. Generally,thereshouldexist com-

mon events in which a p-prim might archetypicallybe used and from which it
may plausibly have been abstracted.This involves both the "vocabulary"availa-
ble for abstractingthe p-prim and the availability of common events that may
be governed by the p-prim. Functionality(discussed later)-that invoking a p-
prim is more thansimply feasible;it is also likely useful-is importantwith respect
to genesis. The principle of unproblematicgenesis is limited by the fact that p-
prims might be abstractedin one class of situations and migrate to others.
Principleof the body.4 Thisis a specializedprincipleof strongvocabulary

(as just discussed) and continuity (discussed soon). P-prims are likely to be
4This heuristic is too close to a programof studyingmeaning advancedby MarkJohnson(1987)

not to remarkon the similaritiesand differences. Johnsonis centrallyconcerned with how meaning
emerges and extends from bodily experiences in the same way I believe experiences-prominently,
but not exclusively, bodily experiences-form the basis of naive causality and, eventually, of the
expert causal sense as well. Three of the centralchaptersin Johnson'sbook deal with issues of force,
issues of balance, and spatial organizationand constraint, roughly parallel with the subsections of
the Elements section thatdescribe differentclasses of basic causal p-prims. I do not pursuea detailed
comparison here. There are, however, three central differences between our points of view. First,
Johnsonis interestedmuch more in "horizontal"issues, thatis, how meaningextends across multiple
domains. I am primarilyinterestedin the details of meaningwithin a selected domain (i.e., mechani-
cal causality). In addition, a fundamentaldimension for me is "vertical."The pursuitof deep expla-
nationsand the relationsof shallow to deeperexplanationsare critical. Johnsondeals with descriptive
phenomena-how one describes events; I deal, more centrally, with explanatoryones-why things
happen. Second, Johnson is not very concerned with what is here most central, change. How are
some conceptualchanges blocked or impeded, and how are some open and awaitingeither particular
moves of instructorsor particularproblem-solvingor explainingexperiences?In particular,I attend
to what happensto naive causal schematizationswhen they are changed and incorporatedinto expert
scientific understanding.Third, Johnson has a vested interest in denying the existence of abstract
(or simply"logicalandrational")schemata.Instead,he concentrateson dynamicflexibilityandmetaphor-
ical extension. These are processes that he claims explain broad use. In contrast, my view is that
p-prims may apply broadly (and may be the root of metaphoricalor analogical transfer), in part be-
cause of theirhighly schematicnature.I am less interestedin accountingfor the full nuanceof mean-
ing and descriptivecapability.I am more interestedin the detailsof specificallyexplanatoryschemata.
I am not directly concerned with how words collect and combine senses or p-prims, or with how
the felt-to-be-explanatorycore of certaindomains, like mechanicalcausality, is similarto and differ-
ent from the cores for other domains, such as law or reason.
abstractedin internallyevidentterms, especiallyearly in development.Thus, agen-

cy, (muscle) tension, and so on are likely to be representedin importantbase
vocabulary for p-prims. The following two principles suggest that considering
how p-prims are used can help identify them.
Principleof functionality. Thisprincipleemergesfromthepresumption

that
the sense of mechanism, intuitivecausality, evolves to serve individualsin deal-
ing effectively with the physicalworld. P-primsthatare "wrong"(by school stand-
ards) are very likely to be better understoodif they are described in terms that
make evident contexts of useful application. Naturally, such applicationshould
conform with the principleof readyavailability(discussednext) and may benefit
from use of the principle of strong vocabulary.
Principleof readyavailability. Oneshouldbe ableto understand

howa pro-
posed p-primapplies to any of its situationsof use on the basis of relatively ready
intuitive representationsof those situations. This depends critically on vocabu-
lary, as discussed before, but also entails analysis of particularsituationsin those
terms.
The next two principles have to do specifically with how the time-sequenced
use of different explanations-in an interview, across months of instruction, or
years of development-can be brought to bear for identifying p-prims.
Principleof continuity. A fundamental

constructivist
principleis thatnew
knowledge arises from old. Similarly, p-prims evolve from earlier knowledge
so thatearlierknowledge provides good hints for later. Understandingthe genet-
ic pathof a p-primcan help explain aspects of its characterthatare not otherwise
evident. Thus, data on children'sconceptualdevelopmentare especially useful.
See also the principle of scavenging data.
Principle of dynamic. The evolution of subjects'explorationsin an inter-

view can give very importantinformationabout p-prims. It is not only the first
or final reactionsthatare relevant.First answersmust makeuse of the most ready
vocabulary, especially if they are firm. P-prims of generally high priority may
be evoked ("thingsthat usually work" are almost always a good guess to start
with), then retractedon closer considerationof the situation particulars.Later
descriptionsare indicativeof reliabilityin the context more thanthey are indica-
tive of direct and simple cuing. Paths taken by subjectsbetween initial and final
stances may indicate which features of the situation, graduallyuncovered, lead
to the cuing or reliabilityjudgments involving particularp-prims. In addition,
many system propertiesare implicatedin such explorations. For example, alter-
native explanationsthatare consideredhelp the researcherdeterminethe aspects
of situationsattendedto; they may show richness or sparseness of the set of p-
prims that apply to a context; and they may show reliability considerationsin
124 DISESSA
"competitive
subjects' Ingeneral,I foundfirstanswersalmost
argumentation."
never exhaustedthe ways people could think about the situationsproposed. The
dynamic was frequentlyextensive and informative. The following methodologi-
cal principlesare very general.However,especiallythe firsttwo haveparticu-
larly apt application to uncovering p-prims.
Principleof invariance. Thisis a generalprinciplethat,if onegetsa descrip-

tion right, the p-prim (or any theoretical construct)will apply in all implicated
contexts. So if a p-primappearsto be used in situationsin which it is not evident-
ly applicablefor us as theorists, some redescriptionof the p-prim may be in or-
der. Similarly, if a p-prim is not observed to be used in a context in which it
should be, given its currentdescription, problems in the p-prim descriptionare
indicated.The principleof invarianceis particularlyapt and strongfor the present
theory because of the importanceand difficulty of getting the basic description
of p-prims correct. Strong evidence is provided when one can invent situations
in which the predictedapplicationof a p-primleads to surprisingbehavior (e.g.,
predicting"misconceptions").See the principleof discrepancy,discussedshortly.
Principleof diverse evidence. This is relatedto, but moregeneralthan,

the principle of invariance. Different problems in which a p-prim is used trian-
gulate on its properties. This principle is sometimes difficult to apply to p-prims
because the diversity of the knowledge system makes it difficult to craft new sit-
uations that cleanly implicate a target p-prim.
and
Principleof redescription. In sparseknowledgesystems,it is important
difficult to get the descriptive frame right. Commonsensevocabularyand intui-
tively ready characterizationsseldom suffice. Thus, tuning and competitive ar-
gumentationconcerningmultipledescriptionsof a p-primcan optimizecoherence
with other principles.
Principleof scavenging data. The humansense of physicalmechanism

is hypothesizedto apply to almost all familiarand unfamiliarphysical contexts.
Although they might not be ideal for analysis, the predictionsand explanations
made by people for almost any situationare relevantto the theory. There is no
need for highly refined and idiosyncraticexperimentalsetups. Thus, data from
large numbers of reported experiments, if reinterpreted,can contributeto the
analysisof p-prims. For example, datafrom many of Piaget'sbooks, before rein-
terpretationinto his theoreticalframework,can providevaluablehelp. Also, many
misconceptionsand alternativeconceptionsstudieshave been useful, such as those
of McCloskey (1983a, 1983b), Roncatoand Rumiati(1986), Viennot (1979), and
Clement(1982, 1987). Again, this is not a data-impoverishedenterprise,and use
of others' data highlights the importanceof the theoretical frame.
Principleof discrepancy. Evidentlywhenpeoplegive nonphysicsexpla-

nations or show nonphysicsexpectations, there is a good opportunityto uncover
explanatoryroots in p-prims. This is basicallythe methodologicalprincipledefin-
ing misconceptionswork. However, the principle of discrepancydoes not over-
ride the principles of obviousness, unproblematicgenesis, and functionality;
p-prims must cover ordinarycases, must be plausibly abstractedfrom available
experience, and must be useful to individuals.
Theprincipleof content over form. P-primsare content-based

analyses.
This view of human causality implicitly denies that it lies in some small set of
universalforms, in analogies, or in the applicationof mappingsjudged by purely
structuralcriteria (Gentner, 1983). Thus, p-prims cannot be removed from an
analysis in favor of general processes such as analogicalreasoning. This perhaps
contentiousprincipleanticipatesa longer discussion in the InterpretiveSummary
section that compares this work with others'.
ELEMENTS
The Basic Forceand MotionCluster

The range of phenomenacovered by the intuitive sense of mechanismis not ob-
viously that of any scientific theory, for example, Newtonian physics. Nonethe-
less, it seems thatintuitiveknowledgeis relativelyrich in the vicinity of Newton's
laws. This richness may well account for the robustnessthat shows up in stu-
dents'difficultiesin changingtheir intuitivephysics to accordwith textbookphys-
ics. The first subsectionthat follows deals with what seems to be the core of this
richness. The p-prims involved are adequateto account for many well-known
misconceptionsconcerningthe dynamics of objects in trajectories,as I describe
especially in the InterpretiveSummary. This first cluster prominentlyfeatures
intuitiveattributeshavingto do with an individual'spersonaland kinestheticsense
of agency.
The second subsectionthat follows deals with situationsthat are handled, by
and large, withoutinvokingagency. Instead,constraintsposed by geometriccon-
figurationof objects have direct causal implications.This turnsout to be the root
of an instructionalproblemexplored in the Development section. In sketch, stu-
dents invoke a strong agentive causality in learning physics, but that causality
mustbe modifiedin two importantways: (a) It mustbe extendedto cover phenome-
na that are seen initially as nonagentive, and (b) the agent-patientrelationmust
be symmetrized to effectively encode Newton's third law.
The third and fourth subsections that follow deal with balance and equilibri-
um. These involve a still more abstractand mathematicalcluster of p-primsthat
concern the causal implicationsof symmetry and asymmetry. As does the sub-
126 DISESSA
section on constraints,these two subsectionsconcernthe implicationsof the naive

sense of mechanism in many situationsthat are not handled in the misconcep-
tions literature.
In this and following sections, I use italics to denote p-prims. I systematically
note, parenthetically,observationsthat relateto the various types of systematici-
ty listed at the end of the Theory section. It is convenient also to extend some
remarkshere briefly to related points about development. More extensive com-
mentson systematicityand developmentare reservedfor theirown sections, which
follow this one.
Forceand AgencyElements
Ohm'sp-prim. I beginwitha prominentandrichlyconnectedelementthat
I call Ohm'sp-prim. It comprises the following subentities:an agent that is the
locus of an impetus thatacts againsta resistance to producesome sort of result.
The major function of this element is to provide for activation(SystematicityA,
mutualuse) of a set of qualitativerelationshipsamong differentialsin the effort
of the agent (amountof impetus), the resistance, and the result: More effort im-
plies more result; more resistanceimplies less result; and so on. The assumption
is that activationof Ohm'sp-prim attachedto particularcircumstanceallows one
to use these qualitativeproportionalitiesas needed for predictionandexplanation.
Ohm'sp-primcan be abstractedfrom any numberof physicalexperiences,such
as pushingobjects.It servesthe fundamentalfunctionsof (a) allowingone to modu-
late appropriatelyone's effort and (b) explaining why that modulationis needed
and effective. One pushes harderto move heavy objects, which "resist"motion
more. It is plausible that this p-prim evolved out of and extends the usefulness
of completely inarticulatecapabilitiesto respondappropriatelyto tasks requiring
differentamountsof exertion. Thus, Ohm'sp-prim may be a common abstraction
over a broad range of already competent sensorimotor schemata. How much
knowledge appearsin compiled form and is only later abstractedinto felt-to-be
explanatory schemata is an importantand open question.
Ohm'sp-prim also seems to interpretintellectualand interpersonalrelations,
such as trying harderand influencing, in additionto directly physical situations.
In these latter cases, as in cases in which Ohm'sp-prim is invoked to serve ex-
planatorypurposes, purely sensorimotorschemataare inadequateto accountfor
familiar human competences.
A situationfabricatedto elicit Ohm'sp-prim clearly is the case of a vacuum
cleaner that is switched on, and then its intake nozzle is covered. Subjects are
asked whetherthe pitch of the motor goes up or down, and why. In the vacuum
cleaner, the motor is a model impetus; indeed, agency in the weak form of an
initiatorof motion is undoubtedlya coordinateattribute,that is, an attributecon-
nected with a high cuing priorityto motors. (Coordinateattributesare a particu-
lar subclassof SystematicityA, mutualuse.) Coveringthe nozzle is an interference
(imposed resistance), which leads some subjectsto predicta reducedresult (cor-

respondingto slower speed and lower pitch); the motor should slow down and
emit a lower pitch.5 An alternateinterpretationof the situationbegins with the
same basic interpretation,but it addsa seeminglyanthropomorphic feedbackloop,
that an increasedresistanceprovokes the motor to work harderto compensate.6
In this case, higher pitch is interpretedas representingthe greater effort. Work-
ing harder is likely an independentlyencoded p-primthat people may recognize
directly in the increase in pitch resulting from the blocking of the hose. Then
Ohm'sp-prim and the anthropomorphictendencyto work harderwhen resistance
is increasedjustify the use of workingharder. Some subjectsuse workingharder
to justify their predictionthatthe vacuumcleaner emits a higherpitch, and some
use it to justify increasedpitch in retrospect,when told that the blocked vacuum
cleaner emits a higher pitch.
Whatis actuallyhappeningis the oppositeof both interpretations.Puttingone's
hand over the nozzle causes the mechanismby which air is pumped, using ener-
gy from the motor, to stop working. (Much less air strikesthe blades, but instead
the air near the blades turns with them.) With less work to do, the motor speeds
up. Thus, this particularinterferenceactuallydecreasesresistance.Readersshould
be aware thatthis explanationamountsto little more than yet anotherapplication
of Ohm'sp-prim, althoughthis one has the propertythat it can be refinedstraight-
forwardly into a legitimate physics explanation.
From a physicist'spoint of view, the intuitiveattributionof resistanceto cover-
ing the nozzle has no justification; a specific mechanism is needed connecting
that act with any force felt by the motor. Intuitively, however, covering the noz-
zle is sufficiently resistance-like to evoke Ohm'sp-prim with that resistance
modulatingthe result of the impetus of the motor. No more specific reliability
loop appears to be needed.
Agency and related notions such as patient and effort seem to comprise a
fundamentalset of attributesat the genetic roots of intuitivephysics (Systematic-
ity B, common attributes).Anthropomorphicand animisticphenomenaprovide
explanationfor many things for many naive physicists, not just children. Work-
ing harder, for example, is seen in the vacuum cleaner by some adults as well
as children. Although agentive phenomenaseem collectively to fade in priority
with developing expertise, still the genetic influence of these ideas is powerful.
5This interpretationprobably forces us to see the impetus as residing in the motor, but not to
coincide with its motion: Motion is a realized impetus. The alternativeis seeing impetus directly
in the motor'smotion. I will not try to settle the questionof which way Ohm'sp-prim applies. Indeed,
it may be that the recognitionof Ohm'sp-prim does not imply that the subjecthas selected an unam-
biguous interpretation.A p-primmay be cued withoutfirm binding of its "slots."Not discriminating
close interpretationsand inabilityto localize attributessuch as agency are typical of the kinds of limi-
tations I posit for intuitive physics. See Appendix A for further discussion.
6In this case, the linking of impetus directly to motion of the motor seems more plausible than
linking it to the hidden cause of that motion.
128 DISESSA
Agency orients thinking about the most fundamentalideas of physics long into
universityinstruction.Particularelements built directly on agentive phenomena,
such as Ohm'sp-prim, find their way into the working vocabulary, if not into
the fundamentalbeliefs of expert physicists. Furthercomments on agency are
found in diSessa (1983), and I returnto the topic later in this monograph,partic-
ularly in the section on development.
Resistance. Spontaneousresistance to forces and influences is a phenome-

non separablefrom the imposed resistance in the vacuum cleaner. The distinc-
tion is in the locus of agency. Spontaneousresistance resides in the patient of
an action, in the mass that is pushed or pulled, not in an external agent that im-
poses an interference.Spontaneousresistance sometimesprovidesthe reasonfor
the returnedpressurefelt when a force is exerted on an object (sometimes called
"reactionforce"). For instance, it explains why a string gets taut when used to
pull an object: The object resists motion. Lighter objects, of course, exert less
resistance and thus allow greater result (faster motion). In this schematization,
two impetuses, a thrustand a resistance, interactcompetitivelyto determinemo-
tion. In a Newtonian schematization,force is modulatedby an entity that does
not have the same agentive status-an inert mass.
Plausibly, resistance's roots are in personal experiences of reacting against
an external agent. If this is the case, however, a fairly early refinementover a
blatantly anthropomorphicagency, reacting to counter a force, is indicated in
some situations;childrenwill often vehementlydeny thatthe wall is pushingback
when they push on it, admittingonly that it is resisting. Adults almost certainly
have both interpretationsavailable:People may resist (actively);objectscan resist
(passively). The difference is that further expectations about the internal state
of the resistor (e.g., its level of effort and that it may cause motion) follow from
active resistancebut not from passive resistance. Surprisingly,some adultsmay
occasionally see very active resistance in inanimateobjects; an object resisting
the tug of a string may leap backwardif the string is cut, as may happen to a
team in a tug of war if the rope breaks. The active-passive distinctionin a patient
seems to be markedin lexical items such as "constrain" (passive)versus "restrain"
(active).
Unlike resistance, many phenomena seem not to have active interpretations
associated with them. Blocking (what a heavy brick does to a hand striking it)
andbouncingimputeno agency, butare kinematic,as it were, describingphenome-
na visually and geometrically. That is to say, these latterphenomenamay be ab-
stracted from situations of observation rather than participation,and one can
maintainthat an object blocks or causes bouncing without any sense of effort
or strain in the object. More on bouncing appears in diSessa (1983). I turn in
earnest to this class of kinematic p-prims in the next subsection.
I believe successful resistance, what a sturdywall offers to a push, is encoded
separatelyfrom the unsuccessful variety that a ball typically exerts in reaction
to a tug on a string. The argumentfor this conclusion is based, in part, on the

general importanceof distinguishingan agent'swinning from losing in situations
of conflict, such as when the tendency toward rest of an object at rest conflicts
with an imposed tendencytowardmotion. LaterI discuss schematasuch as over-
coming, whose existence reinforcesthe idea of such distinctions. More directly,
Talmy (1988) described a large cluster of apparentlyimportantlanguage-based
schematathatcenteron success-nonsuccessjudgments.The implicationof strongly
encoding the distinctionbetween success and nonsuccess situationsis thatpeople
should be in very little danger of confusing the two classes, say, in recalling an
event.7
Force as a mover. Pushingan objectfrom restcauses it to move in the direc-

tion of the push. The p-primabstractedfrom thatbehavior, at thatlevel of detail,
I callforce as a mover.To a physicist,the at-restconditionis an importantprecon-
dition to the alignmentof push and resulting motion. However, it appearsthis
is not true of naive individuals. This leads to various nonphysicalexpectations,
exemplified in the extreme by the universally counterintuitivenature of gyro-
scopes; it is perplexing that gyroscopes do not go in the direction you push
them.8 See also Clement (1982) and Viennot (1979) for misconceptionsinvolv-
ing this primitive. A perspicuoussetting in whichforce as a mover strongly as-
serts itself is when students are given control of a computer-implemented
Newtonian object, a dynaturtle(diSessa, 1982), whose movement they attempt
to control by applying pushes to it. The turtle is Newtonian in the sense that it
obeys Newton's laws, traveling in a straightline with uniform speed when not
pushed and obeying Newton's second law when pushes are applied. The almost
universalexpectationof elementaryschool studentsand of many undergraduates
and naive adults is that the turtle should always move precisely in the direction
of the last push. Instead, pushes actually add to the existing motion to produce
what might be described by a p-prim, deflection (see Figure 2).
It helps to say what is not attendedto in force as a mover. The process of
acquiringspeed is irrelevantto theforce as mover sketch, which prescribesonly
directedinterventionand result. One sees this reflectedin protocolsthatdescribe
a tossed coin, where studentsagonize over the details of change of speed in the
air but accept the toss itself as primitive (see diSessa, 1988).
Expert physicists use force as a mover as a useful shortcutin simple situa-
tions. It would be bizarreto suggest that physicists need to access formal means
71 did not obtain convincing protocol data to supportthe separate encoding of successful and
unsuccessful versions of resistance as opposed, for example, to two separate attributions-one of
resistanceand one of success-nonsuccess. In the same way, active and passive versions of resistance
might be combinationsof independentattributions.Betterdataand argumentare needed to settle the
issue.
8It is more accurateto say a gyroscope does not turn in the directionyou twist it. See force as
a spinner, discussed shortly.
130 DISESSA
O O
o oo
F F
FORCEAS A MOVER FORCEAS DEFLECTOR
FIGURE 2 Alternativemodels of a shove's effect on a moving body.
to predict that an object, when pushed from rest, moves in the direction of the
push. The difference between novice and expert use is that the expert priority
system "knows"much better when to and when not to use that intuition.
Ohm'sp-prim applies to the relationsbetween the agent'seffort and its result
in situationswhereforce as a mover applies. More push means more resulting
motion (greater speed or more distance). Here, the resistance slot in Ohm'sp-
prim is occupiedby the spontaneousresistance remarkedon earlier and is usual-
ly taken, effectively, to be proportionalto the weight of the object.
Force as a spinner. Young childrenoften know thatpushingan object off-

center causes it to spin. This phenomenonis, in many respects, the rotational
version of force as a mover. The cuing priorityof force as a spinner is greatly
heightened in cases of circular symmetryor by other suggestions that the issue
is one of circularmotion.As is truewith most p-prims,contextdetermineswhether
this or competingp-primsare applied, not articulableapplicabilityconditionsor
general methodsof conflict resolution. In the case of the yo-yo shown in Figure
3, most subjectswill simply see thatthe issue is one of spinning, althoughthere
is nothing particularlycircular about a pull to the right. They will predict that
the yo-yo will spin counterclockwiseand, hence, that it will roll off to the left.
A few will admit that, because the object is being pulled to the right, it might
move to the right-an apparentlysimple applicationof force as a mover, which
....
FIGURE 3 Force as a spinner predicts the yo-yo spins counterclockwiseand rolls to the
left. It actually rolls to the right.
provides a correct prediction in this instance. Almost no one who senses both
possibilities has any way to decide which applies. Again, p-prims are relatively
primitive;there are, in general, no methodsor more reliableknowledgeelements
to decide conflicts.
Continuous push. A p-primrelatedtoforce as a moveris continuouspush,

abstractedfrom, say, continuouslypushing a cup across a table. The important
features are a caused motion and a persistent intention (i.e., force) that causes
it. The engine in an automobilecruising along is interpretedas the cause of mo-
tion in this way. (If there is no directionalityattributedto the push of the engine,
this situationprobablyreduces to Ohm'sp-prim.) In a Newtonian view, neither
the handpushingthe cup nor the engine is an ongoing cause for the motion. (The
hand and engine are supplyingenergy but not motion in any direct way. The ef-
fect of forces is, at best, acceleration, not velocity.) Motion perpetuatingitself
with no interventionis the physical notion of momentum,and, in fact, Newton's
laws requirethat there be no net force on an object if it is traveling at a uniform
velocity. Again, Ohm'sp-priminterpretsthe relationshipbetweenpush and result-
ing motion in a continuouspush.
Several importantpoints about p-prims can be illustratedin the comparison
of force as a mover and continuouspush. They are both linked with high cuing
priority to Ohm'sp-prim. Beyond similar connectedness in the knowledge sys-
tem, these p-prims are obviously very similar in internalstructure:(a) They in-
volve the same physicalconfigurationsof agency, intentionor effort, andpatient,
and (b) the mental representationof the intention and the result may be made
in the same geometricvocabulary.These similaritiesmay be representedin shared
abstractionsof, in the first case, realized interventiveintentionand, in the sec-
ond, a particulargeometric directedness, the direction of intention and result.
Possibly, the two p-primseven have a single common abstraction(Systematicity
F, abstraction),a scenario of directedintentionand its realization. Richer areas
of intuitive physics can be expected to be redundantin this way.
The distinctionbetweenforce as a mover and continuousforce is one of pat-
tern of effort. Force as a mover, having been abstractedfrom throwingor boost-
ing situations, involves a burst of effort, then relaxationtypical of a shove or
throw. (The term push is somewhatambiguouswith respect to patternof impe-
tus; shove is not so much so. Strike indicatesa very brief [violent] shove.) Con-
tinuouspush, obviously, involves relativelyconstantamplitude.Whatis implicated
generally is a vocabulary of time-dependentamplitudessuch as those that ath-
letes must be particularlyadept at recognizing and reproducingto tune their ef-
forts. Althoughpersonaleffort is the most obviousplace fromwhichthese patterns
may be abstracted,they or relatedpatternsmust play an importantrole in inter-
pretingspeech (prosody),in dramatictensionin plays andstories, and, even more,
in music. Consideran orchestraconductorexpressingpatternsof volume through
musculartension and physical actions such as punchingor gentle waves, actions
132 DISESSA
that characteristicallyinvolve certainpatternsof amplitude.It is more than that

these acts demonstratethe patternbut that, as they become conventionalized,the
patternexpressed is reliably cued as a coordinate attributeof the acts.
Rapid change of intensity, especially with high peaks, is a patternthat one
might call violence. Aristotle took this to be a fundamentaldistinguishingattri-
bute in his physics, separatingviolent or forced actions from naturalones. An
object thrown (force as a mover)is very likely to be seen as having been impart-
ed a force thatwill carry it along on its own. An object carriedalong (continuous
force) is in a passive relationshipto its motion, receiving it on an ongoing basis
from its carrier. McCloskey (1983a) documentedthatcarriedand thrownobjects
are perceived differently. In the InterpretiveSummary, I return to place this
phenomenon more carefully in the present frame.
It goes without saying that patternsof amplitudeplay a minimal explanatory
role in pure Newtonian mechanics;thus, a developmentis requiredfor expertise
that results in common interpretationsof naively disparatesituations. Any mov-
ing object must come to be perceived as having momentum, whether or not it
moves as a result of violence.
Violence cues strength (the potential for great impetus) with high priority.
Strengthis an attributethatseems to play an importantrole in early thinkingabout
the physical world. It also evidently has an interestingspontaneousdevelopmen-
tal history in view of the fact that adults have a much more articulatednotion
of strengththan children. For example, we find in Piaget that children some-
times appealto weight to explain both the sinking of dense objects and the float-
ing of big ones like ships. In the latter case, I infer an intermediary:Big things
are strong (strong is a coordinateattributeto big), and strong things may do as
they wish in conflict situations,as when gravity "wants"them to sink. In another
situation,considerthe response of my younger son at age 6 in reactingto a query
aboutwhy a magnetcan attractmetalthroughpaper:"Iguess the magnetis stronger
thanthe paper."(In relationto such explanations,see also the discussion of sim-
ple conflict schematain the discussion of dynamicbalance, which follows, and
later on in the discussion of overcoming.)
To hint at some of the developmentthat must take place in changing strength
from a global attributethat explains much about what happens in any situation
of conflict to a much more restrictedone, consider my other son's spontaneous
comment (also at approximately6 years of age): "I figured out why machines
are strongerthan people. It's because machines are metal, and people are plas-
tic." Indeed, strengthis often not a primitiveexplanationfor overcomingor other
conflict resolutionsin the adult sense of mechanism. Instead, conflict resolution
may depend on particularaspects of the conflicting situation (e.g., geometric
aspects such as leverage or positioning) or on intrinsic characteristicsof the ac-
tors (e.g., the strengthof the materialsof which they are composed as opposed
to their agentive strengthor power). Corroborationof the importanceof strength
and generally simple resolutionsof conflict such as overcomingis offered by lin-
guistic structure(Talmy, 1988).
Dyingaway/warmingup. Everyonerecognizesthephenomenon
thatearthly
motion essentially always dies away. Although it can be explained with notions
such as friction and dissipation, dying away is often taken intuitively as a primi-
tive. This p-prim is essentially the stipulationthat a certainpatternof amplitude
(gradualdiminuendo)is naturalfor a particularclass of amplitudes(actions by
inanimateobjectsthatare not subjectto continuousinfluence).Novice adultsoften
treatdying away as a relative primitive. That is, they will often be satisfied with
an explanationthat does not have any particularcause for the dying away. But
if their attentionis drawn specifically to the issue of dying away, they may seek
causes such as the interferenceof gravity or friction. This responsivenessto the
need for deeperexplanationexhibitsa gradientin theirsense of mechanism(toward
more reliable descriptions), albeit not a steep or reliable one.
The situationof bringingan object up to speed, like a car accelerating, seems
to be the occasion for abstractinga warmingup primitive-that it takes some time
for any result quantityto reach its final value when a change in impetus takes
place. Put more succinctly, change takes time. The warming up to speed in a
toss is precisely the phase acceptedas unproblematicin a toss. Althoughweaker
than motion dies away, warmingup is sometimes applied in inappropriatecon-
texts such as the change of acceleration given a change in force on an object.
Some studentsbelieve the accelerationthata force causes in an object, especially
if it is very rapid, continues for a time after the force ceases. To a physicist,
change of force is instantly realized in a change of acceleration.
ConstraintPhenomena
For a physicist,bouncing(a moving objectimpingeson a fixed one andrebounds),
blocking (an object's tendency toward motion is thwartedby another object in
its path), and similar constraintphenomenasuch as supporting(blocking, where
gravity is supplyingthe thwarteddownwardimpetus)or guiding9(such as a tube
does to a ball moving inside) must be reducedto the action of forces. They can-
not be appealedto as primitiveexplanations.Naively, these phenomenaare known
to happen in their respective circumstancesand need no explanationat all, let
alone an explanationin terms of a force-like interaction. When these expecta-
tions are questioned, naive adultsappealto justificationsthat sound more logical
(essentially reductioad absurdum)than mechanistic:"Itwould be absurdfor the
ball to penetratethe wall of the tube."Withinthe naive sense of mechanism,im-
penetrabilitycan be taken to be a mechanicalexplanation,but it is not primitive
in a physicist's world view.
9The use of standardlexical terms to describe p-prims may invite the misinterpretationthat p-
prims are the conventionalmeaningsof these terms. In general, p-primsare a good deal more specif-
ic than word meanings, possibly closer to individual senses of words. Readers should treat p-prim
names simply as mnemoniclabels for particularabstractscenariosratherthantakingnaturallanguage
meanings too seriously.
134 DISESSA
Clamping(holdingfixed) shows plainlythe need for complexreasoningin order

to reduce many common phenomenato a physicist'slevel of fundamentalexpla-
nation. Withinthe naive sense of mechanism,the idea that an object sandwiched
between two opposing forces is held stably in place is primitive. For a physicist,
the existence of two opposing forces does not ensure lack of motion. Zero net
force, in fact, ensuresthatany existing motionwill continue. Even if the clamped
object happensto start out at rest, the naively assumed stability of the situation
requires furthermechanism for a physicist. That is, if a small force is applied
to the clamped object, it remains at rest only in virtue of the clamping object's
developing a new net force to cancel the applied perturbation.To explain how
it is that those new forces arise requiresan attributionof increaseddeformation
to part of the clamping object.
Given the complexity of the reductionto force-like terms and the unification
of a diverse set of phenomenathat reductionrequires, it should not be at all sur-
prising that underminingthe perceived primitivityof p-prims such as clamping,
guiding, and blocking is difficult and takes time. Clement (1987) and Minstrell
(1982), for example, documentedthe misconceptionsmanifestationof this sys-
tematicdifficulty. In diSessa (1983), I notedthatthe reductionuniformlyinvolves
the development of a high priority for the phenomenon of springiness, which
replacesnaive rigiditynearthe peakof reliabilityin the physicist'ssense of mecha-
nism. Springiness'o(deformationwith consequentdevelopmentof restoringforce)
plays the crucial role of generatingthe forces that explain all mechanicalcon-
straintphenomena. Yet even springiness is a nonprimitiveelement for a physi-
cist: It has a reductionto higherprioritynotions.I brieflydescribethe development
of springinesshere in anticipationof a fuller treatmentof central difficulties in
learning Newtonian mechanics (see the Agency in Action and Reaction subsec-
tion of Development) in which springiness has a particularrole.
The inherent springiness of all matter is almost never directly discussed in
textbooks. This is likely due, in part, to the fact that it is largely unnecessary
for problem solving. Instead, studentscan (and often do) settle for magical new
primitiveslike normalforces, which are the reactionof supportingobjects to the
pressure of supportedobjects on them. A second reason that springiness is not
deliberatelydiscussed is that it does not belong to the core notions of Newtonian
mechanics. Even thoughit is centralto overcomingkey difficulties in converting
the naive sense of mechanismto the expert's, it is consideredpedagogicallyunin-
teresting.
Despite the fact that it is seldom taught, most of my freshmen did develop
the springypoint of view. This suggests that expert intuitionsmight develop for
reasonsof increasingcoherenceof the knowledgesystem (SystematicityD, mutual
l'There is probablya basic p-prim at the root of the concept of springiness. I discuss this later
as the spring scale primitive. It is fairly easy, however, to discuss springiness with students. Thus,
it is likely springiness also belongs to the category of more elaboratedand conscious concepts.
plausibility)even though such intuitionsmay not be instructedor even have any

direct instrumentalrole in problem solving. Considerthat normalforces, which
are explicitly instructed,may be explainedand reducedto more core Newtonian
concepts via springiness. Normal forces are simply a conventionalname for the
springy force thatdevelops on a surfacewhen an object is pushedagainstit. Stu-
dents may see the redundancyof the concept of normal force subtly in instruc-
tors' explanations, or they may discover the reduction techniques and thus
understandthe unifying and simplifying properties of a springy point of view
on their own. Similarly, clampingand guiding become examples of springiness's
implications,lending a broadercoherenceto the system. Developmentvia mutu-
al plausibilitygoes beyond the view that competence at solving problems estab-
lishes the sole goal and means for developing expertise. The force of a coherent
point of view, although not as strong as we might like, should not be ignored.
However it develops, the reasoningthatreducesconstraintphenomenato more
centralelements of physics has a strong stabilizingeffect in increasingthe priori-
ty of the more centralnotions and reducingthatof the less centralones. Further-
more, it allows reliable checks on the application of the midlevel knowledge
elements, like normalforces, when necessary. Thus, even if it has no direct role
in routineproblem solving (if one uses normal forces properly, there is no need
to reduce them to more fundamentalideas), learning the reductionvia springi-
ness may be vital for the perceived coherence, stability, and reliability of the
knowledge system.
Balanceand Equilibrium
This section sketches an apparentlyrich and importantclass of p-prims having
to do with balance, equilibrium, imbalance, and overcoming. Again and again,
one hears novices explaining situationsby things being "inbalance"or a system
"returningto equilibrium."
Dynamic balance. Dynamic balancing is likely abstractedfrom situations

in which two opposing forces "try"to achieve mutuallyexclusive resultsbut hap-
pen to cancel each other out. A paradigmaticcase is two people pushing against
each other with equal force, accomplishing nothing. Canceling itself is an im-
portantnaive p-prim (see diSessa, 1982). Cancelingis likely a common abstrac-
tion for many cases of joint, althoughnot necessarily simultaneous,application
of "equaland opposite"tendencies. For a situationinvolving dynamicbalancing,
canceling justifies lack of result.
Dynamic balancing entails agents in interactionand conflict as importantele-
ments in the cuing priority. Recall that the case of a person pushing on a wall
is not, in the naive sense, an instance of dynamic balancing. For the pushed-
upon wall, there is only one agent, the person, and there is no motion simply
because the wall resists the push on it.
"136 DISESSA
One frequentuse of dynamic balancing at novice stages of physics problem

solving occursin explainingcircularmotion. It is saidthatforce pullingtowardthe
center of the circle-gravity in the case of an orbit or string tension in the case
of a rock on a string-"balances out"the centrifugalforce, which pulls the rock
or satellite outward." Centrifugalforce is sometimesjustified as the "resistance"
to the tug of gravity.
In view of the acceptance of balancing forces as an explanationof circular
motion, apparentlycircular motion is taken to be a relative primitive. Novices
who declare"centrifugalbalancescentripetal"do not ask furtherwhat else causes
the circularmotion, as opposed to straight-linemotion, which would be the actu-
al result of balancedforces, or as opposedto no motion at all. The relativeprimi-
tiveness of circular motion is not hard to understand.The world offers many
examples of circular motion where no evident agency is needed to maintain
the motion (e.g., parts of a spinningwheel). If it were not the case that physics
trainingincreasesthe salience of centripetalforces (causingworry aboutthe lack
of motion toward the center), one could just startout with the end stance in the
previous argument:Circularmotion sometimesjust happens. Note furtherthat
clamping might increase confidence that the pair of forces, inward gravita-
tional and outwardcentrifugal, "lock"the object in orbit. Some subjectsverbal-
ize "locking in."
Symmetry,such as havingan obviouscenter,encouragesnovices to thinkabout
circularmotionproblemsso as to view circularmotionas primitive.Manynovices
who propose dynamicbalancing to explain circular motion have great difficul-
ties explainingless naturalellipticalorbits,especiallyif they know or if it is pointed
out to them that the center of attractionis not at the center of the ellipse. Note
here that visual patterns,as opposed to reasoningabout mechanicalor dynamic
causality, play a role in judgmentsof plausibility. Such a role for visual patterns
is a case in which the naive sense of mechanismattendsto substantiallydifferent
attributesthan a Newtonian sense of mechanism. In contrast, whereas F = ma
"The balancing view of circular motion seems more common after the early stages of physics
instruction.A more naiveunderstandinginvolves a force pushing"around," tangentto the circle (White,
1981). I suspect there are coordinatedchanges (SystematicityD, mutual plausibility) that account
for this developmentalpattern.Instructionnames and sanctionsthe force towardthe center in circu-
lar motion situations. Thereafter, the need for centrifugalforce to avoid the simpleforce as mover
prediction (that the object should just move toward the center) is stronger. When centrifugalforce
is questioned, students sometimes explain that, without it, the orbiting object would simply move
towardthe center. The evolution of the concept of centrifugalforce should make an interestingcase
study, because this and other observations suggest it is, at least in part, an unintendedartifact of
instruction.Although centrifugalforce is not a Newtonian concept and, indeed, is explicitly denied
in most contemporarytextbooks, the intuitivelesson studentslearn on their own is thatit is necessary
to make sense of circularmotion. Centrifugalforce seems to be a very robustconcept. Studentsare
frequentlyadamantabout its existence, and popularpresentationsof science (e.g., in the press) use
the concept extensively.
andforce as a mover are technically at odds, they share in many circumstances

a basically mechanicalcausal focus on contact intervention.I continue the dis-
cussion of visual patterns in the Systematicity section.
Gravitationalorbits and ball-on-a-stringsituationsare sometimes seen differ-
ently from what is a physics-identicalmotion, a ball in a circular tube. In the
latter, the guiding primitive and lack of any overt force towardthe center make
the counteringcentrifugalforce also unnecessary, and it may be denied in these
circumstancesby students who affirm it in other situations.
Abstract balance. Consider the following problem: A monkey hangs on

a rope looped over a pulley with a weight of the same mass as the monkeyhooked
to the opposite end of the rope. Initially, the monkey and weight are balanced,
stationary,and at equal height. If the monkey startsto climb the rope, whathap-
pens to the weight?
The correct answer is that the weight rises exactly parallel with the monkey,
at exactly the same speed. The interestingpoint is that studentswho get the an-
swer correctoften appealspuriouslyto conservationof energy to justify the equal
reaction of the weight. In fact, conservation of energy in no way justifies the
spatiallybalancedsolution to the problem. Equalweights can balanceat unequal
heights. My interpretationof this novice explanationinvolves a p-prim, abstract
balancing, which matchesoff amountsin a pair and attachesa balancingimpera-
tive to the two. Abstractbalancing is an assertion of requiredcorrespondence,
a weak form of identity. The imperativenatureof abstract balancing contrasts
with dynamicbalancing: Abstractlybalancingthings shouldor must balance;dy-
namic balancing is balancingby accident or by conspiracy. Abstractbalancing
is a partial, but appropriate,interpretationof algebraic equality.
Abstractbalancingis perhapsthe centralschemein "worth,"as when one asserts
so manyeggs are worthso manydollars.It may be abstractedfrommanysituations
of similarcorrespondenceand serves its role in assertinga requiredcorrespond-
ence, probablywithoutany prejudiceas to the mode of establishingor enforcing
the correspondence(e.g., tradingeggs anddollarsor direct,universalcorrespond-
ences suchas one findsin mathematicalor physicallaws). Piagetianconservation-
numberor amountas a propertynecessarilypreservedthroughtime and physical
manipulationof sets-probably involves this primitive. Similarly, conservation
of energy should appropriateabstract balancing as a distributedencoding, lead-
ing to a generally high priority use of abstract balancing in physical analyses.
So, in the monkey problem, seeing an initially balanced situation, studentsuse
the imperativenatureof abstractbalancing to justify a predictionof maintaining
symmetry (balance). They announce conservationof energy as the underlying
principle because it is an overtly sanctionedversion of abstract balancing.
Conservationof energy is misapplied in the case of the monkey and weight
for a fundamentalreason. Conservationonly applies to the same thing at two
138 DISESSA
different times, not relatedthings in two differentplaces. Novices who use con-
servation of energy as a justification in the case of the monkey evidently have
not linked the "transformation-through-time" perspective adequatelyto the use
of energy. They happilyappealto conservationwhen only their intuitivepercep-
tion of abstract balancing actually applies. It is precisely this partiallearningof
the notionof conservationof energythatsuggestswe musttake a knowledgestruc-
ture such as abstract balancing seriously. If complex notions are graduallyas-
sembled out of more primitive elements, different developmentalstates should
be characterizedby partialassembly. That, in turn, should reveal the pieces of
the whole.
An idiosyncratic example of an inappropriateimportationof balancing into
understandingphysics involved a studentwho constantlypuzzled me by drawing
force vectors on free-body diagramsin the opposite direction from real forces.
The reason for this behaviorbecame apparentwhen she openly declaredthat the
meaningof F = ma for her was that"naturerequiredeverythingto be in balance,"
and the ma was the thing that balanced (apparentlyin the sense of dynamic
balancing-canceling out-but with the imperativeattributeof abstract balanc-
ing) any force. Thus, for any force, she frequentlyidentified the balancingma
force as necessarily existing. This example is valuablein showing how, contrary
to the misconceptions point of view of widely held systematic difficulties, the
present view allows us to understandindividual, althoughsometimes personally
pervasive, constructions as attempts to universalize combinations of common
primitives into a more systematic sense of mechanism.
Imbalanceand Re-Equilibration
Balance is one side of a coin, and imbalanceis the other. Imbalance,in intuitive
physics, has two manifestations.First, in cases of accidentalbalancing(dynamic
balance), there are specific related happeningswhen the balance dissolves. In
the case of abstract balance, the imperativenatureof the balance can be inter-
preted as a tendency ratherthan a strict principle of equality. The balance may
be perturbedby an externalintervention,althoughit is not dissolved by it. Then,
new p-prims describe both the interactionof the perturbationwith the balancing
tendency and what happens when the perturbationis withdrawn.
Dynamic imbalance. Dynamic balancing alerts (SystematicityA, mutual

use) a primitiveI call overcoming,which is whathappenswhen one force is greater
thanthe otherand, perhapsgraduallyat first, "getsits way."This way of describing
overcoming is meant to suggest what I take to be the anthropomorphicroots of
the abstraction.Overcomingis a ratherprimitive scenario for the interactionof
forces or influences in that it prescribesthat one of the influences wins over the
others, and the winner's"intendedresult"is achieved. This works ratherwell for
forces that are directly opposed, but it fails to capture nuances such as either
combining or compromisingamong forces that are not directly opposed. Cate-

gorical predictionsof overcoming, when one or anotherinfluence gets its way,
seem particularlywidespread. This may be surprisingbecause a likely domain
of applicationis social interaction,where complex negotiationscan result from
conflict. But, nonetheless, simple overcomingseems to dominateearly explana-
tions andpredictions.See, for example,the late appearanceof compromisepredic-
tions with a dynaturtle(diSessa, 1982), andalso note thatAristotle'sphysics never
provided any principle for combining influences such as nonaligned forces.
Consideranotherexample of conflict and categoricalovercoming. A frequent
prediction about the trajectoryof an object thrown horizontally over a cliff is
that the horizontalimpetuslasts for a while (graduallydying away), then gravity
"takesover," leading to a near right angle turn. It is importantto add that this
kind of simplificationof interactionsis only relativelystrong, not strictlyadhered
to. In many cases, subjectspromptedto inspect the right angles they have drawn
immediatelypatchthem by smoothingthe corners. Most novices sense that sharp
turnsare very atypicalin situationsof continuousphysical motion, althoughthey
may need to be promptedto cue this reaction.
An exemplar of continuousforce, say, an object being pushed at a constant
speed across a table, is frequently interpretedat a refined level as the push of
the hand overcoming the resistance of the friction of the table. As compelling
as this is, a Newtonian interpretationhas no imbalance. The force of the hand
is exactly equal to the resistance of friction, and an initially supplied motion
(momentum)is left to perpetuateitself. Because it is so familiar and involves
central intuitive schematizations,coming to understandthis simple situationis
surprisinglydifficult, as I elaborate later in the Development section.
The following example of overcoming highlights a function for intuitive
knowledge in physics novices that I believe is very common. Intuitiveschemata
can provide a top-level analysis that serves as a plan that is elaboratedin rather
straightforwardfashion into a complete problem solution. Considera mass rest-
ing on top of a spring. How far must you push the mass down in order to have
the mass jump off the spring when you release it? A novice sees simple opposi-
tion between the force of gravity and the force of the spring pushing upward.
Gravity wants to hold the mass down, and the force of the spring is trying to
toss the mass into the air. If the spring is pushingany amountmore thanthe (dy-
namic) balancingpoint, it will assume the role of the stronger, overcome gravi-
ty, and toss the mass into the air. (This analysis is nearlyan exact quotationfrom
some novice protocols.) At this point the solution is plannedin terms of balanc-
ing and overcomingschemata,and all that needs to be done is to express the dy-
namic balancing point algebraically, which novices do by equating the force
exerted by the spring (kx, where k is the spring constantand x is the displace-
ment from equilibrium)with the force of gravity (mg, where m is the mass and
g is gravitationalacceleration).To a physicist, this equationmerely specifies the
amountof compressionneededto have the springsupportthe mass. Compression
140 DISESSA
a bit more thanthis specified value will result in the mass bouncingup and down
upon release but will not have it tossed into the air.
Abstract imbalance. Elements connected with high cuing priority (mutual

use) to abstract balancing are generally very different from ones connected to
dynamicbalancing. "Outof balance"can mean some perturbingforce is interfer-
ing, but the abstractbalancing tendency can still be in effect. Consider a pan
balancethatis sustaininga temporarydownwardpush from a finger on one side.
The pan is said to be out of balance, but the tendency to balance is still in ef-
fect.'2 After all, one may feel the pressure of the pan "trying to return to
equilibrium."More generally, in cases of abstractbalance, one typicallygets dis-
placement from equilibriumby an amountin proportionto the perturbingforce
ratherthan overcoming. Displacementfrom equilibriumproportionalto the per-
turbationis a p-prim I call generalized springiness, which applies outside of
genuinely spring-likesituations.In contrastto the intuitiveschematizationof im-
balance modulated by generalized springiness, a physicist would informally
describe the out-of-balancebalance scale as a new balance. More formally, it
is described by opposing forces that happen to be equal.
Push a tall block or package of cereal standing on a table toward the side,
and it tilts more until it falls. Because naive physicists assimilate the push and
subsequenttilt to generalized springiness, I suspect many would be surprisedto
discover that, in this case, the greater the tilt, the less hard you are pushing!
Suppose one takes a pair of equal weights and hooks them to the ends of a
ropethatrunsthrougha pulley. The weightsare initiallybalancedat equalheights,
as in the previousexamplewith the monkeyandweight. If one addsanotherweight
to one side, novices frequentlypredictthatthe now heavierside will move down-
ward a distanceproportionalto the weight added. In actuality,that addedweight
will cause the rope and weights to accelerate, allowing the heavier weight to fall
until the lighter weight hits the pulley. (That is to say, dynamic imbalanceand
overcomingwould be a betterinterpretation of this situation.)RoncatoandRumiati
(1986) explained the naive predictionby asserting an intuitive theory of balanc-
ing of potential energy. Indeed, if the heavier weight goes down a distance in
proportionto the perturbingweight, the potentialenergyon one side mightbalance
the potential energy on the other. But according to the present interpretation,
balancing potential energy is a spurious and misleading interpretationof what
novices are doing. They are, instead, using generalizedspringinessto determine
the extentof disequilibriumratherthandeterminingthe positionof a new equilibri-
um by matchingenergies.13 Besides connecting this situationto a broaderrange
12Itis plausible that abstract balancing, ratherthan being ambiguousabout whetherthe balance
is always exact (as in conservation), really comprises a numberof distinct p-prims, some of which
involve strict equality, others of which involve only a tendency.
13Balancingpotentialenergy has anotheruntowardproperty. If we denote the smaller weight by
m, the other weight by M = m + i, the height of the m side by h, and the difference in heights
of phenomenathroughabstractbalancingandgeneralizedspringiness,the present
interpretationavoids attributingmisconceived"theories"involving instructedcon-
cepts to studentswho may have no intuitivefeel for or any real conceptualgrasp
of potential energy.
The fact that abstract balancing can be interpretedas a tendency ratherthan
a strictbalancingmeans it can be temporarilyviolated withoutthe constantexer-
tion of a perturbingforce. Withoutthe perturbingforce, however, things return
to equilibrium,and equilibrationis the operationalprimitive. In contrast,as not-
ed before, physicistsknow stableequilibriamustbe the resultof a relativelycom-
plex mechanism; any return to equilibriumhas a deeper cause.
Equilibrationmay have presentationsthatare very differentfroma panbalance.
An absence or sparseness of material next to an abundanceleads, primitively,
to flow and re-equilibration.The space left by a scoop taken out of a body of
sand or water is refilled. Equilibrationhere can serve as a replacementfor more
mechanisticexplanations,for instance, thatthere are forces that cause the return
to equilibrium. See the discussion of sucking that follows.
In some cases, the returnto equilibriumof a disequilibratedsituationis speci-
fied in more detailedprimitivesthan generic equilibration.Two patternsof mo-
tion are takenas typical of the returnto equilibrium.One is a bobbingmovement
of diminishingamplitude, such as the oscillation of a balance scale or the slosh-
ing of a disturbedpan of water. The other is a nonoscillatingmovement, a stead-
ily slowing movement toward stop at equilibrium, such as a car (with working
shock absorbers)pushed down and then released.
Interestingly,dampedbobbing and the simpler slowing returnto equilibrium
often appearwith very different priority in naive subjects comparedwith more
expert ones. One of my MIT interview subjects, respondingto the question of
how far a mass must be pushed down on a spring in order to have the spring
toss the mass in the air, took the problemas indicatinga simple returnto equilibri-
um. She assumed the latter pattern(no oscillation) as the way of things and re-
fused to do the problemas stated.A physicistcan scarcelylook at a springwithout
thinkingoscillation. This studentnot only denied oscillationbut also maintained
thatthe brick could never be thrown off the spring at all. It would just be pushed
graduallyback up to equilibrium.In other situationsas well, she never spontane-
ously suggested oscillation as a possible motion with springs. Whatis particular-
ly strikingis that this studentdeclared that throwing the brick down might give
it enough "energyto rebound"off the spring. Evidently, thinking of throwing
the brick down evoked a set of p-prims different from that evoked by push and
release. Throwing the brick down (violence?) caused her to see the situationas
a bounce. The stopped state half way througha bounce, which physicists would
for the new "equilibrium"by d, then equalizing potential energy implies the following: mgh =
(m + i)g(h - d). So ih = Md. This equationshows that the displacement,d, dependson the height,
h, at which the apparatusis placed. If you do the experimenton top of a hill, you get a result that
differs from doing it at the base. I doubt even naive physicists would find this congenial.
142 DISESSA
see as identical to the state of an object compressed by hand and about to be

released, is not salient in her abstractionof bouncing.14 Instead of the stopped
state half way through the bounce, apparentlyshe saw something continuously
happeningthat a physicist does not see going on (or does not see as explanatory),
namely, bouncing. That makes the situationdifferent for her from just releasing
the brick from rest.
Fundamentally,the physicist sees the world in terms that do not admit that
being engaged in a process such as bouncingis in any deep explanatoryway part
of "whatis going on" in a system. The rich vocabularyof circumstance-specific
processes of intuitive physics, which often explains situationsin which they oc-
cur, defers in the physicist explanationsto universaldescriptionsof changebased
on state involving only things such as geometry (reflecting, e.g., state of defor-
mation) and straight-line velocities. The emergence of deference of common
phenomenato a sparserset of explanatoryterms is an importantand general pat-
tern that I continue to elaborate later.
DEVELOPMENT
I have sketchedplausiblecontexts for the abstractionof p-prims and commented

on some issues of developmentto the degree these have been immediatelyuseful
in understandingthe natureof the p-primslisted so far. However, largerdevelop-
mental issues need to be addressed. These issues have mostly to do with reor-
ganization. New p-prims are undoubtedly generated to accommodate natural
conceptual development and physics instruction, but this seems relatively un-
problematic.What is more interestingis the restructuringof the system of exist-
ing elements. Following the theory sketch, I describe these changes as shifting
priorities, which may gradually relocate a knowledge element within the
knowledge system.
An insightful way of looking at the shifting priorities of knowledge elements
is to describe the shifts in terms of the changing functions that elements have.
For example, some p-prims might serve transientroles, primarilyto cue the use
of others, whereas some are more fundamentalin that they are broadly applica-
ble andtreatedas explanatory.Learningphysics, I conjecture,requiressystematic
modification from the naive state in the following ways. One starts with a very
shallow explanatorysystem: many p-prims serving as essentially primitive ex-
planationsfor various phenomena. Gradually, p-prims cluster and become or-
ganized as distributedencodings. Recall that distributedencoding refers to the
reuse of intuitive knowledge elements as parts of expertise, such as aspects of
14Recentdata involving children indicate that a stopped state not only is not salient in a situation
of reversal but also may be rejected as implausible when suggested (diSessa, Hammer, Sherin, &
Kolpakowski, 1991).
more sanctionedphysical ideas, special cases, and approximateversions of laws

and principles.The humanencodingof Newton'slaws of motionmustbe assumed
to be very complex comparedwith simple knowledge elements such as p-prims.
Notions of agency get built into the concept of force. P-prims such as force as
a mover operateas special cases for a physicist (pushingfrom rest). Springiness
of matterbecomes more salient than rigidity, and it mediates the reduction of
constraintphenomenasuch as blocking to more centralNewtonianconcepts such
as force. In addition, balancing ideas become refined to play an importantbut
partial role in encoding laws such as conservationof energy. Overall, p-prims
become an essential part of what makes F = ma work in a physicist's head.
The fundamentalchange in structureis that, insteadof a very broadand shal-
low explanatorysystem, whateverp-primsare still used must defer to or become
partof the complex but few subsystemsthatare the encoding of the physicallaws
themselves. Instead of a slew of p-prims, only physical laws are explanatorily
primitive at the highest levels of reliability.'
In this section, I provide more, and more elaborate,examples of p-primstak-
ing on several importantroles in expertise. Aside from serving as part of the
encoding of the laws per se, p-primsmay serve as "legitimizedphenomenology,"
where, like springiness, a phenomenonis known independentlyto happen yet
can itself be explained and connectedto the highest reliabilityelements in virtue
of circumstantialparticulars.16 Other roles include cuing more reliable analyses
(e.g., sensations of abstract balancing cuing conservation)and providing plan-
ning knowledge (more successful versions of the kind of thing demonstratedby
the use of the overcoming schema as a sketch of a solution to the compressed
spring problem discussed at the end of the previous section).
I consider the learning-developmentalhistories of three physics conceptuali-
zations. First, as a brief introduction,I look at sucking, its relation to p-prims
and their explanatorystatus. Second, I consider an extended example of the de-
veloping relationshipbetween a cluster of p-prims and an element of textbook
physics: the simple harmonicoscillator. Finally, I take a look at a particularpeda-
gogical problem, learningNewton'sthirdlaw, which entails a more refined look
at the developmentof the crucial phenomenologyof agency. These case studies
of conceptualdevelopmentextend the list of p-prims in the preceding section a
bit, but, more important,they provide (a) commentson how clusters of elements
interactin particularlines of conceptualdevelopment, (b) discussion of some im-
portantgeneral roles p-primsmay play in development, and (c) some apparently
15Lawsare, of course, not primitive in other than an explanatory sense. They have empirical
supportin the form of evidence. They have logical supportin terms of definitionsof terms, but they
do not have explanatorysupport from more fundamentalprinciples.
161assume it is not harmfulto propose that legitimized phenomenologycan always be reduced,
explanatorily, to "official"principles. This likely catches the general flavor of priority relations in
any case, but it would be interestingif independentworld assumptionsgot built into theories by vir-
tue of the phenomenology that accompanies expertise.
144 DISESSA
general developmentalpatternsand at least a general chronology for some par-

ticular developments.
Sucking
Many of the patternsof developmentthatI believe are importantand typical can
be illustratedin the simple example of sucking, as in sucking water through a
straw. Initially the overt agent, the person sucking on the straw, is viewed (via
an early form of sucking p-prim) as the direct cause of the liquid's motion. At
an intermediatestage, a more elaboratecausal chain is envisaged: Sucking re-
moves or otherwise rarefiesair; then the partialvacuumpulls the liquid. Sucking
is no longer the direct explanatoryanalysis of what happens when you suck on
a straw, but the motion of the liquid is explainedby a general fact that everyone
knows: Vacuums draw gas and other substancesinto them. This is the level of
pop physics.
Laterin development, at the level of a physicist'sinformaldescription, an in-
visible, inanimateagent (i.e., air pressurepushingdown on the liquidin the glass)
becomes the direct cause of motion; it pushes the water up the straw. This hap-
pens when the balance of pressures (pressureon the water in the glass balances
pressureon the top of the liquid inside the straw) is brokenby the overt agent-
the person sucking. It is notable that, at this stage, in contrastto earlier stages,
a vacuum or partial vacuum can in principle never pull (suck) at all. Sucking
is entirely gone as an explanatoryprimitive. In a bit more detail, the new analy-
sis is: The personchangesthe geometryof the situation,for example,by increasing
the volume inside his or her mouth in which a fixed amount of air exists. This
results in decreased pressure on the liquid on the straw-and-mouthside of the
system. A decreased pressure on the inside of the straw and an undiminished
atmosphericpressure on the other side cause the water to move.
Actuallyexpertsknow they must furtherexplainthe propagationof differences
in pressure in the air and in the liquid so that each moving element of substance
experiences the force needed to move it. This is a level of detail that is usually
ignored, because it is known to be a legitimizedpiece of phenomenology.In most
instances, one can simply treatobjects, even things like chunksof liquid, as rigid
objects, ignoring the internal deformationsthat are necessary to produce local
forces that move individualpieces of the chunk. Indeed, even this explanation
omits details in that it employs a version of force as mover-unbalanced forces
cause motion. Instead, unbalancedforces develop accelerations, the legitimate
Newtonian quantitythat responds to force.17 Furthermore,unless prompted, a
physicist would almost certainly use some version of guiding or "channeling"
implicitly to avoid having to explain exactly how the water knows to follow "the
suspect it would be a doable but not a trivial challenge for most physicists to explain correct-
"171
ly why there is no apparentacceleration, given a difference in pressures.
obvious"path into and up the straw. It would take solving field equationswith
boundaryconditions to say exactly how water flows from various places in the
glass into and up the straw. Although such subtle, usually suppressedconsidera-
tions give rise to observablephenomena(e.g., propagationdelays) and give rise
to questionsthatwould be impossible immediatelyto answer (e.g., how the flow
toward the straw "inlet"in the glass proceeds), no physicist would say he or she
does not understandhow sucking on a straw causes the water to move, even if
he or she has never done a detailedanalysis. Instead, one can remainat an intui-
tive but legitimized level of analysis in giving explanations,because one knows
roughly how the story of propagatingpressuredifferences, viscous friction, and
so on, works out so that those explanationsare respectable.
To summarize,learningphysics involves reducingthe explanatorypriorityof
many p-prims, such as sucking, thatprescribedirectconnectionsbetween a class
of actions and resultingmotions. Instead,the Newtonianexplanationsof motion
are channeledinto a single, complex notion (i.e., force), which requireselaborate
situation-specificreasoningto explain familiarphenomena.Despite acknowledg-
ing, if pressed, the need to rely only on basic principles, physicists still use ideas
such as force as a mover, rigid object assumptions, and guiding or channeling
to provide high-level descriptions that avoid many complexities in reducing a
phenomenonto basic causalrelations.The fundamentalreorganizational phenome-
non exhibited here is that p-prims become subordinatedto more core concepts
and principles, yet some selected ones are preservedas legitimized phenomenol-
ogy, so that everyday phenomenado not escape explanationthatis judged New-
tonianwithoutcarryingout all the situation-specificreasoningnecessaryactually
to establish an explicitly Newtonian explanation.
Bells:TowardExpertIntuition
This exampledeals with intuitiveknowledgeelementsthateventuallyplay an im-
portant role in a large and complex piece of expert knowledge. The expert
knowledge is much like what Kuhn called "exemplars"(Kuhn, 1977, p. 298),
"concreteproblem solutions, acceptedby the group [in this case, physicists] as,
in a quite usual sense, paradigmatic."Collecting and systematicallyattachingp-
primsas distributedencodingsfor physicalprinciplesis a structuralandknowledge-
based view of the process thatKuhnidentifiedas centralto learninga discipline,
the process by which studentslearn to see the exemplaroutside its initial context
while problem solving. P-prims interpolate between the world's diverse and
familiar presentations and the highly schematized abstractions of sanctioned
physics.
The particularexemplarat issue here is called the simple harmonicoscillator,
a standardproblemworkedout at greatlengthin every relativelycompletecollege-
level treatmentof mechanics. The simple harmonicoscillator is consideredby
physicists to be fundamentalto a large class of phenomena.Before considering
146 DISESSA
it in detail, let us first look at the intuitive knowledge that genetically predates
that notion.
One of the Montessorieducationalmaterialsis a set of musicalbells with vary-
ing pitches but all of the same size and shape. How do these apparentlyidentical
bells, made of the same material, manage to have different pitches?'8I have
posed this questionto a large numberof peopleat variouslevels of physicssophisti-
cation in addition to the studentsin my formal interviewing sample. Most cor-
rectly conclude the bells must vary in thickness. But the question remains as to
whether the thicker bells are lower in pitch or higher.
Almost uniformly, people answerthatthe thickerbells mustbe lower in pitch.
Interestingly,this is trueeven of those people who do not know thatpitchis related
to frequencyof vibration. Some make analogies to various musical examples of
the principle: xylophones, chimes, or organ pipes, where the bigger the piece,
the lower the pitch. But even those who do not spontaneouslymentionexamples-
indeed, they may have difficulty producingan example when requested-know
thatbigger things have lower pitch. A few subjects, presumablythe more physi-
cally sophisticatedones, produce the chain of implicationsas follows: Thicker
means heavier, which implies slower (lower pitch). But most seem unableto go
beyond "bigger implies lower pitch" as a primitive phenomenon.
The second phenomenonin this chain, that"heavierimplies slower," is a rein-
forcing or justifying primitive that could be abstractedfrom shaking or pushing
objects of differing mass. Note that here, as with force as a mover, there is a
preconditionthat is not abstractedinto the p-prim. In this case, the precondition
is equal force applied to the differing objects. Omitted preconditions, whether
because of systematicinabilityto encode them or because they seem (or actually
are, in most circumstances)irrelevant,is an importantdevelopmentalphenome-
non. Furtherexamples of omitted preconditionsare discussed later and also in
Appendix B, in the section on children'sp-prims.
Note that slow in heavier implies slower is an attributethat applies to slowly
moving objects (speed) and also to slowly vibrating ones (frequency). This is
typical of differentiationsthat do not seem to be useful for intuitive conceptuali-
zation but become more and more importantin physics instruction.Lack of such
differentiationlays the cognitive groundfor conflict. It is possible that an object
vibrates faster but moves more slowly. Some studies exist that show the lack of
differentiationand potentialconflict thatmay arise between versions of slow and
fast in the case of angular speed (frequency of revolution) versus linear speed
(see Levin & Simons, 1986).
A note on abstraction. It is worth a momentto consider the genesis of the

bigger implieslowerpitchp-prim.In particular,one mightquestionits phenomeno-
logical roots. Is this simply an abstractionof a phenomenon, a straightforward
18Thisproblem was first presented to me by Jeanne Bamberger.
schematizationof an event as I characterizedthe generic abstractionprocess for

p-prims?It seems more likely to arise froma comparisonof two perceptions(com-
paring a low-sounding object to a smaller, higher sounding one), thus abstract-
ing the p-primmight imply reasoningbeyond simply describing. The abstraction
process might involve a reflective event, perhapsa conscious search for reasons
that the quality of two events differs.
One can make two moves at this point. The first is to accede to the heuristic
nature of the sketch I made in the preliminarytheory sketch of the abstraction
process as a "minimaland unproblematicabstraction."Sometimes it might hap-
pen in different, possibly more complex, ways. For example, the abstractionof
a p-prim might be the residue of a fairly complex process of reasoning, akin to
Piagetian reflective abstraction(e.g., Piaget, 1967/1971a), say, comparingtwo
situationsandidentifyingan apparentlycausaldifferencethatexplainsotherdiffer-
ences. More than not being ruled out a priori, it makes good sense that the ab-
straction process should be complex in certain instances. Indeed, I have said
nothing aboutthe process that selects particularparts of an event for abstraction
into a p-prim. There must be mechanisms at work in this selection that extend
and refine an individual'scurrentsense of mechanism, for example, determining
what aspects of the event are most useful to incorporateinto causal descriptions.
Expandingthe inquiry concerning mechanisms involved in generatingp-prims
changes little that I say here, because most of what I report is independentof
the abstractionprocess.
Withoutdenying that the genesis of p-prims might be complex, it is possible
to make a more conservative move in this particularcase. We can depend on
the fact that relative features such as bigger and lower seem to have primitive
precursorsas absolute ratherthan as relative attributesin children'sconceptual
development (e.g., Piaget, 1970/1972, p. 30). In situationswhere the absolute
forms of the attributesare salient, a single abstractionof an event is all that is
needed. For example, "(Big)Daddy'svoice is low." Juxtapositionof big and small
objects might serve, via generic processes, to enhance the salience of absolute
big and small attributesby contrast, thus promotingthe incorporationof these
into the abstraction.
Considering absolute attributesas precursorsfor relative ones suggests that
the intellectualdevelopmentthat enables one to deal with relative quantitiesin-
volves schemes that patch together reasoning involving absolute attributesinto
functionallyrelative reasoning. Absolute associationscould be extendedto rela-
tive ones by a genericcapabilityto thinkrelatively,when necessary,or by concept-
specific changes thathave the same effect. This would parallelthe way I propose
naive p-prims are patched together to produce behavior that avoids the pitfalls
of the naive primitives through an augmented system of priorities.
However it is that qualitativecorrelations such as the expectation of lower
pitch for bigger objects come to exist, adults seem to have many items of the
form "the more x, the more y" in their sense of mechanism, as the list that is
148 DISESSA
developed later attests (stiffer implies less motion, heavier means slower, short-
er distancemeans less time). Qualitativeproportionalitieshave been incorporat-
ed into several researchers'models of qualitativereasoning(e.g., Forbus, 1984;
Roschelle, 1991).
A very small proportionof interviewees explain the thickness-pitch correla-
tion by remarkingthatthickermust mean stiffer, which in turn must mean faster
(higherpitch). The latterpartof this phenomenologicalsyllogism mightbe drawn
from the experience of muscle tension causing or being needed to cause rapid
shaking. Indeed, stiffer might really appear as "moretense" in the p-prim, ab-
stracteddirectly from muscle tension. Another possibility is abstractingfrom a
musical situation, say, tighter strings on a guitar making a higher pitch. Some
subjectsindicatea thirdand slightly more complex process of genesis in another
syllogism: Stiffer things do not move very much (in more technical language,
small amplitudevibration is associated with stiffer things), and because that is
true, they can accomplishtheir motion in less time, again implying a higher fre-
quency. The less distancemeansfaster primitiveis typicalof primitiveswith omit-
ted preconditionsthat seem to be behind children'spuzzling responses to time,
rate, and distance problems posed by Piaget (1946/1971b). For example, some
children seem to believe that going faster entails more time (even in cases of
constant distance traveled), which possibly results from a phenomenological
syllogism- faster is associatedwith farther,but farthermeans longer time-that
can be constructedonly with omitted preconditions.
It appearscertainthatmost people have bigger (or heavier)implieslowerpitch
as both a more salientand a more reliablep-primthanstifferimplieshigherpitch.
The latter seems to occur mainly as a "virtualconcept,"the conclusion of an on-
line phenomenological syllogism.
There is potentiallya good deal of mutualplausibilitywork going on in these
phenomenological syllogisms: Based on a given vocabulary, situation-specific
reasoning derives or rederives another element that might then be separately
remembered. In these cases, it seems likely that deriving a previously stored
phenomenonshould increasethe priorityof the elements of the derivation,espe-
cially if the derivationis by novel means. (Again, derivationmust be taken ad-
visably.) By the same token, drawing an unlikely conclusion is as likely to
underminethe reasoning or presuppositionsas it is to change the belief system
directly in a dramaticway (i.e., encoding the unlikely conclusion).
Let us turn to expert thoughtand to the exemplar of the simple harmonicos-
cillator. The underlying model consists of a perfect spring attachedto a rigid
supporton one side and to a particleon the other. The derived behaviorincludes
the fact that the frequency of oscillation of such a device is proportionalto the
square root of k/m, where k is the spring constant characterizingthe stiffness
of the spring, and m is the mass of the particle. My claim is that experts, when
they need this informationqualitatively, use intuitionsof precisely this form-
heavier implies slower and stiffer impliesfaster-but that their confident use is
generallyrestrictedto the specificcontextof a simpleharmonicoscillator.That

is, expertsattachintuitionsof the samekindas (andmaybeeven identicalto)
thosethatnoviceshaveto morespecificcontexts.Theexpert'sconfidencein the
use of the intuitionsis linkedto havingan elaborateknowledgesystemthatcan
validatethe simpleharmonicoscillatorcontextin anyparticularcase, andit is
linkedto beingableto justifythe qualitativeresultswithhigh-reliability notions
via a morecarefulderivationif necessary.
Wecanmodelanexpertanalysisof thebellproblemcrudelyas follows.Sound
andfrequencyissues,andpossiblyotherrecognizedconfigurations suchas struck
objects,cue vibrationas a problemcontext.Vibrationcues simpleharmonicos-
cillatorknowledge,becausethatis the expert'sfundamental modelof vibratory
behavior.The expertreasonsthat"thethickerbell will be moremassive,so it
willtendto vibratemoreslowly;buttheextrathicknesswillmakeit stiffer;hence,
vibrationwilltendto speedup."Of course,thisreasoning mightconsistminimally
of only threementalevents,the recognitionof the two qualitativecorrelations
in the perceivedsituation,andthe recognitionof a conflictbetweentheirsug-
gestedconclusions.Thereafter,theexpertmustreasonwithgreaterprecisionto
resolvetheconflictingtendencies.It turnsoutthatstiffnesswinsthecompetition
becausestiffnessincreasesas thesquareof thickness,whereasmassonlyincreases
linearly.Thatis why the thickerbells have a higherpitch.19
Theexpertmayneedto checkthatthesituationis trulylikea simpleharmonic
oscillatorvia reliabilityloops. Butthe fundamental pointis thathe or she may
use
confidently essentially the same intuitionsthatnaivepeoplehave,butrestricted
to a specificclass of situationsthatcan bothbe elaboratelyspecifiedandthat
allowjustificationof the intuitionif necessary.
Inrefiningthismodel,it seemsas likelythatanexpertmightmakethequalita-
tive reasoningsbasedon recognitionof vibrationalone,withoutany mediating
cuingof the simpleharmonicoscillatornotion,exceptin post hocjustification
of his or herintuitions.Indeed,we shouldquestionthe very meaningof "cuing
the harmonicoscillator."An exemplaris muchtoo largea knowledgestructure
to be activatedin anycontext-freesense, the way I supposep-primsare. More
likely, somesubpartsof "harmonic oscillator"maybe cued,suchas a subvocali-
zationof its nameor possiblyactivationof an appropriate germof the particle-
on-spring model. Other parts of the exemplar would be activatedaccordingto
need andcircumstance.
It is worthemphasizingthataddinga complexjustificationcontextsuch as
an exemplarto the knowledgesystemdoes not meanthatit is alwaysinvoked.
Indeed,for the purposeof rapidlypursuingpossibilitiesratherthanlaboriously
19Aslightly differentexpert analysis might observe thatthe wavelengthof the fundamentalmode
of the bell remains roughly invariantas the bell is made thicker. Because wavelength determines
frequencyvia speedof motion(speeddividedby frequencyis wavelength),the questionof pitchreduces
to what happensto the speed of sound in the bell. That, in turn, reduces to the same ratio of mass
to stiffness considered in the text.
150 DISESSA
pursuingnecessities,one can expect expertsto developthe abilityto judge heuristi-

cally the applicabilityof the intuitive componentsof the exemplar according to
features of the context. To cite an extreme case mentioned previously, no ex-
perts would hesitate to use force as a mover in a situationof pushing from rest,
and they would not, under most conditions, need to resort to a law or principle
to justify that prediction. In this case, something similar to a naive at-rest attri-
bute could do the trickof judging applicability.In othercases, new p-primsmight
be specifically developed for the purpose of making such judgments. These p-
prims would be legitimatepartsof an expert'ssense of mechanism,althoughthey
might be quite specific. I am suggesting here that it is implausiblethat experts
would avoid using surface featuresin encoding dispositionsfor problemsolving.
See Chi, Feltovich, and Glaser (1981), Roschelle and Greeno (1987), and Smith
et al. (in press) with regardto the plausibility of experts using surface features
as part of their expertise.
In addition to adding a complex justification context to the basic intuitions,
it is clear that a general shift in priorities has occurred, particularlyin the case
of stifferimpliesfaster comparedwith heavier (bigger)impliesslower. The former
notion is at best marginallyrecognizableto naive people. The curriculumof naive
physics seems to teach heavier implies slower much better than stiffer implies
higher pitch. Understandingwhy this is so makes a good case study, which I
briefly pursue, of the circumstances that explain relative priorities of various
p-prims.
A principalcontributionto the fact thatheavier implies slower is more salient
thanstifferimplieshigherpitch is thatthe intuitiveattributeheavieris more salient
than stiffer (SystematicityB, common vocabulary). As noted in diSessa (1983),
stiffness seems not to be well elaboratedin naive senses of mechanism, tending
to be perceived in categories such as (absolutely) rigid or squishy, like clay.20
More particular,the relative stiffness of objects that make musical sounds, such
as Montessoribells, is completely imperceptibleto naive subjects. Certainlywe
have better access to sensing variations in mass through kinesthetics. We have
even better access to patternsthat are spatially describable, such as "moremo-
tion," because of our highly developed visual capabilities. This would lead to
an increasedpriorityfor spatiallydescribablep-prims,even if the motioninvolved
is too small to be directly perceived. Indeed, stiffer implies higherpitch is likely
to be derivedindirectlyfrom the spatialintermediaryof stifferimpliesless motion.
To add to the differentialbetween heavier implies slower and stiffer implies
higher, it seems an indisputablefact thatit is more important(and also, thankful-
ly, easier) for people to control motion through modulatingforce and reacting
to mass thanto controlpitchby modulatingstiffness. Functionalityand instrumen-
tal access select one p-primas more importantthanthe other. All in all, it should
20Theintuitive phenomenologyrelated to materialproperties-rigidity, squishiness, brittleness,

and so on-would seem to make a rich and interesting study.
be no surprisethat heavier (or bigger) implies slower has naively much higher
priority than stiffer impliesfaster.
To summarize, the simple harmonicoscillator shows several developmental
changes in the sense of mechanismthat are instigatedby learningphysics. Some
naively recognizable phenomena(heavier implies slower) become appropriated
as distributedencodings of physically sanctionedideas. In complementaryman-
ner, an exemplarsuch as the simple harmonicoscillatormay serve intuitiveideas
as a complexjustificationcontext. New p-primsor p-primswith greatlyincreased
priority, such as stiffer impliesfaster, also come to play importantroles. These
may be of naively low priority for reasons such as their involving low-priority
attributes(stiffness) or simply because they are not importantin everyday life.
I noted that many p-primsinvolve particularqualitativeproportionalitiesamong
intuitive attributes:More x begets more y.
The discussion touched on several general developmentalphenomena. Many
early p-primscontainomittedpreconditionscomparedwith more expertschemes.
Some differentiation(e.g., slow speed vs. slow frequency) is also implicatedin
development. I also noted examples of a fairly general mechanismof coherence
generation,phenomenologicalsyllogisms, in which two p-primsthatare roughly
of the form "AttributeA implies AttributeB" and "AttributeB implies Attribute
C" may lead to the encoding of a p-prim- "AttributeA implies AttributeC." Fi-
nally, I discussed reasons to believe that, sometimes at least, the genesis of p-
prims might be more complex than a simple abstract schematization of a
phenomenon.
Agencyin Actionand Reaction

Agency is a crucial attributein the development of many aspects of cognition.
It is likely that developing some sense of personal agency is one of the first im-
portantlearning tasks for babies. Piaget and his collaboratorsfocused much ef-
fort on this topic and on the subsequent equally importantlearning task of
discovering how and when we come to attributeagency to aspects of the world
outside our selves. Various forms of overextension may be seen as anthropo-
morphic or animistic mistakes. On the other hand, these extensions of agency
to the world are also insights about how, in some way, agency is a good model
for many physical regularities.I offer here a few commentson some of the ways
agency appearsto contributeto and to develop as part of the physical sense of
mechanism. This section follows up on the introductionof many p-primsin the
Elements section that used agency as part of their encoding. It explains some of
the broader developmentalissues surroundingagency.
I have argued(diSessa, 1980) thata particularform of agentivecausalitycom-
mon in intuitive physics comes to dominateearly encounterswith school phys-
ics. Schematically, the basic causal syntax includes an agent (an initiatoror an
impetus-containingobject), a particulardirected and legitimized kind of action
152 DISESSA
on apatient,whichis anobjectin whosebehaviortheeffectof theagent'saction

is at issue. "I[animateagent]hit [legitimizedinteraction] theball [patientwhose
resultantbehavioris at issue],andit moved[theresultof thecausalinteraction]."
"Johnpushedthe rock (andit thereforemovedbecauseof the pushor did not
'for otherreasons')."21
The causalsyntaxis a primaryinterpretiveframeworkfor F = ma. Some
agentis responsiblefor exertinga force, whichhas the effectof acceleratinga
mass.Newtonianmechanicsneednotbe interpreted in thisway. In fact,thisin-
is a
terpretation effectively pedagogical decisionto provideanoverarching frame-
workfor interpreting Newtonianmechanicsin termsof ideasprovidedby the
naivesense of mechanism(diSessa,1980). Thereare otherways to makethat
transition,otherchoicesof intuitiveframesto importintoa Newtonianmechan-
ics, thatprovidean initialintuitiveaccessibility.All of theseframes,of course,
needrefinement; noneof themcanproperlybe identifiedas aspectsof Newtoni-
an mechanicsthatwe knowbeforeschooling.
Oneof the implicationsof legitimizingcausalsyntaxis thatagentandpatient
continueto be stronglydifferentiated in earlylearningof Newtonianmechanics.
The effect of a forceon the patientis far moresalient(becauseit is the focus
of the syntax)thanreturneffectson the agentof the interaction.Indeed,the ef-
fects on the agentof the interactionare interpreted as anotherindependent ac-
tion, this time with the initial agent as the patient.22Thus, an unavoidable
symmetryof Newtonianmechanics,Newton'sthirdlaw, is expressednotdirect-
ly, butby addinga mandatory butinverselydirectedasymmetrical relationship:
"Forevery action,thereis an equaland oppositereaction."
This patch-assertinga mandatoryreaction-representsa relativelygood
pedagogicalmove, but it entailsa host of relateddifficulties.These all stem
fromthe fact thattop-downcoherence(Systematicity C) takesa long time to
achieve.Studentsfirstlearn"actionandreaction" largelyas a slogan.Themany
coordinations neededto effectivelyencodethethirdlaw only graduallyfall into
place.
The firstdifficultyis withpathscuingreactionandwiththe persistence(ap-
propriatereliability)of thatcuing. Studentstakea long time beforethey come
to applythe reactionpatchinstinctively.Reactionsdo not becomesalientand
arenotfeltto be necessarymerelyby virtueof the factthatthebookandinstruc-
tor say theyexist. Actionscontinueto be seenas directed,andit takesconstant
promptingby instructorsto considerthe reaction.
"Actionsandreactions" maybe interpreted naivelyin termsof violence,either
as a cue or as a justificationfor the reactionof the patient.If violencebecomes
21SeeTalmy (1988) for anotherschematizationof this set of relationsin a surprisingcontext (i.e.,
grammaticalstructure).In Talmy'sforce dynamics,the agentor locus of controlis entitledantagonist;
the patient, defined by the locus of interest in terms of effect, is called the agonist.
220ne can see the depth of the problemin that even our naturalvocabularyseems to pick direct-
ed actions as the basic unit, with mutual actions described as pairs or clusters of inter-actions.
partof thecuingof reactions,thenwhengradualeffectsareat issue,theremight

seemto be no needfor reaction.The gravitational pull of the earthcausingthe
fallingof a ball might not seem to occasion need for a reactionfromthe earth.
Morelikely, a reactionmightseemnecessaryif the fallingballcrashedintothe
earth(althoughhere, the actionis likely to be interpreted as the blow the ball
effectson theearth).In contrast,if violencebecomespartof thejustificationfor
reaction,theearth,as big as it is, wouldseemunlikelyto respondto aneveryday
objectthathappensto be pulledby the earth.Here,partof the difficultyis also
thatreactionis intuitivelymeasuredovertly,by evidenteffect, and,indeed,the
earthwill nevershowmucheffectin reactionto everydayobjectsfallingnearit.
Theseconddifficultyentailedby usingcausalsyntaxandtheaction-reaction
formulation is thatrecallingtheaction-reaction slogan,or evenbelievingit must
governall interactions, is stillsomedistancefrombeingableto identifythereac-
tionproperly.Otherinverselydirectedactionsaremistakenfor theproperreac-
tion. Forexample,if I pushon anotherperson,his or herinstinctivepushback
may incorrectlybe interpreted as the reaction;or if I pushon a cup, movingit
acrossthe table,the frictionforceof the tablepushingbackon the cup maybe
consideredthereaction.A reaction,in naiveterms,is only"oppositely directed"
and"somehowcausallyrelatedto the action,"whichare insufficientcriteriato
distinguishthe real reactionforce fromothers.
In a physicist'sparlance,the expression"theforceof A on B"selectsa con-
venientasymmetrical pointof viewon whatis fundamentally a symmetrical rela-
tion. If the complementary asymmetric view (the force of B on A) not
is
immediately available with appropriate identities(themagnitude of the forceis
the samein bothcasesandthephysicallocationsof thetwo forcesareidentical),
theknowledgesystemwill haveineffectively encodedthefundamental symmetry.
Thethirddifficultyis thatthenotionof agencymustbe broadened to see many
reactionsas plausible.(Thisis a specialization of the firstdifficulty,withcuing
andreliability,to theroleof agency.)Thefourthdifficultyis thatcompeting naive
schematizations of eventsmustbe undermined to preservetheplausibilityof ac-
tionandreactionacrossa diversityof situations.I illustratethesebothwithex-
amples.
Considerthe transitionto understanding Newton'sthirdlaw in the case of a
cup supportedby a table. A physicistsees thata tablepushesup on a cup so
thatit does not move (i.e., fall down).To a novice, this reactionof the table
to the cup is not plausiblein view of an evidentlack of agencyin the table
(thethirddifficulty).The tableis not a likelyanimator;it exhibitsno senseof
strain.Furthermore, not movingis not a prototypicalresultof a push, and it
seemsentirelyimplausiblethatunderany circumstances (say, gravitywerein-
stantaneously turnedoff) a tablecouldcausethe cup to fly up intothe air. Be-
sides,thissituationis alreadycoveredby an"adequate" naiveschematization (the
fourthdifficulty):Thetableis simplyblockingthe fall of thecup;no furtherex-
planation is necessary.
154 DISESSA
We see clearly in this example that distributedbut coordinatedchanges in the

sense of mechanismare necessaryto encode Newtonianthirdlaw symmetry.Many
coordinate-attributeties must be individually broken, or a way must be found
to effectively weaken them all simultaneously.Among these are expectationof
strain in agents and, at least in some circumstances, the "it will cause motion
if opposing forces are diminished"expectation(overcoming)for force-like inter-
ventions.
A dramaticexample of a number of the just-listed difficulties, centering on
agency, occurred in my interviews when a subject refused to admit that a metal
cabinetcould be said to pull a magnetto it, even to the point of denying she could
feel the pull on her hand when I handedher the magnet and had her bring it near
the cabinet. She said she was just feeling the magnet pull the cabinet. Presum-
ably, she did this because she knows thatmagnetismresides in the magnet;hence,
it must be viewed as the initiatorof magneticinteraction.This is remarkablylike
children who refuse to believe a wall can under any circumstancespush back
on a person, even if they see a protuberancefrom the wall deforming a piece
of clay held in the hand. Clement (1987) also cited the fact that novice physicists
refuse force-providingstatusto passive objectseven when the effects of the force
are evident (e.g., you break your hand in hitting the wall).
Even if agency is attributedto a salient patientand competingp-primsare un-
derminedin some situations, other p-prims in other situationscan still interfere;
the fourth difficulty is a broad one. This is a canonical effect of a broadly dis-
tributedsense of mechanism, and it furthercomplicates the encoding of princi-
ples like the third law. The schemataof balancing and overcomingdeclare that
the cause of motion is typically an imbalance-things move undera push because
they do not resist enough, because the return force is not equal to the applied
force. To watch a cup being pushedacross a table and not see the handovercom-
ing the resistanceof the cup or the frictionmarksquite an advancetowardexper-
tise. Because the balancingassertedin "equaland opposite"may be confusedwith
dynamicbalance, novices may wonder out loud how anythingcan possibly move
if every force is balanced with an opposite one.23
The fifth problem with symmetryby patchingtogether two asymmetrieshas
to do with the long-term viability of the fundamentalconceptual commitment
representedin adopting causal syntax and the action-reaction schematization.
These conceptualizationsplay into naive expectationsthatare, over the long term,
23Someof the difficulties here have become more or less officially recognized by the physics
teachingcommunity(of course, not describedin these terms)and are indirectlyreferencedin physics
texts. The horse-and-cartparadox-that if the horse and cart are exerting equal and opposite effects
on each other, nothingshouldmove-is used to counterstudents'impressionsthataction and reaction
may constitutea dynamicbalance. The elephantparadox-that two elephantspullingon oppositesides
of a rope do not effect more tension in the rope than one elephant pulling on a rope attachedto a
tree-is an occasion for attemptsto get studentsto see a nonagent,the tree, as equivalentto an agent,
another elephant in the tree's place.
underminedby the rest of the Newtonian sense of mechanism. It seems that a

reaction might follow an action, but the distinctionbetween initiator(agent) and
patientlies entirely outside the core Newtonian frame and must graduallycome
to be perceivedin thatway. Initially,comfortableexcuses (SystematicityE, com-
pleteness) thattend to preserve the intuitivecategorizationof action and reaction
must fade. For example, studentsmay say, "Onecan tell the action from the reac-
tion accordingto who acts first."Or, as describedearlier, they may have notions
about what should be considered a naturalagent (e.g., animate things). These
or refinedandextendedversionsmustall lose credenceas one putstogethernewly
learned pieces such as that action and reactionare always simultaneous24 or that
situationsthat appearentirely inanimatemust be interpretedby the same notion
of force.
Althoughthe thirdlaw is an inextricablepartof the Newtoniannotionof force,
it appearsto need special treatmentto become even remotely intuitivelyaccessi-
ble (via pairedasymmetries). Given the systematicshallowness exhibitedby un-
sophisticatedsenses of mechanism, it should not be surprisingthatthe thirdlaw,
even if it is effectively used, may develop its own independentexplanatorypow-
er and not find its properrelationshipin the developingknowledgenetwork.This
is a sixth difficulty. Like many p-primsbefore it, studentswill often be satisfied
with the thirdlaw as a primitiveexplanationfor motion.Why does a rocketmove?
Studentssay, "becauseof action and reaction."Many studentswill resist finding
the force on the rocketthatmakes it move, becausethey interpretactionand reac-
tion as its own causal primitive. The force on the rocket is in realityeasy to iden-
tify: It is that of the hot gasses thatpush againstthe forwardwall but not against
the (absent) back wall of the rocket's engine.
One of the general developmentsthat aids novices in seeing the world in less
directed, less agent-initiatedterms is to shift toward seeing geometry as playing
a causative role. This extends anothercommonsense form of causality-that sit-
uations are sometimes taken to be the initiators of events. One can say, "That
Joe was in such a bind caused him to rethinkhis attitude."Studyingphysics en-
courages studentsto see, "Thatthe cup and table are in a state of mutualcom-
pression implies that there is a force on each of them." Springiness, as in this
example, is part of this broader move. Consider also coming to see a change
in mouth geometry as the instigatingevent in sucking. More generally, gravity
and electrical attractionare effects of the proximity of objects, and, indeed, all
Newtonian forces are functions only of the spatial arrangementof objects. If
geometry is taken as causal, neither of the interacting objects in an action-
reaction pair needs to be seen as the instigating agent. Within the naive sense
of mechanism, the use of geometry and configurations as causal initiators is
lower in priority than causal syntax in mechanical situations. Moving toward
24Anaspectof causal syntaxI have neglectedis the presumptionof time sequencingbetweencause

and effect, whereas, in Newtonian mechanics, there can be no time lag.
156 DISESSA
causativegeometryis a neededsteptowardcompatibilitywith the Newtonianworld

view.
Related work. Brown and Clement (see Brown, 1990; Brown & Clement,
1987, 1989) studiedthe difficultiesstudentshave in coming to see a reactionforce.
Their empirical work shows many studentexplanationsfor why they do not be-
lieve the table can provide a force. These explanationsalign with the interpreta-
tions providedhere (e.g., Brown, 1990), althoughthese researchersdo not suggest
the underlyingreasonsfor theseexplanations.Minstrell(1982)also provideda
long list of studentreactionsto third law situationsthat can be comparedalmost
pointfor pointwiththe list of p-primsanddevelopmental
difficultiescataloged
here. In addition, Brown and Clement'swork on pedagogical interventions(fol-
lowing on earlier work by Minstrell) shows how the model of objects as springy
can effectively attack many but not all of the difficulties described here.25
Springiness (see discussion in the Elements section, p. 134 and following) pro-
vides a model of blocking that is more compatible with Newtonian mechanics
than the naive blocking p-prim. Springiness explains how a table may be seen
more naturallyas agentive. In contrastto the present framework, Clement and
Brownexplainedthe effectivenessof theirinterventionsonly in termsof the gener-
ally positive effects of analogies and of the root springinessintuitionin particu-
lar. They did not reportthe positive influence of p-primssuch as resistance(e.g.,
to a shove) in providing intuitive access to the third law. They did not interpret
the effects of springiness in terms of its competitionwith other p-prims such as
blocking or in the context of the central issue of agency, as sketched here.
To summarize, impartingintuitive accessibility to Newton's laws by building
in a fundamentalreliance on causal syntax has profoundeffects on subsequent
learning. Even to state the third law requirespatchingin an intuitively arbitrary
assertionof symmetry. A relatively global change such as the shift towardseeing
geometry as causative is helpful. However, a host of other patches and refine-
ments is necessary (a) to extend the circumstancesin which the third law will
automaticallyand justifiably be seen to apply (expandingcuing and reliability
paths), (b) to encode aspects of the law that are implicit in its statement(e.g.,
requiredsimultaneityof action and reaction) and to provide for smooth reason-
ing in problemsolving using the law (e.g., appropriateshift of attentionto retrieve
the reactionforce), (c) to refine agency so that its own cuing and reliabilityhelp
25Forexample, they do not identify difficulties that may be provided by alternateschematiza-

tions such as violence, dynamicbalancing, overcoming,and so on. These researchersconsiderneither
the long-term effects of underminingnaive assumptionsof causal syntax and causal sequence nor
other difficulties such as the thirdlaw incorrectlyassumingits own primitiveexplanatoryforce. They
also consider, by and large, only one way of overcoming the broad set of difficulties listed here-
using "bridginganalogies."I have been deliberatelynoncommittalaboutinterventionprocedures.See,
in this regard, diSessa (1980), which suggests a global alternativeto importingcausal syntax into
Newtonian mechanics.
make many reactionforces salient and plausible, (d) to make peace with compet-
ing p-prims (blocking, dynamicbalance, and overcoming)that provide compet-
ing analyses, (e) to come to grips with the fact thatthere is really no Newtonian
distinctionbetween action and reaction, and (f) to keep the law from assuming
its own independentstatusas a causalprimitiveratherthancontributingas a proper
extension of F = ma.
SYSTEMATICITY
I have alreadymade a numberof observationsaboutsystematicity.For example,

I have considered: central elements (Ohm'sp-prim, force as a mover) and less
central elements (wobbling, warming up);26importantattributes(agency) that
may give rise to centralcausal schematizations(causal syntax); classes of recog-
nizablepatterns(amplitudepatternslike violence and dampedbobbing);and even
a few mechanismsfor producinglimited systematicity, such as phenomenologi-
cal syllogisms and slogans (equal and opposite reaction). In this section, I con-
sidera pairof more global systematicitiesin detail. The centralquestionis whether
these systematicitiescan be localized, say, as relatively modularsubsystems or
as being due to the extended and integratedinfluence of a few elements. I argue
that, in some restrictedcases, there is evidence for modularity.Yet overall, the
quality of knowledge systems such as the sense of mechanismmakes it unlikely
that broad and deep systematicitiescan be localized. The questionhas important
pedagogical implications with respect to our expectationsthat narrow, focused
interventionscan achievestrongresults.It is also importantin decidinghow simple
an account of the development of knowledge we can expect to give.
The first systematicityis a static predisposition that seems to skew descrip-
tions and explanationsof dynamic events toward interpretationsinvolving static
phenomena.I presentan extendedset of examplesof this predisposition,because
it seems to capture, in some degree, many misconceptions that have been
cataloged. My analysis of this systematicityis made by proposing an attractive
way of localizingit to a basic categoricalstructure,namely, an ontology of change
and cause. However, careful considerationsuggests thatthis systematicityis, in-
stead, the result of cumulative but relatively isolated effects.
The second systematicityinvolves a class of primitives that are also static in
nature.These have to do with spatial/visualpatternspresentedby dynamicevents.
Figuralprimitives appearto be of a ratherdifferentdescriptiveclass from many
of the dynamicp-primsalreadyconsidered. Because they may also be perceived
as a class by students, and, further, because they relate all in the same way to
26Strictlyspeaking, lists of central and less central elements (high- and low-priority elements)
may explain systematicitiesin databut do not constitute,per se, relationalsystematicitiesamong ele-
ments in the system, as listed in the Theory Sketch section.
158 DISESSA
a Newtonian sense of mechanism(they are essentially irrelevant),figural primi-

tives may behave largely as a group in the evolving sense of mechanism. Thus,
in contrast to the static predisposition, my analysis suggests figural primitives
may constitutea module with significantunit properties, albeit of limited impact
in the net transformationfrom a naive to a Newtonian sense of mechanism.
The Static Predisposition
Railroadcar in space. Considera railroadcarmovingwitha hugeveloci-

ty on a track past a planet as in Figure 4A. Of course, gravity from the planet
attractsthe car and things in it. Suppose there is a mass on a spring inside the
car, and the mass is constrainedfrom the side so that it can only move up and
down. What will a trace of the motion of the mass look like as the car rushes
past the planet?
The most frequently given answer shows a compression of the spring that
is symmetrical about a point of maximum compression nearest the planet (see
Figure 4B). This solution seems to derive from a spring scale primitive-that
squishy or springy structurescompress according to the force or weight put on
them.27This much can be read very directly from some protocols (see the ex-
ample that follows). The spring scale primitive is probably learned with some
version of weight, bigness or massiveness being the controlling attributeof the
compressingobject, but the transferto a force interpretationin the course of phys-
ics instructionappearsto be unproblematic.If asked for a justification for their
initial responses to this problem, studentsoften use Hooke's law, that the force
exerted by a spring is proportionalto compression; because gravitationalforce
varies symmetrically on either side of the planet, symmetrical compression is
predicted. That is to say, there is a garden-pathroute using instructedconcepts
thatjustifies the incorrectintuition. The bug in this justificationis that the force
of gravity is not the same as the force exerted by the spring. If it were, there
would be no motion. Thus subjects seem to be assuming a dynamic equilibrium
between gravity and the spring force.
The correctansweris shown in Figure4C. The springreachesmaximumcom-
pression after the car is past the planet, and in fact it oscillates up and down.
An importantobservationto which we returnshortly is thatthe incorrectanswer
is actually correct in the static limit, when the car is moving arbitrarilyslowly
past the planet.
The incorrectanswer shows remarkablerobustness. Studentsliterallydiscuss
the problem for hours without coming to see the oscillation entailed. None of
the studentsinterviewed on this problem as part of the formal study even men-
tioned oscillation as a possibility during the initial ?/2 hr with the problem. One
27Thisdoes not necessarily entail seeing the springy object as an agent pushing back, as is re-
quired of more sophisticatedversions of springiness.
(A^)
-----------
AAA-r
AAAAAA
AA-^r
^^A^AA^
AAA-r
FIGURE 4 A boxcar in space whizzing past a planet.
might have expected oscillation to be attachedto springs as a default behavioral

attribute,much the same as impetus or intention ("wanting"some result) is at-
tachedto electric motors. This would cause oscillationto occur at least as a weak
hypothesisaboutthe behaviorof the spring, even if it were delayed in considera-
tion or rejectedeventuallyin favor of the spring scale. But it appearsthat oscilla-
tion is much less strongly attachedto springs by novices than by experts. This
empirical fact is consistentwith the proposalmade in diSessa (1983) that the ef-
fective novice version of the fundamentalbehaviorof springsis more like squishi-
ness (large hysteresis) than springiness, so oscillation would be systematically
underrepresentedin answers. Recall the studentwho thoughtaboutthe brick that
was pressed down on a spring and then released in terms of a gradualreturnto
equilibrium-the dampedshock absorberimage. In an attemptto help her think
about the problem dynamically, I literally built an analogous system, a weight
hung from a rubberband. She refused to believe the system would oscillate if
pulledaway from equilibriumbut, instead,maintainedit wouldjust returnsmooth-
ly to equilibrium.
During the course of discussing the railroadcar in space, I uniformly tried
various techniques for cuing a perspective different from the static spring scale
image. Following are some sample probes and typical responses, the latter in
italics and parentheses:
160 DISESSA
Imagine the car is going faster still. Would that make a difference? Can
you motion in the air how the mass will move? (Veryfast hand motions
up and down are typicallyproduced.)Whatis responsiblefor the mass mov-
ing upwardmuch more quickly in thatcase? Isn'tit the spring?Is the spring
getting stronger?(Yes, the spring is whatpushes the mass up. No, it's not
gettingstronger. Themassjust movesupwardfaster because the car is mov-
ing past the planet faster.)
Note how the primitivenessof the spring scale, that it "justhappens,"is here
replacing a physical analysis in terms of force and acceleration, the only legiti-
mateprincipleof (changeof) motion. Althoughadmittingthatit mustbe the force
of the spring that causes return to rest length, that pronouncementof cause is
not systematicallymaintained:Insteadof observing that the spring force would
have to change in order to produce quicker returnto rest length and then con-
sidering the possible cause of that change (there is none!), the studentmaintains
the nonphysicsanalysisby returningto the springscale image. This kindof fluidity
of causal linkages seems typical of intuitive analyses. Again, novices use primi-
tive phenomenain place of analyses based on higher reliability notions. Their
sense of mechanismdoes not requirethe use of the more restrictedset of qualita-
tive primitivesused by experts, even when apparentlyattemptingto check a pro-
posed solution thoroughly. Schemes such as spring scale are cached as likely
solutions at a level of confidence indistinguishablefrom physicallyjustified ex-
planations.
Imagine you, yourself, are the mass on the spring. What do you feel as
far as the force of gravity is concerned? Eventually I convince students,
if they do not see it for themselves, that when the speed of the car is very
high, gravity acts just like a hammerblow to the mass, a downwardforce
that comes and goes very quickly. (Studentsstill maintainthe symmetrical
solution, that after the force of the hammerblow is gone, no matterhow
quickly, the mass must have returnedto an uncompressedstate.)
Note how in this context studentssimply do not perceive the force of gravity
as supplying an impetus, which must then die out or be counteredby the force
of the spring. Intuitivephysics is context fluid (with respect to the way physicists
define contexts), and, in this case, spring scale dominatesand suppressesmore
dynamic reasoning.
Describe to me the motion of the mass if therewere no springat all present.
(The mass continuesmovingdown until it hits thefloor, possibly long after
the car has passed theplanet). Now, what if one inserteda very, very weak
spring?(Reversionto symmetricalprediction.) Isn'tthat discontinuousbe-
havior puzzling? (Sometimesit is perceived as puzzling, but typically not
enough to overcome confidence in the symmetricalsolution.)
A quotationfrom a protocol of a studentwho had, duringabout45 min, been

broughtalong this path of questioningshouldbring home the point. Whatmakes
this protocol particularlystrikingis thatthe subject, D, was an excellent physics
student.He showed no traceof inappropriate force as a moveranalysesand, as far
as I could tell, reasonedflawlessly aboutthe basic effects of force on objects. He
did not haveany troublewith reactionforces. To indicatehis level of sophistication,
when posed the questionaboutwhatwould happenif therewere no spring,he ob-
servedthatthe mass would thenbe in orbitandtravelon a hyperbolicpath(exactly
right).Whenqueriedaboutwhatwould happenwith a very weak spring,he revert-
ed to the symmetricalpredictionbut was troubledwithoutbeing prompted.How
could this minor change of insertinga very weak spring cause such a big change
in the resulting path? After spending about 5 min worrying about this, D said
(Ellipsesdenotepauses;parentheticalremarkshave been addedto clarifycontext.):
You know, I've been explaining it to myself as basically the compression

of the spring is a measure of the force, and so ... almost ... basically
like having a bathroomscale with the mass sitting on it. You would weigh
the most (most gravitationalforce) at the closest approach(to the planet)
... and I think that is an accurateway of looking at it. But, um, I don't
know. I really can'texplain why basically a small change-having a spring
and adding a little bit of tightness to it28would move that point (the point
of most downward motion) so far (from infinity, to near the planet).
D straightforwardlyandconsciously examinedhis reasoning,even to the point

of describingthe primitiveanalysis (bathroomscale) in essentiallythe same terms
I do. Yet he did not retreatfrom that into an analysis in terms of forces and ve-
locities, which I have no doubthe could have done if directly pressed, an analy-
sis which wouldhave derivedoscillation.Instead,he judgedthe p-primsufficiently
reliable to stand on its own, on a par with basic physical law. D ended the 1-hr
interview concluding that he still believed the symmetric solution and thought,
somehow, the discontinuity problem could be solved. A p-prim here simply
replaced, by virtue of context (the springproblem), his knowledge aboutthe de-
tails of force andmotion,knowledgethatwas intactandworkingin otherproblems.
The fact that even this relatively sophisticatednovice could be satisfied with an
analysis on the basis of a relatively noncentralp-prim such as the spring scale
ratherthan moving to more physically central analyses is a strong indicatorof
the continued shallowness of novice explanation.It also indicates the difficulty
of building the deep explanatorynetwork of experts, deep in the sense that ex-
perts can always be more careful, retreatingat need to sanctionedand richly en-
coded notions such as physical quantitiesand laws.
28Wehad been talking about the case of no spring as having an infinitely loose spring and then
increasing its "tightness"just a little bit. D had no difficulties thinking about infinity and limits.
162 DISESSA
It is worthunpackingthe notion of a physicist "beingmore careful."The point

is not that novices do not write down equationsto derive oscillatory behavior;
experts do not do that either in problems such as these. The point is that an ex-
pert'squalitativeanalysiscan be improvedby using intermediatepriorityphenome-
na (distributedencodings), as well as by resorting to the highest levels of the
knowledge system, say, expressed in analytic versions of physical laws. Con-
sider the following two examples of phenomenahaving to do with momentum
that help experts deal with dynamic situationsbetter than novices.
First, momentummay be perceived phenomenologicallyin the tendencyof an
object to keep moving in the face of resistingforces. That"brakingscenario"may
sound trivial, yet it is precisely what the novice mentionedbefore does not see in
the downwardmotion of the weight in the boxcar. The springis supplyinga brak-
ing force to the downwardmotion suppliedby an impulse of gravity. More cases
of ignoring momentumin this phenomenologicalsense will be introducedlater.
The secondcommonsensephenomenonthatturnsout to be more commonsense
for physiciststhannovices involves the case of a moving objectcompressingsome
other elastic, spring-likeobject. In such circumstances,momentummay be per-
ceived as being "usedup"or, better, "pouredinto"the compression, which leads
to a rebound. "Momentumpoured into compression"is a midpriorityanalysis
in expert thought that is the replacementfor naive bouncing.29
Brakingand momentumpouredinto compressionare more importantto phys-
icists thanto those with a naive sense of mechanism.On the otherhand, although
they are not quite naive phenomena,these scenarios also cannotbe mistakenfor
conservation of momentum or energy in any formal sense.
Other examples. I believe there is an importantsystematicityreflected in

the fact that the intuitive response to the railroadcar problem is the static limit
of the correct response. Let me mention some other examples of situations in
which phenomenologyabstractedfrom static situationsis applieduncriticallyto
dynamic situations. Figure 5 depicts an extremely compelling image of tides on
the earth as caused by the moon. Water simply sloshes over in the direction it
is being pulled by gravity, as if the moon and earthwere nailed in place. In fact,
the earth is falling toward the moon due to gravitationalattraction.This causes
two tidal bulges, one on the side of the earth facing the moon and one on the
other side.30 This situation is complicatedby the fact that, even if one thought
29Readersmay protestthatit is really energy ratherthanmomentumthatis "pouredinto"the com-

pression of the spring. This is a properrefinementof the intuition;however, at the level of a recogni-
tion that would cue different behavioral assumptionsin problems, I do not believe the distinctions
among motion, momentum, and energy are encoded.
30Becausethe earthexperiences a strongerpull thanthe bits of water farthestfrom the moon and
less strong than the bits closest to the moon, it falls at an intermediaterate. (Imagine three cars of
decreasing power lined up at a stoplight. On the light turning green, the cars spread apartjust as
the "innertide" gets ahead in falling toward the moon, and the "outertide" lags.)
EARTH
"BULGE"OFTIDAL
WATER
FIGURE 5 An image of tides: The moon's gravity pulls Earth'swater toward it.
of the situationdynamically, the motion of the earth is usually ignored in com-

parison with the larger motion of the moon. "Verybig thingsjust do not move,"
might be a verbal expression of this heuristically useful phenomenologicalas-
sumption. Tides would be easier to understandif there were water on the moon,
the outer lunarexisting because of the centrifugalforce from the moon'sorbiting.
A well-knowntricky situationfor novices and even unwaryexperts involves a
weight falling while attachedto a string as in Figure 6. The temptationis simply
to assign tension to the string as if it were holding the mass at rest. In reality,
becausethereis acceleration,the tensionin the stringis less thanthe staticsituation.
A staticproblemsolutionis here applieduncriticallyto a dynamicsituation,which
suggests thatthe knowledge system either does not easily recognize dynamicsit-
uations or does not know that they are essentially different from static ones.
Theforce as a moverprimitive, which shows itself so prominentlyin dynatur-
tie problems and in nearly everyone'sjudgmentthat the behavior of gyroscopes
is counterintuitive,is a prime case of a phenomenonthat is (a) abstractedfrom
. .......... ...
... ..... .... ............... ............
.. ...... ...... ..... . . ....
.. . .. ...... ...... ........................
. ............................................................................................... ...........................
.
...........
.....................
.... .... ...........
............... ........ ..... ... ...... ............
....
........
................. .....
.........
... .. .............................................................................................
....... .... ... .. ...
................ ................................... ..............................
........................................................... ............ ................................................................................................................................
........................................
.... .............. .... ........ ..............
...
.................
..... ...... ...........
.........................
................. .. . .... ..... ........
.............. ............
. ................................. . ....... .................
........... ........ .........................................
. ............ . ........
...
.......
.................
..... ........ ...........
................
..........
. .................
FIGURE 6 What is the tension in the string holding up M?
164 DISESSA
static situations (pushing on objects at rest) and (b) applied indiscriminatelyto

dynamic situations.
Force as a mover or a closely relatedphenomenonappearsin a problem first
studiedby Viennot (1979) and later by McCloskey: Whathappensif the support-
ing string of a pendulumis cut at various stages of its back and forth oscillation?
(see Figure 7A). Figure 7B shows the prediction of force as a mover: Gravity
pulls the pendulumstraightdown, independentof existing momentum.Because
gravity is generally not thought of as a violent push or shove, I believe it is a
separateprimitivephenomenondoing the work in this case: Released objectsfall
(straight)down. The differencebetweenthis andforce as a moveris thatthe former
does not concern itself with agency but merely asserts as a default assumption
that things fall down. Undoubtedly,however, subjectswould appeal to force as
a moveras justificationif the shallowerprimitivewere questioned.Released ob-
jects fall shows precisely the same omitted preconditionas force as a mover: It
is semistaticin that it ignores initial velocity. So even if released objectsfall and
force as a moverare separatelyencodedphenomena,the two notionsnicely parallel
and reinforce one another.
A surprisinglyfrequenterror is to pronouncethat the accelerationof a body
at the peak of a toss is zero (Reif, 1987). In actuality, the accelerationof a body
duringa toss is always a constantdownwardquantity.This error fits the pattern
of focusing on static aspects of a situation(a point of rest), ignoringthe dynamic
context (the point of rest is of infinitesimaldurationand occurs within a process
of constant acceleration).
Finally, I mention a class of interview problems from my study that I have
not yet discussed. These have to do with weighing in a broad range of circum-
stances, in particular,weighing objects in downwardmoving or downwardac-
celerating elevators. What is the reading of a scale supportingan object in an
elevatorgoing down? The most primitiveresponsesare topological: If the object
is on the scale, the scale should show the object'sweight. Obviously, only a su-
perficial mechanismis evident in such an explanation.Althoughthose who pro-
posed such predictions often realized that the elevator and attachedscale could
"runaway" from the object, and that as soon as contact was broken, the scale
could no longer read the weight of the object, the dynamics of the situationap-
(A) (B) (C)
FIGURE 7 (A) The string is cut on a pendulumat various stages of its swing. What hap-
pens? (B) The pendulum falls down. (C) The pendulum falls "away."
peared irrelevantto them up to the "break-away" point. Their reasoningwas es-

sentially static: "If the object is on the scale, the scale measures its weight; if
it is not on the scale, the weight is not measured."Before examining the roots
of the static predisposition, we turn to a simpler but related issue.
Figural Primitives
The second class of systematicityappearsin the use of figural patterns,judging

plausibilityof motionson the basis of the form or overallvisual patternof a trajec-
tory. One example is the acceptanceof circular motion as a primitive in situa-
tions in which obvious circularsymmetryexists. It also seems plausible that the
left-right symmetryof the railroad-car-in-spaceproblemdirectly encouragesthe
symmetricalpredictionof the static springscale phenomenonratherthanthe evi-
dently asymmetricaldynamics.
Emphasis on figural patternsratherthan on the mechanismor dynamic pat-
tern generally plays a role in the static predisposition.I propose, however, that
thatrole is separablefrom the semistaticphenomenologydescribedearlier. I pro-
pose that figural patternsconstitutea relatively primitive and modularpiece of
the developing sense of mechanism.
Althoughchildrenoften use judgmentson the orderof "itlooks right"to make
decisions about what happens in physical situations (Piaget & Inhelder,
1966/1971), physics studentsseem to lose confidence in the explanatorypower
of figuralpatternssoonerthanwithcore andevidentlybehavioralstaticallyoriented
phenomenasuch as force as a mover. Thus, McCloskey found Figure 7C as a
responseto the pendulumproblem,whereasnoneof my presumablymore sophisti-
cated MIT subjectsgave such a response. I have found, however, even sophisti-
catedadults, such as nonsciencegraduatestudentsat MIT, who thinkthatan orbit
around a cubical planet should be notably square (Figure 8) and that elliptical
orbits can only exist if the shape of the attractingbody is nonspherical. So the
I
/I
! ,\I
I
/
(A\
I)B I\
(A) (B)
FIGURE 8 A predictionthat an orbit arounda square planet should be figurally similar.
166 DISESSA
downgradingof figuralpatternsin the sense of mechanismmight specificallyhave

to do with instruction.
It is worth commenting on an importantpossible exception to the modular
reductionin priorityof figuralprimitives.Abstractbalancingseems to have much
in common with figural primitives; it is cued, at least under some conditions,
by spatialsymmetry, and it is more a mathematicalconstraintthana mechanism.
Yet, if it belongs to the class, it must become separatedfrom other figuralprimi-
tives, because experts have a ratherspecific need for it in the more mathematical
side of theirreasoning.Further,symmetryconsiderationseventuallybecomelinked
to deep notions of mechanism in physics, so some aspects of figural reasoning
might well be encouraged and survive.
There are reasons not to be too concerned with the apparentexception ab-
stract balancingprovidesand with the potentiallink figuralprimitivesmighthave
with symmetry.Abstractbalancing is abstractand may be relatively easily sepa-
rated from genuinely spatial figures if the connection is ever strong. Indeed, as
pointed out earlier in the monkey problem (p. 137), one thing studentslearn is
not to interpretabstract balancing as relating directly to spatial symmetry. In
introductorymechanics, one uses the primitive mostly in the context of algebra
(e.g., expressing conservation of energy). Learning that one may use abstract
balancingin such a narrowcontextshouldbe relativelyunproblematic.Symmetry
considerations, which might reinforce figural considerationsmore widely, are
usuallybarelytouchedin elementarymechanics.It is more prominentin advanced
mechanics (Hamiltonianand Lagrangianmethods) and in field theories such as
electricity and magnetism.
How do we understandfigural patternsto constitutea module in terms of the
basic assumptionsaboutthe sense of mechanismexpressed in the Theory Sketch
section, particularlythe assumptionof fragmentarylearning, mostly p-prim by
p-prim?First, all figuralprimitivesrelate to Newtonianmechanics in essentially
the same way. They are irrelevant.Thus, learningwill not force studentsto treat
them individually. Second, it does not seem unreasonableto assume that figures
might be naively perceived as a class of recognitions. Thus, students could be
addinga prominentmeta-p-primthatdeals with a commonabstractionof the many
figural primitives: "Figuresare not relevant to how mechanical systems really
work." Naturally, one would not expect to see a monolithic change in the status
of figural p-prims in any case; some figures may lag or lead development. But
a class abstractioncould go a long way toward creating a module of phenome-
nology.
Ackermann(Ackermann-Valladao,1981) has done some careful studiesof the
developmentof the notion of horizontalityin children. These studies show some
aspects of the early use of figural primitives and their demise in thinkingabout
physical situations.The particularquestionshe considersis how the level of water
appears in a jug if the jug is tilted or even inverted. The youngest children al-
ways use the canonical, static image of water in a jug and draw the water
level near the base of the jug, parallelto it. Near-transitionchildrencan be pro-
voked to changetheir answersby remindingthem of phenomenasuch as the fact
that "waterfalls down" or of images such as drinking from an inverted bottle,
in which water must move to the mouth in order to be drunk. These children
sometimes revert to the jug-frame image when encouragedto draw again, sug-
gesting thatthe act of drawing,functioningsomethinglike a modality,encourages
static imaging and suppresses more dynamic insights.
More generally, Ackermanntraces the genesis of absolute-framehorizon-
tality to gestural and dynamic imagery and sees the suppression of the mode
of thinking associated with drawing, sometimes even self-conscious suppres-
sion, as importantto the eventual understandingof horizontality. An analysis
parallel to that suggested before is that children have or develop a family-
resemblancesense that detects "drawinginstincts"as a particularclass of think-
ing, thus allowing the systematicreductionin their priorityin the case of physics-
like questions.
Churchand Goldin-Meadow(1986) uncovered strikingphenomenaconcern-
ing such modalities. They show that children who give evidence in gestures of
knowledge that differs from their verbal explanationsare often near transition
to a new stage of understanding.Such children are substantiallymore receptive
to instructionthanchildrenwho give no signs of having divergentconceptualiza-
tions in their gestures. However, these authors make little of the difference in
modality, other than providingevidence of knowledge presagingtransition.For
example, they make no observationsrelevantto children'sawarenessof "differ-
ent ways of knowing." For my purposes, unfortunately,it is less relevant that
there are modes in knowledge than that knowers perceive such modes.31
These considerationsare speculative. The data even underminethe modular
contention somewhat in that, if children come to perceive gestural or dynamic
primitivesto be more relevantto situationssuch as the horizontalityof waterthan
figural ones, that clearly does not spreadbroadlyto the use of figures like circu-
lar patternsin orbits. (At least it does not spreaduntil the more systematicinter-
vention of physics instruction.) Nonetheless, the sensitivity to modes of
understandingwould representa possibly high-leverage development of meta-
knowledge. Generally,phenomenologyof phenomenology-abstractionsthathelp
define classes of phenomenaand their applicabilityto classes of circumstances-
may contributesome of the broadestand deepest modularitiesof naive physics.32
31Theexistence of modes-significant chunks of mutualuse systematicity-constitutes an oppor-

tunity for modularreductionof priorities through mechanismsother than meta-conceptualizations.
This is not pursued here.
32Empiricalpursuit of these questions should include investigating the robustness of, say, the
figural answers to the pendulumproblem, in response to interventionslike Ackermann's:If stu-
dents are asked the pendulum problem, say, in the context of dynamic simulation rather than in
McCloskey's paper-and-pencilstyle, answers may change. In this context, see Kaiser, Proffitt, and
Anderson (1985).
168 DISESSA
Explainingthe Static Predisposition

The tentative descriptions of figural primitives as a module leaves the broader
questionof the origin and meaningof the staticpredispositionlargely untouched.
I make my proposalusing an abbreviated"competitiveargumentation,"building
on the examples presented so far and a few related ones, to draw the issues as
sharplyas possible. The competitionis less fair than it might be, because I could
not locate any substantialpublisheddefenses of the competing positions by the
criteriaI propose. However, I believe these competitorsare interestingproposals
that will serve to heighten by contrast some of the points I would make about
my own proposal in any case.
Ontologicalexplanations. I beginwithtwo proposalsthatareontological

in the sense thatthey propose thateitherphysics-naivepeople simply do not have
motion as a basic category of thought, or they have it placed in a fundamentally
wrong category with respect to its causal nature. Learningmust happen by ac-
quisition of a new ontology or a major shift of a category from one ontology
to another. A specific version of the former, which has been offered in the case
of young children (Henriques, 1984),33is that reasoning about motion must be
done as a sequence of staticconfigurationsand thatno propertiesintrinsicto mo-
tion other thanthose evident in discrete change can be considered. This amounts
to always thinkingaboutmotion as "snapshots"in time sequence, with reasoning
done by comparingthose snapshots. Such an argumentmight conclude that in-
finitesimaldurationsof rest are inconceivableinasmuchas a single snapshotshows
absolutely nothing about motion, and the pair of snapshotsthat shows an object
at rest also shows an interval of time over which the object is not moving. This
reasoning could produce the novice bug of not seeing acceleration at the peak
of a toss but seeing instead an at-restcondition of some extended duration.One
could arguethatthe commonanswerto the railroad-car-in-space probleminvolved
reasoning simply on the basis of a sequence of static images of the mass and the
spring in gravities that are stronger at first, then less strong.
The second ontological explanationof the static predispositionis that motion
is conceived early on as belonging to the category of "aneffect"and that, univer-
sally, effects need a cause. This conceptionis a direct abstractionof the continu-
ousforce p-prim, effort engaged in the productionof motion, althoughit elevates
the level of abstractionand universalizesthe relationship.In contrast, according
33Otherswho hint at ontological explanationsof developmentinclude Hayes (1979a) and Larkin

(1983). The former suggested that the Piagetiandevelopmentof conservationmay reflect a change
in ontology for liquids. The lattercharacterizednovices' physical knowledge as strongly constrained
by time-sequencedcausality(cf. with evidence presentedhere for the use of figuralprimitives,which
involves limited sequentiality,at best), whereas experts rely only on time-independentinferencing.
Chi (1992) has recentlyofferedan ontologicalaccountof difficultiesin learningthatcontrastsstrongly
with the view presented here.
to Newtonianphysics,motionis a stateof the systemandneedsno intervention

to perpetuate.Only changesin stateneed causes(forces).34Learningphysics
meansmotionmustno longerbe conceivedof as an effect.
Becauseit is not specificaboutcausesof motion,this secondontologicalex-
planationmay be furtherspecifiedby any numberof kindsof causalrelation-
ships,althoughpresumably theprincipleloses explanatorypowerandempirical
as
plausibility a sense of mechanism if it straysfar frommechanical
(forceful)
interventionsas the type of cause. "Theobject'swhim"as a causecan explain
arbitrarymotiontoo easily.
A counterproposal. Ratherthanproposinga competitorfortheseimmedi-

ately,I will let it emergefromcriticismof them.Threeclassesof criticismsare
presentednextas follows:theoretical inadequacy,empiricalinadequacy,andsub-
sumption into a better articulated
competitor in
("knowledge pieces").Inthecourse
of argumentation, I notethreebasicdifferencesfromthe case of figuralprimi-
tives.First,theseontologicalproposalsareso broadanddeep,coveringthewhole
of the senseof physicalmechanism,thata monolithictreatment,say, as a single
modality,is extremelyunlikely.Second,unlikefigures,it doesnotappearplau-
siblethatanysinglemeta-conceptualization canbe responsiblefor a categorical
unity.Third,the developmentof an expertsenseof mechanismcannottreatall
naivep-primsthatmightadhereto the naiveontologyuniformly.Someof them
areessentiallycorrectand,unlikefiguralprimitives,areneededat relativelyhigh
levelsof reliability priority.Hence,learningwillbe moredistributed,
distinguish-
ing the useful p-primsgraduallyfromthe less usefulones.
1. Theoreticalinadequacy.TheprogramI set forthin the secondsectionof

this monograph providesa set of questionsto ask of any proposeddescription
of a knowledgesystem.Thetargetof thesequestionshereis the encodingof on-
tologicalstaticpredispositions.To takestepsbeyonda commonsense epistemol-
ogy, one has the to
obligation beginmakingmoreprecisedescriptions of proposed
mentalstructureslike ontologies.
A refinementof the commonsensedescriptionof theseontologicalproposals
mightbe attemptedalonga numberof lines, say, as specificrulesor proposi-
tions, or as a propositional
system.The latterhas not, to my knowledge,been
attempted.Ratherthanpursuethese(in my view unlikely)possibilities,I briefly
investigaterefinementalongthe lines of the p-primstheorysketch.The major
difficultyherelies in describingthe integrationof a diverseset of p-prims,the
presumedoriginof the naivesenseof mechanism,into a monolithicontology.
34Philosophers,notably St. Thomas Aquinas, have considered the universality of the proposi-
tion thatmotionneedsa cause in proofsof the existenceof God: "Theremustexist an unmovedmover."
Similarly, the claim that even existence needs a constantcause leads to the same conclusion. I do
not take these argumentsto have great intuitive force, despite their intuitive roots.
170 DISESSA
Any proposalthatrequiresvery broador universalapplicationof categorizations

(change, cause), as ontologies do, is suspect on groundsof genesis and develop-
ment. The processes that decide whether something necessarily must be inter-
preted as a change must be elaborate to be reliable. One needs to accumulate
significant stores of knowledge specific to the many special contexts of applica-
tion in order to classify reliably. Beyond classification, the spread of confident
use of a p-prim such as motionneeds a cause requiresmore than seeing the term
in the situation. It requires seeing the entailmentsof the p-prim as basic and ex-
planatoryof all those circumstancesthatinitiallywouldbe coveredby morespecific
p-prims. Consider the argumentsmade about Newton's third law. P-prims are
simple to generate but very difficult to elevate to universal status. Similar criti-
cisms apply to any proposed universal method of viewing phenomena. For ex-
ample, the ability to formulateand confidentlyapply a reduction, such as seeing
all change as sequentialsnapshots,must be complex. These criticisms are played
out empiricallywith referenceto particularproblemsof universalizingacross con-
texts next.
This line of argumentationis subjectto a substantialchallenge in that motion
needs a cause may be a meta p-prim, an abstractioncommon to many unproduc-
tively causal p-prims. Causal syntax is a similarcommon abstraction.However,
the story of causal syntax regardingdevelopmentto expert thinkingis not nearly
so simple as that of figural primitives. In the first instance, causal syntax is in
some ways productive. It is certainly encouragedat early stages of learning F
= ma, and it is even correct if change is interpreteduniformlyin the Newtonian
way (i.e., as acceleration). More generally, one simply cannot treat all p-prims
adhering to causal syntax uniformly in looking toward expert understanding.
Althoughmanyprimitivesthatobey causal syntax(e.g., suckingandperhapseven
continuousforce) must be reducedin priority, others (force as a mover)are cer-
tainly used by physicists. Moreover, even in the cases in which causal syntax
succeeds, deep and complex refinementsare necessary, as indicatedin the dis-
cussion of action and reaction in the Development section.
It may generallybe easier to dismiss a module from the naive sense of mecha-
nism thanto integrateit properlywith Newtonianmechanics. Althoughone can-
not rule out a priori that there exists a naively perceptible category other than
causal syntax that unifies motion needs a cause in such a way as systematically
to undermineexactly those primitivesthatneed underminingand to advancethose
thatneed advancing, such a prospectis dubiousgiven the case studyof the causal
syntax as a precedent.
2. Empirical inadequacy. On the basis of the empirical work already dis-
cussed, I can be very specific about some deficits of these ontological explana-
tions. These will be the complementof the argumentsjust made. Not only is it
unlikely that motion needs a cause is a disapprovedmodule in the expert sense
of mechanism, but also the idea is hardly a universally approvedmodule in the
naive sense of mechanism.
It seems clear thatphysics-naivepeople have dynamicp-primsthat violate the

motion needs a cause dictum and the specificationthat dynamicphenomenacan
only be considered through the lens of sequences of static states. One of the
findings-more properly, it was a nonfinding-of dynaturtleexperiments(diSes-
sa, 1982) was that, althoughpeople often ignored the preexisting motion of an
object when it was pushed, they had no objection at all to the continuedmotion
of dynaturtlein the absence of pushing. Within the small confines of motion on
the screen, not a single subject questionedthe turtle'scontinuedmotion or sug-
gested any reason for it. An object continuingforever is one thing, but relatively
limited nonforcedmotion is not only not problematic,but it is also expected. And
there was no indicationthat subjectswere inclined to combine the cause of con-
tinuingmotion with externallysuppliedpushes. If one retrenchesto allow nonlo-
cal causes such as the initial push of a hand in a toss, Newtonian physics is not
in contrastto motionneeds a cause, because one may properlyspeak of the push
as the supplier of momentumand, hence, as the cause of the motion.
Anothercounterexamplewas providedby the studentwho had the more static
shock-absorberimage ratherthan oscillations attachedto springs as default be-
havior but who, nonetheless, conceived of other dynamicevents such as bounc-
ing as "whatis going on" when one throws a brick onto a spring. Except that
it occurs in a particularcontext, there is little basis for distinguishingthe release
part of the press-and-releaseconcept from a situation such as a manualtoss of
an object not involving a spring, in which there would be no doubtthatthe object
could be thrown into the air. More particularly,there cannotbe a distinctionbe-
tween the release phase of press and release and the reboundphase of the throw
onto the spring based only on a sequence of static states. The maximally com-
pressed spring is static-stateidentical in the two cases. To the student, the two
are distinguishedby one being in the midstof a bounce. The distinctionmay sound
strange,but it is identicalin form to the physicist'sinsistencethat velocity, being
in the midst of linear motion, distinguishesmotion from rest. The choice of what
a thing may be in the midst of is the only distinction. Indeed, the very notion
of a body experiencing a force, which is naively unproblematicin some cases,
implies a class of goings on different from what can be "seen"in snapshots.
In experiments on circular motion-a ball on a string, say-children at vari-
ous stages of development see different things going on. Thus, what is natural
change and what needs interventionis neitherfixed nor universal. Earlyon, chil-
dren may see circular or at least curvilinearmotions as primitive goings on. At
more advanced stages, more often they see the object engaged in linear motion
that is disruptedby the string. In both cases, children often follow (sometimes
articulately)a heuristicthatcontradictsmotionneeds a cause. They say thatwhat
is going on will continue unless some sort of interventionis made (Globerson
& diSessa, 1984). In this context, we saw young childrenwho predictedthatmo-
tion should stop altogetherwhen the string holding it in the circle was cut, not
because the cause of motion had ended, but because the ball "would not know
172 DISESSA
which way to go." More telling, reactionsto a simulationof a ball stoppingwhen

the string was cut were not approvalbut subjects'announcementsthatthere must
have been an interventionto stop the object. Absence of motion, in the context
of very rapidcessation, needs a cause other than the cessation of an externalin-
tervention.
To summarize, the sequence-of-static-statesdictum predicts an unacceptable
restrictionon the class of expectationsaboutmotion in view of the rich phenome-
nology that the empirical bases of this work show. The naive sense of mecha-
nism respondsfavorablyand primitivelyto many essentially dynamicgoings on.
Some may need causes of certain kinds, but they by no means all categorically
need constant, interventivecauses. To patch the argumentby assertingthat only
unnaturalchangesneed interventionsbegs preciselythe same questionsaboutwhat
constitutes the categorizationinto naturaland unnaturalchanges.
3. Knowledgein pieces. It is time to discuss a specific competitorto the on-
tological proposals, which, naturally, is drawn from the theory sketch and de-
tails of the naive sense of mechanism as they have been laid out. My basic
contentionis thatthe static predispositionis a relatively broadbut still idiomatic
systematicity that is not categorical. In particular,I identify a general but not
a universal dynamic impoverishment.The elements and, more generally, attri-
butes of the naive sense of mechanismare skewed towardstaticphenomena,par-
ticularly at the highest priority levels. Thus, althoughdynamic phenomenaare
perceptibleand readily used, when it comes time to think carefully and explain,
thinkingmoves (a) toward staticallyperceptibleattributessuch as geometric po-
sition and orientationand (b) toward the higher priority causal primitives such
as the interventiveforce as a mover or continuousforce. Thus, althoughevery-
one knows that motion perpetuatesitself, it is tempting to think in terms of an
internal interventionas a deeper mechanism if the issue is specifically raised.
But, although it may be suggested, there is simply not enough structurein the
knowledge system to supportthat interpretationuniversally, and changes in the
situation would revive less mechanistic p-prims, such as simple "returnto
equilibrium."
This impoverishmentwith respectto dynamicdescriptionsand gradientin con-
fidencefavoringstaticallydescribablephenomenashouldnot be surprisingin terms
of the properties of our sensory system with its strong reliance on vision and
spatialrelations.Phenomenologyconsists, at first, of minimallyabstractinterpre-
tations based on strong vocabularies. If our touch sensitivity were as structured
as vision (e.g., one could see forces as positions in force space), and, indeed,
if we had remote touch-sensing capabilitiesinstead of being confined to where
we can put our fingers, things would likely be different. Instead, we make use
of whateverwe can readilysee thatcorrelateswith the structureof motion. Thus,
for example, figural p-prims stand prominentlyto be abstracted,and these are
only underminedand replacedvery slowly by the weak force of evolving priori-
ties. I have already gone some distance toward assimilating the insights of the
two ontologicalproposalswithin the broaderframeworkof knowledge in pieces.

Let me continue doing so.
First, I have localized the change needs a cause dictumto phenomenasimilar
to force as a mover and continuousforce.35 This move has the advantagethat
what counts as change is specified by the local knowledge elements, not by some
universalframe. We do not have to decide what qualitiesof motion are essential
to motion in orderto define the static predisposition.Defining qualitiesessential
to motion was difficult for Aristotle, and even if we, as theorists, can do it, it
seems unlikely that children solve the problem intuitively in any general way.
If we take this conservativeposition, thatthe problemof causalityis at best solved
locally, it need not concern us if in a situation of clear symmetry, circular or
return-to-balancemotion needs no interventiveexplanation,but outside of those
circumstancesit requires intervention.We need not concern ourselves with the
dubious task of defining change in terms that contrast short-termcontinuations
of motion (which in general need no explanation)from perpetualmotions (which
more often appearto need explanation). The knowledge-in-pieces view is that
many kinds of change need explanationbut thatthere is no universalcharacteri-
zation of change that covers the circumstancesin which people feel the need,
or not, to look for deeper causation. Instead, these predispositionsare distribut-
ed accordingto the kind of change; that is, they are embeddedin the primitives
that connect to the particularcircumstance.
A piece of insight at the root of the snapshotontology is that, within the naive
sense of mechanism, the at-rest condition is a very special goings on. Being at
rest is an importantattributeof situationsthat determinesthe way people think
about them. Thus, rest accumulatesmany coordinateattributesspecial to it. In
general, although not universally (e.g., sudden stopping is an exception), rest
needs no explanation,andit participateswith interventionsin very particularways.
For example, rest is expected to be a terminalconditionwith indefinitecontinua-
tion in noninterventivesituations. (Motion dies away to rest, which continues
indefinitely.) Or, rest is self-perpetuatingprior to an intervention.These termi-
nal and self-perpetuatingattributesfor rest make it difficult, althoughnot forbid-
den, to conceive of rest as a transientphenomenon, as at the top of a toss.
In contrast, rest and motion are only accidentallydistinguishedin a Newtoni-
an frame. The at-reststate of an object is, to a physicist, not a particularlydistin-
guished class of goings on. It is in no way intrinsic to the phenomenology of
any situation.The very fact of nonmotionis dependenton the accidentof picking
a frame of reference with the same velocity as the at-rest object. To a novice,
things such as accelerationevidently belong to the phenomenologyof things in
35Thisneeds qualification. There seems little doubt that change needs a cause is an early de-
veloping notion. Childrentake spontaneousmotion of certain sorts as somethingthat needs explana-
tion; in fact, they take it as a primarycue to "alive,"at least in situations in which the issue is to
determinewhether a thing is alive or not. But it is one p-prim among many; it does not follow that
children cleanly and with high reliability encode the general principle that change needs a cause.
174 DISESSA
motion. They need specific considerationin cases of rest ("topof the toss" im-
plies no accelerationmisconception)to be seen as applicableat all. A physicist
generallydoes not treatat-restsituationsdifferentlyfrom those involving motion.
One of the theoreticaladvantagesof this assimilationof the insights of onto-
logical proposals to the knowledge-in-piecesperspective is that one thereby has
a more uniform view of the naive sense of mechanism. Instead of singling out
a particularclass of elements, relations, and systematicity(e.g., static elements,
some particularnotion of change or cause) for special treatment,we can see all
of the things that people know about motion in a uniform light. We no longer
are challenged to compare the status of categorical aspects of the knowledge
system-aspects such as ontological knowledge-with all the bits and fragments
that we can easily see people using in reasoningaboutthe physical world. Facts
such as that objects move continuouslyor that there exist only certain kinds of
causal interactions(mechanical, not directly willful causes for an inanimateob-
ject's motion) do not need to be patchedinto the snapshotview as subsidiarycon-
ditions.
Finally,the ontologicalproposalsmiss entirelythe substantialstructuralchanges
thatoccur in the sense of mechanism.The knowledge system startsas distributed
and fragmented,and it evolves to one that, on the basis of a broad integration,
manages to think about a very broad range of situations on the basis of a few
universalcausal mechanisms. These physicist universalsinclude the lack of need
to explain uniform motion in a straightline and the possibility of explaining all
otherchangeson the basis of a very restrictedclass of interventions(i.e., mechan-
ical, gravitational, or electromagneticforces). In contrast, ontological and in-
deed all categorical proposals treat naive and scientific senses of mechanismas
if they were on par, suggestingthatone learnsby tradingone ontologyfor another.
COGNITIVEMECHANISM
The ultimategoal for the work representedin this monographis a computational

theory of common sense and intuitiveknowledge and its evolution into scientific
understanding.Althoughthis work is still relatively far from implementablepre-
cision, one may still benefit from clarificationand furthermotivationof the as-
sumptionsaboutmentalmechanismthathave been used as a descriptivelanguage
for empirical observations. This would both clarify the meaning of statements
made in terms of the mechanismand better specify the intendedrange of the the-
oretical perspective. I take it as a nontrivialcontributionof this monographthat
it joins importantphenomenologyof learningphysics to simple, formalizablecog-
nitive mechanisms(structuredpriorities). With the exception of the final subsec-
tion, this section will likely be of interestmostly for those readersconcernedwith
issues such as choice of modeling languageand the relationof cognition as mod-
eled to cognition as abstractedfrom protocol behavior.
The predominantuse of mechanismin this section, of course, refers to mental

mechanismratherthanto the physical mechanismthathas been the dominantfo-
cus up to now. This monographis recursive, however, in that it attemptsto pro-
vide a sense of mechanism about certain activities of mind. This section is an
attemptto pointthe way from an orienting,althoughnot compactlyand rigorously
specified, sense of mental mechanism toward a simple core.36
I improve the descriptionof the model of mental mechanismin steps: First,
I review the motivationsfor the programof researchthat are relevantto select-
ing the way cognitive mechanistic substrateis modeled; second, I clarify the
mechanisticmodel itself; third, I clarify the context for interpretingdatain terms
of the model; and, finally, I abstractand extend slightly the observationsthat
have been made about the systematicityof naive physics to the point of under-
standinghow they might be relevantto other areas, such as intuitive"socialrela-
tions" and schemes of story comprehension.
the ModelingLanguage
Motivating
That knowledge comes from experience is a truism. Yet much of the research
in cognitive science in the area of developing expertise, mental models, and
novice- expertdifferenceshas been on the basis of assumingpowerfulknowledge
representationschemes and processing organizationsthat seem to presume far
more uniformitythan the texture of experience indicates. Compareproduction
systems'universalactivationin well-specified circumstancesto the vague impres-
sions one has of productiveor unproductivedirections in problem solving or of
relative satisfaction with a solution in contrastto terminationof an algorithm.
How do we model impressions and their effects?
Empirical results, such as those presented here, indicate complex, conflict-
ing, andunreliablestrandsof reasoningin which studentsmay be guidedby aspects
of the circumstancesthatthey cannotarticulateor check. This also suggests care
in modeling. Do we simply not consider the fact that studentsmay never be able
to reinstantiatea particularmental state that accountedfor a judgment (a "once-
in-a-lifetime"production?),in strong contrastto experts who can simply "solve
the problem again"?
Using powerful processing or knowledge representationschemes withoutthe
constraintof understandinghow the particularforms of knowledge and reason-
ing are learnedmay well miss some of the most basic and difficult partsof learn-
ing. Even if we can roughly describe the data and processing schemes of one
level of competence, understandingthe developmentto the next level may force
us to look for invariantsin the pieces of encoding. How does the descriptivesys-
36Thecore, however, cannot operate as a theory without interpretingphenomenasuch as those

describedin previoussections. This is similarto the relationof formulasandpropositionsto the broader
sense of mechanism that I have argued is importanteven in expert physics.
176 DISESSA
tem that specifies the conditions for actions develop? The descriptionsof new
conditionsare unlikely to be simply constructiblecombinationsof attributeseas-
ily available from prior stages of competence, at least if substantialconceptual
change is at issue. Instead, one may need to rebuildon lower level or just differ-
ent primitives. Sensory schemataat fairly low levels may need to be recrafted
or reselectedto "seethe world in a differentway." Buildingthe descriptivecapa-
bility and the systematicitynecessary to instantiatea rule reliably in a relevant
contextmightbe a substantialproblemfor knowledgesystemsof the type described
in this monograph. Hence, it would be very easy to gloss this concept-specific
developmentwith the insertionof a rule and presumptionof powerful and gener-
ic rule-processing capability.
Similarly, goals as computationalconstructsthat operate in relatively clean
planningor problem-solvingschemes may well be some of the last kinds of men-
tal objects createdin a knowledge system; fundamentally,they may be optimiza-
tions in the overallcontrolof an already-operating systemwithadequatedescriptive
capability. Certainly it makes less sense to have goals if one cannot reason
hypotheticallyaboutalternatives.But, consideringsome of the empiricalfeatures
of intuitivephysics, reasoningaboutthe merits of alternatives,say, differentin-
terpretationstrategies, seems a ratheradvancedstate. (See the furtherdiscussion
of "aesthetics" thatfollows.) It seems more likely thatless reliable,more situation-
specific "accidents"of cuing drive a naive attemptto understanda system than
a collection of preformed goals that organize search.
To take anotherreasoning scheme, predicate calculus is as difficult to learn
as the physicallaws whose developmentis at issue here (Wason& Johnson-Laird,
1972). We certainly should not assume that kind of reasoning infrastructurein
modelinghow studentslearnanduse physics knowledge,most especiallyin every-
day contact with the world.
Put in computationalterms, the symbols or data structuresone writes in mak-
ing a model of a knowledge system may seem compellingly to contain the
knowledge we attributeto them. However, dataare dependenton the interpreter
thatoperatesover them for any meaningthey might have. I believe we have little
reason to suspect that any natural, general, highly reliable interpreterexists in
the mind. Certainlythereis, if anything,negativeevidencethathigh-levelschemes
such as predicatecalculus are a built-in interpretercapability. So our epistemo-
logical programmustincludebuildingthe interpreter(or more likely, manymicro-
interpreters)as well as specifying the data, rules, or propositions.37
Take the case of productionsystems in a bit more detail. Structuredpriorities
is a much restrictedmodeling language compared with typical productionsys-
tems. To show this, we can model a p-prim as a productionthat fires (turnson)
37Theconnectionbetween data and proceduresoperatingon them is so intimateat the low levels

of mechanism presumed here that the distinction is useful only to contrast the modeling language
used here with high-level systems that depend strongly on such a distinction.
undercertainconditions.In a productionsystem,however,knowledgeis dis-

tributedacross(a)theset of productions,(b) a languagein whichto writecondi-
tionsfor the productions,and(c) a languageof mutationfor the actionside of
productions,thatis, a particularsystemfor changinginternalstate.In a struc-
turedprioritysystem,theelementsthemselvesconstitute thedescriptive language;
thereis no separatelanguage,if we thinkof thisas a "language" at all, to rapidly
puttogetherdescriptions of conditions.Thelanguageof mutation-theset of ac-
tionsthatcanbe taken-is alsoexceedinglylimited.Roughlyspeaking,a p-prim
maybecomeactiveor it mayturnoff. Furtherimplicationsfor the stateof the
systemmustfollowfromthe specificconnectionsin the networkandsubsequent
firingsof otherp-prims.
Is it goodto useprimitive,noncommittal modelinglanguages,suchas a cuing
network?It is clearlynot a good ideaif the aimis to modelknowledgesuchas
routineskillor "instructions following"whereonecanpresumesomebasicepis-
temologicalproblems(e.g., reliable,reinstantiatable representation) havebeen
solved, and, therefore,the processesthatsolve themcan be absorbedinto the
modelinglanguage's capabilities.Ontheotherhand,in somecasesit is appropriate
for thetheoristto takemuchmoreresponsibility for understanding theevolution
andmicrostructure of thedescriptive andmutation languages, in which case,using
structured makes
priorities more sense. The to a
impediment learning production
thatis partof expertproblemsolvingmightnotbe lackof anoccasionforlearning
it buttheverycapability to describeandcheckthecircumstances in whichit is apt.
I amnotmakinganin-principle argument againstanyparticular modelinglan-
guages, butI am highlighting the issueof choice of grainsize and the importance
of separating themodelingsystem'sintrinsiccapabilitiesfromthosemodeledby
the system.Thus,predicatelogic andperhapsproductions are abstractions and
formalizations of a certain level of humanreasoning.But, if we are to under-
standthe genesisanddetailedfunctioningof thatlevel, we mayneedto pushto
a morefine-grainedlevel whereit is not at all obviousthese are perspicuous.
We mustbe constantlyon guardto preventthepowerof the modelinglanguage
fromtriviallyexplainingandglossingthe detailsof thehigherlevel functioning
("Allyou need is X production,Y axiom").
In net, my choiceof grainsize andof assumedinherentprocessingcapabili-
ties forthepurposeof investigating theintuitivesenseof mechanismis different
fromthosefor whichmanyothercurrentknowledgeandprocessrepresentation
schemesseembestsuited.Althoughone couldpossiblyhandlethis withcarein
the use of the modelingsystem,I have chosento handleit by avoidingthose
schemesandby assuminguniformbutveryweakrepresentation andprocessing
capabilities.Luckily, these seem to dealrelativelyeffectively with theempirical
phenomena at hand, such as context lack
sensitivity, of articulateness, andlack
of confidentapplicationin manyinstances.
Pop epistemologicalnotions such asfact and theoryare yet more problematic
as termsin which to investigatethe foundationsof knowledge. They are problem-
178 DISESSA
atic, in particular, for understandingthe systematicity or lack of it in naive

knowledge. Unless one gets the grain size and the knowledge system'srelational
structureat least approximatelycorrect, studyingsystematicityis a hopeless task.
Concepts, theories, and facts are loaded with unspecified assumptions about
mechanism,modularity,and systematicity.Facts are independentitems thatmust
be true or false; theories presume some unspecified but rich internal structure,
presumablya structuresimilarto theoriesin the highlycultivatedand socialprocess
of professional science.
Considerthe characterizationof force as a moveras a fact. If a fact is known,
it shouldpresumablybe takento be true independentof mentalcontext. Yet, that
may grandlyoverstatethe circumstancesin which a fact will be thoughtrelevant.
How should one describe the relation betweenforce as a mover and alternative
notions thatsometimes compete with and replace that fact, such as figuralprimi-
tives? One needs somethinglike applicabilityconditionsor a weakerversionthere-
of, such as cuing priority. Characterizationas a fact capturesnothing of genesis
and encoding, such as the use of attributeslike patternof effort, aspectsthat may
help explain the characterand applicabilityof the notion and some of the larger
scale characteristicsof the system of which it is part. Empirical work couched
in pop epistemologicalterms standslittle chance of refiningthose termsinto tech-
nically acceptablecomputationalnotions that do justice to the empiricalfeatures
of the sense of physical mechanism.
My intention is to build an epistemology based on fragmentsof experience,
with a commensurategrain size, that makes no ab initio assumptions(withinthe
focal sense of mechanism)about modularity,universalityof reasoningprocess-
es, and other such large-scale issues; the theorist'sresponsibilityto describe and
display systematicitywill then be evident. One needs a mental mechanismthat
provides a similarly neutral and unassumingepistemological base on which to
pursue questions of the propertiesof the knowledge system expressed in terms
of that mechanism.
As describedbefore, learningis a particularlyimportantconstraintwhenmaking
assumptionsaboutmentalmechanismand, hence, implicitlyaboutthe grain size,
basic relationalstructure,and assumed(ratherthanmodeled)capabilities.Again,
unless one gets such assumptionsat least approximatelycorrect, development
may appeareither magical, "insertingnew knowledge,"or hopeless. I have been
carefulto buildthe conceptof p-primso thatgenesis is relativelyunproblematic-
establishing small, rememberabledescriptions of events. Similarly, the model
provides for development in terms of simple learning events throughchanging
priorities bit by bit, elevating some knowledge structures toward universal
prescriptions. Further, any reasoning built into the system will have to be con-
structedfrom low-level patterns.Indeed, the developmentof intuitiveknowledge
should not at all be expected to reach the systematicityof professional scientific
ideas, as we are used to seeing them displayed, in breadth, depth, or quality.
Contrastthe systematicitythat arises from common attributes(base vocabulary)
to, say, an axiomatic deployment based on operationaldefinitions.
I presume that conceptualdevelopment is a large-scale phenomenoninvolv-

ing substantialreuse of knowledge elements that existed for their roles in previ-
ous stages of competence. If it is true that expertise consists of very large
vocabularies(e.g., Chase & Simon, 1973), then we should be preparedto look
for similarlylarge systems constitutingthe competence of common sense. After
all, expertise and common sense are built by the same machineover similarperi-
ods of time, although their operating characteristicsmight be very different.
An independentstrandof motivationfor the presumedmentalmechanismcomes
from considering the basic machinery humans have available to implement
knowledge. One of the most remarkablefacts abouthumancognition, especially
reflected in expert problem solving, is the quicknessand versatilityof the infor-
mation processing. Yet all of this must rely on neural machinery, which has a
time-dimensiongrain size on the order of a few milliseconds. In comparison,
the time-scale for operations involved in present computer implementationsof
cognitive models is aboutthree orders of magnitudeshorter, at the microsecond
level. Evolutionarily,slow machinerysuggests that in order to react quickly to,
say, a glimpse of a tiger, humanswould need large numbersof recognizers and
comparablylarge connectivities so that any of a huge numberof perceptualpat-
terns could constructon short order an appropriateflee-from-the-tigerresponse.
The interest in connectionistmodels of cognition is inspired, in part, by similar
feelings that some basic architecturaldifferencesbetween humanprocessing and
von Neumannmachineshave consequences(e.g., Rumelhart,McClelland,& The
PDP ResearchGroup, 1986). The latterhave arbitrarydataconnectivity,butthey
are serial and hence have limited proceduralconnectivity.38Structuredpriorities
are a mechanismthatshouldbe relativelyeasily refinableinto connectionist-style
machinery, which does not have limitations typical of von Neumann formula-
tions. I returnto these real-timeconstraintsin the InterpretiveSummaryas they
relate to core properties of the sense of mechanism.
of Mechanism
Clarification
I have so far mixed functional and structurallevels of description of p-prims.
Thus, although explanationand providing a "sense of satisfactorydescription"
are functions of interest, structured priorities are intended to be function-
independentcomputationalterms. In this andthe following subsections,the struc-
tural mechanismsand their relationto my use of them are clarified. First, I give
a notion of p-prim in purely structuralterms and discuss the changes in connota-
tion thatresult. This descriptionis of a greaterdegree of precision thancurrently
necessary to supportthe interpretationsI intend but less precise than necessary
to implement these ideas. The point is to establish a context for interpretation
38InLisp, for example, any data element may be connecteddirectly to any other, but procedural
entities are connected only in series or recursive hierarchies.
"180 DISESSA
thatwill correctsome misimpressionsaboutthe meaningof p-primsthatone might

get from the examplesgiven. This also sets a bettercontextfor the intendeddirec-
tion of futurework. Second, I review the connection of these ideas with the em-
pirical context.
The model of mechanismthatfollows is a fairly generic connectionistsystem.
Consider a set of discrete mental entities, each with an activationlevel. These
are to representthe structuralequivalentof p-prims, and if it is importantto dis-
tinguishthem from the functionallydefinedobjects, I call them structuralp-prims,
or s-p-prims. In standardconnectionistterminology, these would be called units
or nodes. S-p-primshave an externalactivationlevel, in additionto their internal
one, that is supplied by connections to other s-p-prims. The external activation
should be something like the sum of the (external) activations of all connected
elements weighted by positive (activating)or negative (suppressive) factors. A
dynamic is requiredthat specifies the time-varyinginternalactivationof an s-p-
prim on the basis of the histories of the externaland internalactivations. A sim-
ple model would involve discrete time steps, and it might assign the value of the
external activation at one step to the internal activation at the next. Or, in the
case of a continuousmodel, one could use the currentvalues of the externalacti-
vation to specify the derivative of internal activation with, say, the derivative
being proportionalto the differencebetweenthe internaland externalactivations,
which would lead to the internalactivationtracing the external with some lag.
A plausible featureof the dynamicwould be a nonlinearthresholdbeyond which
externalactivationwouldcause the s-p-primto "lockfully on"anda lower threshold
below which externalactivationwould allow the s-p-primspontaneouslyto return
to zero activationor at least to well below lock-on levels. Of course, manyvaria-
tions and complexities might be convenient, but they are beyond the needs of
the current exposition.
This model has a naturaldirection of activation from node to node, with no
feedbackunless by specific connections. In it, cuingpriority refers to the topolo-
gy and the weights of connections into a selected element from other elements
whose activationsare not relatively directly affected by the state of the selected
element itself. Reliabilitypriority refers to the topology and the weights of con-
nections from elements that are (relatively) directly connected back to the ele-
ment of interest. That is, reliabilityrefers to the structureof feedbackpaths that
returnto the selected element in short order from activationsthat proceed away
from it. A contextwould heuristicallyreferto an ambient,althoughsubthreshold,
external activation(in most cases of relatively long-termduration),which func-
tionally preparesan element to fire on the basis of some critical set of activation
links. Strictly speaking, the model is the network of nodes, their connections,
and weights. Cuing, reliability,and context specificationsare qualitativedescrip-
tions of the networkthatare intendedto have fairly simple interpretationsin terms
of subjects'behaviors, such as fleeting responses (high cuing in the context) or
confidence (high reliability).
I spoke before about"relativelyclosely connectednodes"to imply that effec-

tive cuing from the activationof any single element will generally be limited to
directly connecting s-p-prims or, at most, chains of s-p-prims a few elements
in length. Longerchains will exist only in cases of well-practiced,probablyenor-
mously redundantand reliable activities, or else they would depend in a critical
way on contingenciesthatoughtto be capturedwith referenceto the cuing possi-
bilities of specific other s-p-prims.
To this point, I have implicitlyidentifiedp-primswith specificknowledgestruc-
tures that are basically abstractdescriptions of entities, their relations, and at-
tributes.Impressionsof meaningfulness,however, are irrelevantto the structural
interpretationof p-prims. Indeed, these impressionswill be supportableonly in
the more dense regions of the knowledgenetwork.It is unlikely we can use stand-
ard lexicon (e.g., "identificationof some class of objects engaged in a class of
actions")to securely define individualp-prims. Instead, the higher level cogni-
tive structuresimplicatedby terms such as identifying,objects, and actions in-
volve relatively integratedsystems with substructureto allow reliability loops
to check applicability(e.g., "Isthis a teapot?"),specifically to supportappropri-
ate persistence (e.g., "ShouldI abandonthis line of reasoning?"),and to support
use in multiple contexts and with other subsystems, such as language, logic, or
other reasoningpatterns(e.g., the very capabilityto verbalize considerationsas
noted before). The particularp-prims I have described here are likely to be of
intermediatestatus,less datadrivenandcontextlocalizedthanelementsdescending
towardlow-level sensory elements but with substantiallyless surroundingstruc-
ture thanconsciously accessible notions. They are likely to be some of the highest
priorityelements that still are generally hidden from introspectionand probably
not sufficiently capturedby describing them as recognizing standard-ontology
entities, such as objects, attributes,categories, and actions. This skewing of ex-
amples towardthe top of the phenomenologicalheap (butbelow the "consciously
conceptualheap")may, in part, be due to methodologicaldifficulties:Lower level
mental entities will not be directly reportedby subjects even in approximation,
and they will not typically direct the flow of reasoning the way the currentlist
of p-prims can when they constitute significant insights into the way a physical
system works. In consideringlower priorityelements, we as theoristsmust face
in extreme form the problemthatthe words we have readilyavailableto describe
p-prims' meanings are essentially all drawn from higher priority levels.
To correct this in-principle skewing of examples, some examples of less
meaningfulprimitivesthatare likely to exist maybe helpful.Considertwo different
actions (or meanings)thatmay be appropriatelycued by some configurationthat
happensto be both easy to see (involves elements with high cuing and reliability
priorities from sensory elements) and characteristicof both actions (meanings).
It may be useful specifically to encode thatconfigurationas a p-prim. For exam-
ple, fight and flight responses are both readied by perceived aggression. So a
fight-or-flight state element would be useful. The two cued options, however,
182 DISESSA
belong to ratherdifferent action patterns. Thus, our usual strategies of disam-

biguatingideas might push the classificationof the fight-or-flightelement toward
one or the other meaningor, more likely, towarda very differentstate: the logi-
cal disjunctionof two higher priority notions as considered by a sophisticated
human. Disjunctionsuggests a modularityof processing, checking two separate
conditions, that might not exist.
Closer to the realm of researchedtopics, remembrancesof a word or person's
name might include s-p-prims having to do with broad, possibly idiosyncratic
classes of letters or sounds. M, N, or L might share a common prim because
of common phonemic characteristics.In fact, experimentsby Klahr and Chase
(1983) suggest that access to various pieces of informationabout the alphabet
occurs first throughthe particularchunk in the "alphabetsong"in which the let-
ters occur. One may interpretthis in the presentmodel by assertingthe existence
of chunkprims attacheddirectly to each letter that can initiatesomethingsimilar
to an internalsinging of the letter'ssegment. This, in turn, gives access through
monitoring to needed informationsuch as "predecessor"and "successor."39
What is described as the tone of a situationmight be elements of a class of
prims partially representingsituations through extremely ambiguous combina-
tions of factualor emotionalattributes.Here, ambiguousmeanscapableof describ-
ing a broadrangeof situationsratherthaneitherunspecifiedor consistingof several
disjunctions.Dejavu could be incidentsof extraordinarilyrich activationsof these
prims, which, however, do not serve to lock in some particularmemory.
More generally, s-p-primscould representcontexts in the sense of classes of
mental states that have broad, uniform affects on other aspects of processing.
Thus, s-p-primsmightbe involvedin the implementationof reliableserialprocess-
es, markingphases of a deductionor inference, say. The meaning of those s-p-
prims might be indescribableoutside the connectivity of the activationnetwork
in terms of preparingand suppressingpossibilities. In other cases, a p-primmay
encode ambiguousbut not useless informationaboutthe environmentwithoutthe
overhead of becoming assertions or even conscious hypotheses. Whereverthey
come from and whatevertheir functionalniches, a rich vocabularyof mentalcon-
texts would seem extraordinarilyuseful to building fluid and rapid control
mechanisms.
From a more developmentalpoint of view and, once again, closer to the level
of empiricalinvestigationrepresentedhere, those p-primsthatconstituteessential
39Thereis a subthemein this model of alphabetknowledge that is suggestive of problems of us-

ing intuitive knowledge. If humanshave these alphabetchunk prims, why do they not use them in
other circumstances,for example, as redundantcodes for recalling the spellings of words? The rea-
son may be that the only mechanismof access involves the time-consumingact of mentally singing,
a complexcoordinationof metricalandotherinternalsensationsandresponses.Althoughthe knowledge
is there, the mechanismof access is so clumsy as to be practicallyunworkablein general situations.
Alternatively, humans may simply never have learned that such a bizarre thing as singing a song
of letters might be useful in memorizingspellings. In either case, mechanismsof access limit useful-
ness of the knowledge either throughintrinsicclumsiness or throughlimited, context-specificcuing.
OFPHYSICS 183
AN EPISTEMOLOGY
TOWARD
partsof thesenseof mechanism

atonestagemightmergeintothebackground
of internallyusefulbut not legitimizedhigh-priority elementsof the next. So,
for example,agency,prunedandrefined,servesa rolein the distributed encod-
ing of F = ma. It tags a member of the pairof entitiesinvolved in a situation
of force.Thistaggingmayfunction,forexample,to allowretrievalof themem-
ber for its role in laterstagesof problemsolving,say, whenforceneedsto be
computedandpropertiesof the agent,suchas its distancefromthe patient,are
relevantto thatcomputation. Despitethisproductiverolein theimplementation,
agencyplays no sanctioned role in the explicitscientifictheoryof Newtonian
mechanics.In general,conceptualchangemayor mustreuseelementsfrompri-
or systemsin substantial ways,yet it maychangetheoverallcontextof operation
enoughto renderelementsof the previoussystem,by themselves,meaningless.
If we interprets-p-primsas symbols,theyarenotat all arbitrary butaredepend-
entontheirmentaloperating context;hence,thatcontextconstitutes a strongcon-
strainton the development of new symbols.Thestrongversionof thisprinciple
of continuityis thatthe symbolsthatconstitutethe encodingof expertisemust
be builtgraduallyout of the operatingsymbolsof intuitivephysics.
In a similarway, s-p-primsneednot be generatedas meaningful,abstracted
descriptionsof observedphenomena,as hasbeentheprototypeto thispoint.To
talkof description in termsof elementsandattributes mightimplyuniformmecha-
nismsfor localizationandattribution thatprobablywouldnotexistat lowerlev-
els of a knowledgesystem.One shouldmoreproperlytalkaboutattribute-like
aspectsof internalstatethanimplythatthesementalstatesmayapplyto a broad
rangeof objectsor thatthe knowledgesystemhas the capabilityto determine
whatobjectof a sensoryarrayis responsiblefortheattribute.Certainlyit makes
littlesenseto applytermssuchas true,false, or evenappropriate to anarbitrary
mentalactivationengenderedby a sensoryexperience.
Thelong-termprogressof thisperspectivedemandsexplicationof anygener-
al mechanismsthatmightexist (description,attribution, retrievingbinding,and
so on)andof howtheydevelopandareimplemented fromsimpler,epistemolog-
icallyneutraloperations.Needlessto say, thereis no convincingaccountof this
class of processesin neutralterms,4 so I have not complicateddiscussionby
avoidingwhatI taketo be less problematic termssuchas abstraction anddescrip-
tion. In the long run,however,suchtermsare subjectto the samecriticismsas
fact and theory.
Clarification
of EmpiricalContext
The elaboratedstructural
modelmightseemto be intendedto providea micro-
mechanicalmodelof cognition.Certainly
manyconnectionists
havethisintention,
4?See,for example, Norman'scommentson type-tokendistinction,variables,and otherdifficulties

of connectionistmodelingin the conclusionof McClelland,Rumelhart,and The PDP ResearchGroup
(1986).
184 DISESSA
and, for example, similar "spreadingactivation"models (Anderson, 1984) are

intendedto accountcompletelyfor very shorttime-scalephenomenasuch as recall
and priming.
This expectation poses problems if we put it together with the fact that the
descriptionsproposed here are intended to capture aspects of protocols lasting
an hour or so. A complete micromodel of this would involve huge numbersof
nodes. Prioritiesexpressedas contingentactivationwould necessarilybe extraor-
dinarily complex. For example, the activationof an element might in principle
take place under a huge range of circumstances, each of which might warrant
description. Furthermore,the implicationof the previous discussion of mecha-
nism is that there is undoubtedlya large vocabularyof s-p-primsthat is beyond
empirical investigation in the mode of my interviewing studies.
The resolutionof these difficultiesrelies on two assumptionsaboutthe interpre-
tive context. First, note that the micro-mechanisticview allows that there may
be majormodularities;theremay be very significantslices of any particularmental
statethatare irrelevantto the futureactivationof any given elements. In particu-
lar, the critical episodes in the protocols involve judgments of plausibility41 and
judgmentsof having adequatelydrawnout causally relevantstructure.A subject
may do a lot of things to generatecandidatedescriptionsand behaviorsto judge.
She or he may consider many different versions of the problem that might cue
different p-prims, or she or he may try to find featuresof the problemsthat are
more relevant than the first-noticed ones. But, in each version of the problem
and in each hypothesizedbehavior, a judgmentmust be made aboutthe natural-
ness of some particularhappening. These judgments are, we presume, the ex-
pressionof the underlyingsense of mechanismandare mostlymodularwithrespect
to other knowledge. To the extent thatother reasoningis evident in the protocol,
we hope to captureits net effect as presentingdifferentbehaviorsor circumstances
to judge or by slightly shifting the salience (degree of activationof featurescon-
tributingto p-prim activation)of various aspects of a single situation, perhaps,
therefore, altering plausibilityjudgments.
This suggests, in fact, that sparsenetworksmight well be easier to investigate
and describe thandense ones (providingthat we can tell, at least approximately,
when they are contributing).For example, in terms of its mental encoding, the
naive sense of mechanismmay be easier to investigate than the expert sense of
mechanism.42If one can capitalize on presumedrelative sparsity and modulari-
ty of the s-p-primnetwork, describingan s-p-prim'sprioritiesrelativeto one or a
41Inthis section, I use judgment as an intuitively accessible prototypicalactivity of the sense of

mechanism, without intending to be exclusive or definitive.
42Thiscontrastswith the presumptionthat, because of its explicitness, systematicity,and closely
structurednature,it is easier to get a handle on expertise. These are not in fact conflicting presump-
tions. We may know more about expertise but, at the same time, know less about the encoding of
expertise.
TOWARDAN EPISTEMOLOGY
OF PHYSICS 185
few particularcontexts will provide reasonableinformationwithout unduecom-

plexity and commitmentto modeling all aspects of the behavior.
Second, there are steps we can take to highlightthe role of the sense of mecha-
nism in our protocols and in their interpretation.We can set up our intervention
to maximizedependenceon the sense of mechanism.We can, for example, select
questions that depend critically on judgments, not posing problems that are in-
trinsicallycomplexor ones thatinvolve muchexplicitsearchor practicedmethods.
We can use whatever we know about where and how the sense of mechanism
operatesto selectively interpretprotocols, focusing on patternsof consideration
andjudgment. We can attendparticularlyto those aspects that are the best indi-
cators of thejudgmentsbeing made, such as confidence or the basis for terminat-
ing or extending a line of considerations. We can put aside (tentatively) other
aspects of the protocol, such as strugglingwith which particularword to use or
such as the processes that generate candidatebehaviors to judge. So, for exam-
ple, the order of selecting candidatesis secondaryor irrelevantto the judgments
on them. We can, as suggested earlier, attemptto summarizelines of reasoning
into a "netresult"relevantto judgments,such as heightenedor diminishedsalience
of attributesof the situation relevant to judgments made.
To summarize, I make two presumptionsthat together make interpretations
of protocols in terms of structuredpriorities plausible. First, the modularity
presumptionis that, even if elaboratereasoningis partof intuitiveproblemsolv-
ing, still a substantialpartof the basic physics is encoded in cuing and reliability
relationsin a relatively sparse part of the active knowledge system-the partwe
call the sense of mechanism.This is tantamountto sayingjudgmentsof plausibil-
ity and similarreactionsconstitutethe core of intuitivephysics ratherthandeduc-
tion, analogy, solution methods, and so on. Second, we can bootstrapon what
we know about the operationof the sense of mechanismin setting up and inter-
pretingprotocols. We accept the fact thatthere might be many irrelevanciesand
idiosyncrasies about a particularprotocol, but we still maintainthat the kinds
of questions we ask, including the selective focus we may take on interpreting
behavior, may zero in on the critical sense of mechanism. These assumptions,
of course, will need checking as our understandingevolves.
Systematicityin WeakKnowledgeSystems
Most of the discussion here has been about specific elements and specific sys-
tematicities. In contrast, however, one can attemptto predict macrobehavioror
typicalpatterns,as opposedto particularones, based on entailmentsof the general
modeling scheme, together with gross assessments of the operatingcharacteris-
tics of the system. A huge rangeof system types may be constructedout of struc-
turedpriorities.Whatfollows are some moves to reducethe ambiguityin general,
structuralterms: guesses as to structuralcharacteristicsof the intuitive-knowledge
186 DISESSA
system abstractedfrom our empirical base. The two structuralcharacteristicsI

concentrateon have been mentionedpreviously: sparseness and the absence of
high-priorityknowledge structures.
Sparsenessin this context meansthat, althoughthe whole system may encom-
pass manyknowledgeelements, few are invokedin any given situation.Systematic
combinationsare relativelyinfrequent.In particular,intuitivephysicsis not charac-
terized by methods or extended reasoning involving the coordinatedsequential
activity of a significant number of elements. As a contrast, consider the com-
plexity involved in being able to heed advice such as, "for ballistic problems,
pick a frame of reference, choose a coordinatesystem, write down F = ma, and
solve." Instead, interactionof intuitivephysics elements is typically idiosyncrat-
ic to the situation, mediatedat most by general reasoning processes that do not
have much physics in them. Sparsenessmeans, however, that applicationof p-
prims can have the characterof insights as it will be relatively localizable, at
least in principle, and have importantorganizingeffects on the thinkinginvolved
even if it does not determine sequentialcoherence. Sparseness also means that
one cannot get a good view of the system from small numbersof contexts and
that systematicityin general must be somewhatweak, not based on coordinated
activity over relatively long time scales, like methods.43
I have alreadyremarkedthat, althoughintuitiveandexpertphysics sharemany
features, indeed, many common elements, one of the most crucial differences
is that the former lacks elements comparablein reliability to some expert ele-
ments. This contributesto the fluidity and data-drivenappearanceof the system.
No strong commitmentcan be generatedfor most interpretations.Even if such
commitment is generated, justification would be inexplicit. A change in point
of view, either by changing the problem slightly or by a spontaneousshifting
of perspective, might lead to a different interpretation.Many studentsin my in-
terviewing study frequentlytalked themselves into qualitativelydifferentunder-
standingswithoutseemingto notice. The systematicityof the knowledgeis simply
not of the quality to supportrelatively long-term coherent activity. This is not
to say that naive and novice studentsare necessarily incoherentin their reason-
ing about physical situations but that whatever coherence is imposed does not
come directly from the sense of mechanism:the coherence is not, for example,
establishinga knowledge-basedhypothesis(e.g., "Conservationof energy should
work")or pursuing specific tests ("CanI find the kinetic and potential energies
before collision?") that establish or reject the hypothesis.
More generally,one way of concretizingthe implicationsof a sparseknowledge
system with limitedhigh-priorityelementsis to say thatthereare no hypotheticals.
43iMeans-endsand other weak methods seem inappropriatefor a weakly systematic knowledge

system that, like the sense of mechanism, lacks hypotheticalsand explicit access to its elements, say,
operators. Still there must be appropriatestrategies for getting the most out of the knowledge one
has. Looking for such strategies(e.g., looking for applicablep-prims in analogies, simply repeating
an analysis over and over, waiting for alternativesto be cued, or systematicallychangingthe situation
to enhance these cuing possibilities) constitutes an importantavenue of future investigation.
For our purposes, a hypotheticalis a knowledge element that is cued in circum-

stances in which its applicabilityis not obvious and thathas an appropriatedegree
and kind of persistence to allow substantialreliability checks before it obtains
the status of an assertion. By contrast, low-priority p-prims are never cued in
a situationin which they are not more or less apparent;typically, the constituent
and cuing parts of the p-prim are greatly overlapping. And one cannot follow
up the hypothesis that a p-prim applies. If one does not know that a structure
such as Ohm'sp-prim exists, much less what its constituentsare, how can one
check circumstancesfor application?This means the system cannot supporten-
coding of strong inferences like deductions. Only when the connectivity of the
system has broadenedconsiderablyso that elements will be cued in nonobvious
situations,and when systematic, knowledge-basedmeans exist for assertingwith
confidence that a situationmust be governed by a particularknowledge element,
does it make sense for a knowledge system to bother encoding necessary entail-
ments. In contrastto Piaget'sview that hypotheticalthinkingemerges as a gener-
al capability, the present view is that hypothetical thinking is, as a default
assumption, specific to individual knowledge systems.
To say thatthere are no hypotheticalsdoes not meanthat one cannotentertain
hypotheses aboutthe behavior of a physical system. Indeed, it is one of the cen-
tral organizers of intuitive physics problem solving to consider the plausibility
of some particularbehavior.The limitationis in bringingexplicitly into consider-
ation the reasons for such behavior. Although behaviors (describedin terms of
high-reliabilityspatialor other terms) may be hypothesized, individualp-prims,
in general, may not.
I call a knowledge system with the just-mentionedcharacteristicsan aesthet-
ic. The term is meantto be evocative of the functionalcharacteristicsof rich but
structurallylimited knowledge systems, which, notwithstandingtheir richness,
appear fluid, data driven, and involve situation-specificreasoning (as opposed
to plans and general methods) and idiosyncraticjustification. These systems are
bettersuitedto providejudgmentsof similarity,of like or dislike, of relativecon-
fidence or insecurity in an analysis. They provide hunches and intuitionsrather
than pronouncementson what is true.44Similarly, aesthetics, such as a sense
of mechanism, can reasonablyencode a gradientof confidence, judgmentsthat
some ways of viewing a situation are likely to be more reliable. But they are
far from appropriatelyencoding a reductionistprogram in which a broad class
of circumstancesmust be seen as being explained by a limited class of mecha-
nisms. Thus, a novice may sense thatimprovedexplanationof an event is needed
and may search for that improvement.However, even when the search for im-
proved explanationsucceeds, which, in general, it may not, novices may settle
"44Artor literary criticism may serve as a heuristic model for such knowledge systems, hence
the name. Indeed, these might well be good examples of aesthetics in this technical sense. I believe
the meaning of such disciplines is lost unless one can project back to the p-prims of tone and feel-
ing that must be the essence of the art. Papert (1981) discussed aesthetics as it relates to mathe-
matical proof.
188 DISESSA
for increasinglymore plausibleanalogiesto understoodsituationsinvolvingsimilar

or identical p-prims. Experts' knowledge, in contrast, supportsjudgments of
isomorphism(i.e., that an event must be seen in a certainway, and events simi-
larly analyzed are not just alike but also explanatorilythe same). In such cases,
an expert terminateshis or her analysis for reasons other than having done as
well as possible in limited time or having run out of ideas.45
Are there candidatesfor knowledge systems having the characteristicsof aes-
thetics other than intuitive physics?
"*I suspect that knowledge about interpersonalrelations is structurallyvery

similar, with abstractionsfrom social phenomenaprovidingthe phenomeno-
logical primitives. People perceive threatsand friendlinesson the basis of
inarticulatecues, and they take actions to cause other people to do things
as if those actions stemmedfrom a sense for the psychological mechanisms
that actually cause people to act in particularways. Some of these intui-
tions are felt to be very secure and reliable. Yet others are vague, lower
in general reliability, or insecure in their applicationto a context.
"*Possibly relatedto social phenomenology,Wendy Lehnert'sanalysesof sto-
ries (Lehnert, 1981) identify what may be the equivalentof p-prims, some
inexplicit but relatively high-reliabilitystructuringschemes involving con-
figurationsof agents and sequences of actions abstractedto levels of who
does good or bad to whom with what reaction.
"*A phenomenologyof personalfunctioning,reflectionson the stateandactivi-
ties of one's own mind, could providethe basis for an intuitiveepistemology
thatmayconstitutethe meta-knowledgepeoplehaveaboutknowingandlearn-
ing. In the same way that intuitive physics knowledge can be more or less
productive,intuitiveknowledgeaboutlearningand knowing(diSessa, 1985)
can affect how studentsgo aboutlearningand, thus, what they learn or fail
to learn. A simple example might be the feeling of losing one's train of
thoughtafter an interruptionin a conversation. In analogy to the questions
I have pursuedwith respectto intuitivephysics, we must ask how one learns
to recognize such situations,what the internaldescriptivevocabularyis that
allows recognition, and what mental actions follow such a recognition.
"*I believe that much of the heuristicknowledge of expert physicists remains
at the level of an aesthetic.
"*A class of aesthetics worth investigating involves those relatively small
knowledge systems that decide whether or not an instance is an example
45Fora contrastingview, see Clement (1986). More recent work from Clement (see Brown &
Clement, 1989) is more compatible with allowing (in experts) structuralisomorphism, ratherthan,
for example, only increasingly more plausible analogies, as the basis for causal explanations. The
distinctionbetween plausibilityjudgmentsandjudgmentsof isomorphismis importanthere in distin-
guishing aesthetics or aesthetic-likebehavior from behaviors that are producedby more systematic
knowledge systems.
of a more general kind. Family resemblanceought to be a kind of aesthetic,

the mechanismsof operationof which, if not the mechanismsof genesis, ought
to be like those operatingin the sense of mechanism.For example, in mathe-
matics, definitionsare supposedcategoricallyto determinethe set of exam-
ples of a class. But if studentspossess structurallylimitedknowledgesystems,
judgmentswill frequentlybe made on the basis of differentand inarticulate
knowledge, even if definitionsare overtly endorsed. In the way a balancing
situationmay resemble a conservationof energy situationto a novice phys-
icist, so may mathematicalstructuresbe seen throughintuitivesubstructures.
Aesthetics seem good candidatesfor empirical investigationdespite the fact
that even the highest priority elements may not appearexplicitly in verbal pro-
tocols. This is the generalizationof the remarks in the immediately preceding
subsectionaboutthe plausibilityof investigatingphysics p-primsbecause of their
sparseness. Compare cognitive processes at the level of general reasoning and
logic. There one should find a structuredpriority system so well orchestrated
that we are unlikely to be able to see the structuraltrees throughthe functional
forest. Sparsenesscan be an advantageif it allows individualelements to stand
out as, for example, insights. Anotherreason general-reasoningp-prims, if they
exist, would be hardto investigateis that such knowledge is almost by definition
applicable across a broad range of circumstances. Thus, the functioning forms
of general reasoning are likely to be very refined, remote from whatever
phenomenological roots it might have had.46
Among aesthetics, intuitivephysics is at a particularadvantagewith respectto
offeringa good empiricalwindow, becausethe phenomenologyon which it is based
is more open to inspection. The phenomenologyis in some sense out in the phys-
ical world, which makesp-primseasier to guess and describe. (The problemwith
describingphysicalphenomenologyis more to resist assimilatingelementsto con-
ventionaldictionaryor physics-class vocabulary.)Comparea phenomenologyof
personal functioning, for which we have little conventionaldescriptive vocabu-
lary. Similarly, it is easier to manipulatethe world or to describe various physi-
cal situationsverbally in order to probe a subject'sphysics knowledge than it is
to manipulateor describe internalstates or social interactions.Argumentssimi-
lar to these, althoughfrom a contrastingperspective,are given in Hayes (1979b).
INTERPRETIVE
SUMMARY
The remainderof the monographtakes a number of orienting perspectives on
this work. In this way, I review basic featuresof the set of ideas presentedhere,
commenting on strengths and limitations. First, I provide a sketch of the fun-
"46Incontrast,however, bugs revealingcontext-dependentinconsistenciesin people's logical abil-

ities (Johnson-Laird,1983) could well offer interestingopportunitiesto unpackthe encoding of logi-
cal knowledge.
190 DISESSA
damentalreasonsthatsomethinglike a sense of mechanismexists. Next, I review

the basiccharacteristicsof p-primsandthe systemof whichtheyarea part,highlight-
ing the role of the modelinglanguage,structuredpriorities,in statingandvalidating
these claims. Then I make some comparisonswith a few noteworthycompeting
perspectives. Finally, I close with brief comments on educationalimplications.
Theoretical Claims
Thesense of mechanism. Inbroadbrush,basiccharacteristics

of thenaive
sense of mechanism and its evolution into expert physics are determinedby a
number of fundamentalcognitive and physical facts. To begin, the function of
the naive sense of mechanismis as an explanatoryframeworkfor reasoningabout
the world. After all, humanshave a lot of interchangewith the physical world,
and understandingit bettercan serve them in many ways. Understandingshould
evolve toward compactness, involving few principlesthat are as general as pos-
sible. In a sense, compactnessis the essence of explanation, identifying general
mechanismsbeneathdifferences. Compactnessshould imply abstractnessin the
sense that it is unlikely deep principles could be immediatelyevident in particu-
lar details presented in each of the full range of instances of the principles. If
a breadthof situationsis covered by a single principle, one would expect much
specific knowledge (distributedencoding) that connects the principles to differ-
ent situationsof application.This specific knowledge might be particularforms
of the principles for particularsituationsor strategiesadaptedto do the work of
interpretingdiverse situationsin common terms. Energycomes in variousforms,
each of which is differently recognized and computed.
Although compactness is an in-principleadvantagefor explanatorysystems,
the physical world is, unfortunately,extremely diverse. One measure of this is
how complex it is to use Newton's laws across the full range of their application.
Fluid dynamics, to take an example, might in many respects be reducedto New-
ton's laws. Yet it is worth multiple textbooks by itself. Another measure of this
diversity is the richness empirically found in the naive sense of mechanism. A
compact sense of mechanismthat still deals with this diversity must be tremen-
dously coordinated,with a wide range of specific knowledge properlyconnected
to the compact core. If such coordinationis hard to achieve, including finding
the right abstractand general principles, the easier alternativeis a broadbut rela-
tively shallow sense of mechanismsuch as appearsto exist in physics-naive in-
dividuals.
It happensthat Newtonian mechanics is, by and large, relatively compatible
with the naive sense of mechanism. This provides a great opportunityto de-
velop expertise by revampingnaive knowledge, both to encode basic laws and
to connect those laws to specific situations. Continuityis a fundamentallearning
principle. Knowledge is complex and difficult to come by. Humansneed to build
new from old. To be sure, there are hitches in that rebuildingthat may appear
as "misconceptions."Yet, I claim, much of an expert's sense of mechanism is

reused intuitive knowledge.
Why should an expert's sense of mechanism not completely merge with his
or her other physics knowledge (like knowing algebra and, literally, F = ma)
but retain characteristicsthat make intuitive knowledge elements evident in it?
Most prominently, I believe that diversity still rules, given humanprocessing
limitations. People have limited ability to quickly compute highly reliable, yet
situation-specificinferences from a small set of principles. They need to have
levels of analysis such as legitimizedphenomenologythatgloss details of usually
ignorablephenomenain favor of quick, situation-specificreasoning.Yet, experts
must, upon need, be able to refine that specificity to more careful and general
analyses. The effect of compactnessis achievedby experts, not by reducingeach
situationdirectlyto first principlesbutby coordinatingquickerand more situation-
adaptedknowledge. Suckingfrom a strawis, in the first instance, a phenomenon
that is immediatelyfamiliar.At the next level of explanation,air pressure"causes
liquid to move"(force as moverp-prim)along "channeledpaths"(similarto guid-
ing). Beneath that, field equations determinepressure gradientsthat accelerate
elements of liquid, and retardingviscous forces are generatedthat damp poten-
tial oscillations. The reductionto more general principles proceeds only when
it is required and, even then, in stages.
Elementsand system properties. Perhapsthemostcentraltheoretical

claim
of this work has to do with the quality and grain size of elements in the naive
physics knowledge system. Here, I heuristically situate the size and quality of
p-primsin comparisonwith more familiarunits(i.e., chunksof language).Roughly
speaking,p-primsare aboutthe "size"andcomplexityof words, althoughin several
senses they are clearly smaller and simplerthanwords. In the first sense, lexical
items often have clusters of meanings; polysemy is general, if not universal, in
word meaning. P-prims are, by contrast, more comparableto a single sense of
a word; they are the smallest, context-invariantmental activations. Consider a
word such asforce comparedwith p-prims such asforce as a mover, continuous
force, force as a deflector or as a spinner, or force meaning violence. Consider
balancing compared with dynamic or abstract balance (the latter likely itself
decomposable), or comparedwith overt balancingactions such as a person act-
ing to keep a tray in balance or the nonagentive "gradualoscillatory returnto
equilibrium."See also the multiplemeaningsof resistance listed in the Elements
section.
Wordsalso benefitfrom a structuringthatallows themto be integratedsystem-
atically into larger structures.In particular,they have grammaticalclassification
attachedto them. No comparable structuringappears to exist for the intuitive
sense of mechanism. Instead, a typical relationshipin a problem scene would
be the simple joint assertion of a numberof p-prims, respecting the contingen-
cies of cuing and reliabilityamong the set. (Recall thatcuing refers to processing
192 DISESSA
that leads to activation of an element, and reliability refers to postactivation

processing that might "check"that the conditions for using the element are satis-
fied.) Even what I have referredto as a phenomenologicalsyllogism (stiffer im-
plies less motion, less motionimplies faster)may amountonly to the simultaneous
use of two relatively strong coordinate attributebindings rather than a true
deduction.
More substantially,p-primsare weakly persistent.Words, on the other hand,
by virtue of the ability to keep a word literally in mind, allow substantialreliabil-
ity, hypotheticaluse, and even explicit reasoningabouttheir use. One can easily
ask, "Isthat really an instanceof a 'teapot'?"The equivalentquestionscan hardly
be posed for, say, Ohm'sp-prim: Even if one were conscious of the notion, the
scene featuresthat cue it are not explicitly known; how would one know where
to focus one's attentionto check applicability?It has been the contention here
that these limitationsin intuitivephysics are substantialand characteristicof the
system and thatbuildingtowardsome of the capabilitiesthatwords seem to have
is one of the key developments that must occur in learning school physics.
Although debatingthe fine points of a comparisonbetween language objects
such as lexical items and p-prims is problematic without common theoretical
ground, these comparisonsbecome less problematicas one moves up the ladder
of complexity. One would be hardpressed to think of a p-primas a proposition.
Even if one accepts a form such as "Incircumstancesx, y happens,"which might
subsumemanyp-prims, the inexplicitnessof circumstancesandlack of otherreli-
ability checks make such attributesas universality,truthvalue, and participation
in general reasoningpatterns(approximationsto logical reasoning)highly dubi-
ous. Needless to say, I believe use of even higher level terms to describe the
system of p-prims, such as theory, is unenlighteningand misleading with regard
to the level of systematicity that actually exists.
The notion of structuredpriorities is a linchpin of my theory sketch. It has
played two roles. First, structuredprioritiesis a deliberatelyweak, epistemolog-
ically neutrallanguage for describing structuralrelationswithin knowledge sys-
tems. In terms of it, we can discuss such propertiesas sparseness and limited
reliability. In the Cognitive Mechanism section, I described the advantagesof
epistemic neutrality,particularlyin comparisonwith using informal terms with
vague and implicit assumptions.I have also discussed the advantagesof a weak
mechanismsuch as structuredprioritiescomparedwith otherconventionalmodel-
ing languages such as propositional calculus and production systems which,
althoughnot ruledout on in-principlegrounds,would seem to be tailoredto higher,
more systematic levels of cognition, levels that may benefit from assumptions
of uniformconnectivity of declarativeknowledge and strong reasoningengines,
such as predicate logic, that operate over it.
The second role for structuredpriorities is as a more or less direct interpre-
tive frame for expressing empiricalresults. Elementsthat appearearly in a pro-
tocol have been interpretedas having a high cuing priorityrelationshipto salient
featuresof theproblemsituation.Similarly,laterappearing elementsor onesthat

persistthroughout the protocolhavebeeninterpreted as havinghigh-reliability
relationsto aspectsof thesituation thatI havetriedto specifyin mostcases.Defin-
ingrelativelyglobalpriorities(withrespectto theparticular problem)tiedto salient
elements(technically,elementsof highcuingpriority)in the problemsceneis
ratherstronglydependenton the knowledgesystem'sbeing sparse.
The senseof mechanismis onlypartof physicscognition.Modelingit is not
the sameas modelinggeneralmethodsor strongreasoningpatterns.To the ex-
tent thatstrongreasoningpatternsare importantto the naivesense of mecha-
nism, structured prioritiesis an incompletemodel.The claim,however,is that
they are not important,althoughthey cannotbe ruledout as contributing.
To drawan analogy,the senseof mechanismis morelike a judgethanlike
a lawyer.Itsjob is to judgeplausibility,not to do the workof marshaling data
or arguments. Someonetryingto understand a situationmightinvokemanymore-
or-lessconsciousstrategies,invokingmetaphors, tryingoutdifferenthypotheses
andkindsof description,or thinkingabouta relatedproblem.Thesemovesare
not explained,in themselves,by chartingthe senseof mechanism.Instead,the
judgmentsoccasionedby thesemovesareexplainedby the senseof mechanism.
(However,I also intendsome of the functionsof these moveswith respectto
the senseof mechanismto be explained,for example,thattheyactivaterelevant
p-prims.47)
Becausestrongmethodsand reasoningare not encompassed,the sense of
mechanismevidentlydoes not capturemanyaspectsof an expert'sNewtonian
mechanics.Thisis mostevidentin thepartthatis formalizable, although,I reiter-
ate, that is hardly all of expert Newtonian mechanics. Formalizable aspectsof
mechanicsignoreatleastmanyheuristicconnections expertsuseto solveproblems
andignorethe connectionof the formalizedtheoryto experiencenecessaryto
solve problemsposedin commonsense terms.I havetriedto sketchsomeof the
connectionsbetweenmoreevident,articulable,andformalizable partsof expert
cognitionandp-prims,for example,in the discussionof the simpleharmonic
oscillator.
P-prims,it seems,lie systematically at the interfacebetweenexperienceand
formalizable physics,bothin thegeneticsense(providing animportant knowledge
basefor learningphysics)and,later,for interpreting the realworldin termsof
theformaltheoryandvice versa.Theyinterpolate betweenthe"blooming, buzz-
ing" world of sensoryexperience and the much sparserworld of conscious,ex-
plicitideas. Structured priorities constitute an attempt to see development from
the formerto the latterin uniformterms.
47Roschelle (1991) provided a model of learning to problem solve that separatesthe sense of
mechanismfrom (a) literaland articulatedescriptivecapabilitiesand from (b) specificproblem-solving
schemata. He showed, in some degree and in a particularcase, how these differentmodules interact
and how basic to understandingthe sense of mechanismis.
194 DISESSA
A gap in the present sketch, as opposed to a presumptionor choice of what

to model, is the lack of specification of the mechanisms that decide when and
how priorities should be changed. Although the model can easily accommodate
descriptions of various kinds of change, specifying exactly which change hap-
pens when constitutes a central unelaboratedaspect of the theory. To be sure,
probably useful heuristics are not hard to come by. Use begets increased pri-
ority. Conscious reductions of an explanationbased on one set of p-prims to
another based on a different set probably enhance the priority of the latter set
and reduce the priority of the former. Similarly, I have not troubledto model
exactly how activations of p-prims are read out into feelings of satisfactionor
dissatisfaction.
Structuredpriorities are adaptedto describe what may be called problematic
learning-learning involvinginitiallysevere systematiclimitations,such as sparse-
ness and descriptiveimpoverishment,or involving the need for substantialreor-
ganizationof an existing adaptedknowledgesystem such as preconceptionsreveal
aboutintuitivephysics. Contrastproblematiclearningwith skill acquisition,which
seems to work relatively reliably and has been capturedto some degree by such
learningmodels as Anderson's(1983) ACT* theory. Therehave been fewer well-
motivatedtheoreticalinvestigationsof problematiclearning, despite the fact that
deep and long-standingproblemssuch as that of generalintellectualdevelopment
may yield results when viewed as problematiclearning. More conservatively,
I hope and intendthat the presenttheory sketch will contributeto understanding
meaningful learning in which students perceive a subject such as Newtonian
mechanics to be a naturalevolution of their own sense for how the world works
rather than a competitor or a totally disconnected and abstract subject.
Finally, I have deliberately decided to model p-prims as "pointalist"nodes,
connected, nearby,to their constitutingfeaturesand, more peripherally,to nodes
that representother p-prims related by cuing and reliability relations. I do not
mean to join argumentsabout localized or distributedforms of representation
in connectionistnets by this move. I have no vested interestin nodes per se com-
paredwith p-primsbeing "resonances"in a field or other ways of modelingthem.
Assuming nodes simply seems to be the easiest first step that gets at some of
their essential properties.
Competing Points of View

The theory sketch as it stands is incomplete in view of the previous discussion
and in otherways. By nature,it is also, like biological evolution, not ideally suit-
ed to critical experiments and direct confirmationor refutation. For example,
elements are by hypothesis hard to observe; they may be essentially impossible
to manipulateexperimentallyone by one, and they are dependenton the charac-
teristics of their equally intractable"neighbors"for their meaning and implica-
tions. In this light, may one not view the data presented in this monographas
yielding to a loose descriptive frame ratherthan supportingan evolving theory?

Could any view of intuitive physics not be recast as a p-prims view?
Partof the leverage of a theory is what it rules out or makes implausible.This
section takes thattack. A numberof views exist that are nearly directly antitheti-
cal to the one presentedhere. Contrastingviews can show that the presentview
is not compatiblewith any data or with any explanationof that data. I treat one
comparison, that with the views of Michael McCloskey, in some detail. Other
cases are abbreviated.
McCloskey'simpetus theory. MichaelMcCloskeymay still be the most

widelyreadandcitedresearcherconcernedwithnaivephysics,despitethe fact
thathis workwasdoneduringtheearly1980s(McCloskey,1983a,1983b,1984;
McCloskeyet al., 1980).Hisviews, beyondthedatalevel, contrastalmostpoint
for pointwith those developedhere. He claimsthatthe core of naivephysics
is a "surprisingly articulatetheory,"whichvariesonly a bit fromindividualto
individualandwhichstronglyresemblesthe medievalimpetustheoryof profes-
sionalscientists(philosophers).
Schematically, theimpetustheoryholdsthatthefreemotionof bodiesis caused
by aninternalforce,calledimpetus,thatis impartedto theobjectwhenoriginal-
ly impelledin its course.The impetusgraduallydies away, whetherdue to an
intrinsictendencyor interference of otherinfluences,suchas gravityor friction.
An excellentexampleof impetusthinking,perhapsthe best, is foundin stu-
dents'descriptions of a verticaltoss.Students will frequently
declarethatthetossed
objectrisesbecauseof the forceimpartedto it by thetosser.Theimpetus(sub-
jects almostalwaysuse the termforce), however,graduallydies away. At the
peakof the trajectory,the impetusis exactlybalancedby gravity.Gravitythen
overcomesthe upwardimpetus,causingthe objectto fall downward.Earlyand
enlighteningcharacterizations of impetus-likeexplanations as involvinga "sup-
ply of force" were providedby Viennot (1979),followed by Clement(1982)and
McCloskey. There is no question thatstudentssometimes give descriptionslike
these. The issue is whatone makesof such data.
McCloskeyclaims that there is a circularbranchof the impetustheory.
Whenan objectis constrained to move in a circleandthenreleased,it has ac-
quired a circular impetus. That impetusis saidto impelthe objectto continue
moving in a circle for a periodof time (althoughthe objectusuallyis described
as spiralingoutward).Severalsituationsareusedto supporttheexistenceof cir-
cular impetus.A ball may be spunhorizontallyin a circle on a string, and
the stringbreaks.A versionof this is thata ball is propelledthrougha tube
bentintoa partof a circle(C-tube),andtheballemergesfromthetube.A more
thansuperficially distinctprobleminvolvesthependulumproblemintroduced in
the Systematicity section.Recall,thequestionis whatpaththebobtravelsif the
string holding it breaks. McCloskey says many answers to these follow impetus
analyses.
196 DISESSA
A finalpieceof theimpetus ina minimal

theoryaccounts wayforanapparent
toimpetus
exception theorypredictions.
Objects froma movingobject,
dropped
suchasa bombfroma plane,arefrequently notseentohaveacquired
anyimpe-
tus. This is explainedas a separateprinciple:"Carriedobjectsabsorbno impetus."
Let me beginmy critiqueof McCloskey'stheoryby notingsomeverybroad
differences.McCloskey(implicitly)acceptsconventional
andpopularepistemo-
logical distinctions as being clear and relevant. The intuitive impetus theory is
false and has been supersededby Newtonian mechanics. He accepts that it is ap-
propriateand informativeto describe intuitivephysics as a theorywithoutelabo-
rationof what, exactly, that entails. He expectsintuitivephysicsto contain
presumed relatively universally valid notions. Thus, McCloskey's ideas are an
example of what I call "theorytheories,"which claim intuitiveideas are theoreti-
cal. And McCloskeyhas, in fact, soughtto find the origins of naive ideas in direct
misreadingsof reality,in visual illusions(McCloskey,Washburn,& Felch, 1983).
In contrast,I claim thatall such intuitiveepistemologicalterms, theory among
them, are problematicin their imprecision, inexplicitness, and, most especially,
in their applicationto sparseknowledge systems such as intuitivephysics. I claim
intuitivephysics is the residue of a complex and extensive process. Simple read-
ings or misreadingsof reality will not do to explain it. Although genesis of in-
dividual elements may come closer to "readingsof reality,"the prioritynetwork
is establishedin a gradualsortingof ideas to build an abstractexplanatoryframe-
work, albeit one with some fairly dramaticlimitations of systematicity.
Let me now considerin moredetailthe differencesbetweenknowledgein pieces
and McCloskey's theory theory. I believe the case that follows is already com-
pelling, even thoughI have not explicitlyundertakento collect datathatchallenges
McCloskey.48But this expositionis also intendedto providea frameworkto sug-
gest particularlyinterestingfocused experimentsand lines of criticism that may
be pursued in future work.
How can knowledge in pieces deal with the basic phenomenon,the kinds of
predictionsthatpeople give in responseto tossing problems,pendula,or C-tubes?
For this, we need only look at those phenomenawith p-prims in mind. It is not
difficultto see in the toss many of the generallymost prominentp-primsdescribed
here. Force as a mover describes the hand-in-contactthrow part. Although this
part of the motion requires explanationfrom an expert'spoint of view, it is en-
tirely unproblematicfrom the intuitive point of view. Thus, the throw per se is
not mentionedprecisely because it is unproblematic.We should be chastenedto
481mention, but will not here pursue, a different line of criticism. McCloskey's citations from
protocols that supposedlyimplicatean impetus theory have much more charitablereadings. The key
to these readings is realizing that impetus and momentum(or energy) are, in fact, much alike. In-
deed, it is technicallycorrectto say thatforce is a flow of momentum"thataccumulatesin an object."
It is also appropriateto say, informally, that momentumcauses an object to continue moving. So,
the linear impetus theory looks like a ratherbenign use of nonstandardterminology along with a
confusion of a flow (force) with an accumulation(momentum).
notethat,althoughtheexpert'sstancetowardphysicsproblemsis to reducethem
to explicit,fundamental explanatoryprinciples,thenaiveexplainerwill general-
ly not commentat all on whatis evidentlyunproblematic andwill pursueonly
whatmay be less immediatelyassimilable.
In this light, the intuitivethinkerdoes have a problemto solve in the toss.
Thereis a conflict,posedprobablyby continuous force, in thattheballgoes up
for a while, whereasgravitywouldcauseit to go down.This conflictis some-
timesstateddirectly,andI believeit is the sourceof muchof theexplainingstu-
dentsdo. Theyneedto explainhow it is thatthe ball can act as an independent
agentagainstgravity.Of course,theyknowthisagencyhascomefromthetoss-
er, so thatfactis expressedas a transferor communication of someformor other.
Thereis otherevidentphenomenology in the toss. Thinkingintuitively,the
top of the toss fairlyexudessomeequilibrium or balancing.49Indeed,the impe-
tus or internalforcemightbe the thingthatis balancinggravity.It is, afterall,
thatwhichcounteractsand overcomesgravityat the startof the toss.
Someweakening of theimpetusis implicated duringthemotionfromthethrow
to the apex, a transitionfromovercoming(thehandovercomesgravity)to dy-
namicbalancing.In thatit is commonknowledgethatall inducedactionsdie
away(dyingawayp-prim),impetuscomingintobalancewithgravitymakesgood
sense.Finally,as theupwardimpetusdiesawaymore,gravityovercomesit and,
finally, gets its way.
Thisredescription of thetoss hasthefollowingproperties.First,it highlights
some of the prominentattributesof the situationthatare relevantto the sense
of mechanismas sketchedhere. Thereis conflictand a restrictedagencyin a
tossedobject.Second,it gives a point-by-point analysisof the processof toss,
now decomposedinto severalp-primsthathavealreadybeenarguedfor by in-
dependentmeans, notablyforce as mover, dying away, dynamicbalancing, and
overcoming.And critically,the analysissets a contextin which impetus,as
describedby McCloskey,can essentiallytie up all the loose ends by giving a
nameto theagencyimpartedto theobjectthatcounteracts, balances,andfinally
is overcomeby gravity.
Compactly,I maintainimpetusis nota systematicandcoherenttheory.It is
anabstraction particularlysalientin a relativelysmallclassof situations,of which
the toss is archetypical.Essentiallyall of impetus'spropertiesfollow fromthe
confluenceof independent p-primsthathappento applyto the situation.In this
reinterpretation,impetuswill onlybe observedoccasionally;it may, in fact, be
encodedin somedegreeseparatelyfromtheconstituting p-prims(inwhichcase,
its genesismightbe preciselyas notedhere,fromconsideration of theproblematic
aspectsof a toss). But the impetustheorywill not exhibitstrongintegrity.
49Inrecent work, we videotapeda sixth-gradestudentinsisting an object stops and "teeter-totters"
for an instantbefore reversing direction. Some college studentsin recent interviews have explicitly
implicatedbalancing at the peak of the toss as the root of believing there is an impetus-like force
in the tossed object.
198 DISESSA
Decomposition into independentlymotivatedelements is the core of my cri-

tique. Let me continue based on some of the criteria discussed in the Method
section.
1. Principleof invariance.As notedin the priordiscussionof each of the previ-

ously mentionedp-prims, we see these in other situations. We do not see an in-
tegrated impetus theory. The sound of a bell dies away; the force of a spring
may overcomethe force of gravity to toss a brick in the air (p. 139). Arguments
on the basis of equilibrium, in particulardynamic balance, are prominentin a
broad range of circumstances,notablyrelatively static ones that do not resemble
the toss in any direct way but in which dynamic balance fits appropriately.So,
I argue that p-prim pieces better fit the criterion of invariance.
Contrastingly,the impetustheory must have extensive and as yet unarticulat-
ed applicabilityconditions. In particular,the brick-on-springsituation(How far
down must you press the brick on a spring to cause the spring to toss the brick
into the air?) can be interpretedas an impetusproblem. From a physicist'spoint
of view, it is appropriateto say the issue is whetherthe springhas suppliedsuffi-
cient impetus (momentum)to the brick by the time it has extended itself so that
there is a remnantto carrythe brick into the air. (More technically,if the momen-
tum of the brick is positive at the point the spring is fully stretched, the brick
will continue into the air.) But instead of impetus in students'explanations, one
sees only dynamic balance and overcoming p-prims.
The spring in the railroadcar is a better example. (See discussion in the first
subsectionof Systematicity.)There, the gravitationalforce of the earthprovides
an impetusto the attachedobject, which shouldbe, accordingto impetustheory,
graduallyused up in continuingto compress the spring after the push of gravity
has passed. But this is precisely what studentsmiss. Instead,they see springiness
and a figural symmetrythat dominateimpetus, even when the violent, blow-like
nature of the gravitationalforce is pointed out to them and accepted by them.
As noted, impetusdoes not seem to apply frequentlyin cases of carrying.This
is an unmotivated"patch"to the core impetustheory, which McCloskeyexplained,
as mentionedearlier, as deriving from a visual illusion. Within a knowledge-in-
pieces perspective, it makes more sense. In principle, we expect fragmentedap-
plication. More particularly,if there is no violent (force as a mover)toss, there
is no evident agentive source of impetus. In addition,an independentp-primthat
unproblematicallyapplies to everyday situations, released thingsfall (straight
down), can help account for different predictions in these cases.50
50Tobe fair, McCloskey's proposalthat the "straightdown"belief comes from a frame of refer-
ence illusion seems to have been moderatedin later work. In Kaiser, Proffitt, and McCloskey (1985),
naive stagesof comprehensionare described,moreplausibly,simply as a holisticbelief thatall released
objects fall straightdown (cf. released objectsfall p-prim). The visual illusion is said to act only
to preventsome learningthatmight overcome this belief. Generally,this laterwork seems to be more
sensitive to complexities in the development of naive notions. But, for example, in citing
The principleof invarianceappliesproblematicallyto circularimpetus.A more

knowledge-in-piecesinterpretationis thatpeople have abstractedthe loose prin-
ciple that things keep doing what they have been doing (see the discussion of
empirical adequacyin Systematicity,p. 171, and also generalized momentumin
the list of p-prims in Appendix B), but people do not hold strictly and blindly
to suchprinciples.Everyoneknowsthata sling flings a rock away fromthe slinger.
Many people have a sense for centripetalforce that may be seen to pull an object
out of a circularpath. So they reason a nice compromise:An outward-spiraling
path makes more sense than strictly to adhereto circularimpetus. I pointed out
as well in the previouslycited section on empiricalinadequacythatchildrensome-
times reactvery differentlyto simulationsas opposedto the visual gestaltpresented
by static drawings. "Theories"that depend on the mode of problempresentation
are perhaps more delicate than warrantsthe use of the term.
Carriedobjects appearnot to acquire impetus in linear situations. Then why
should similarlypropelledobjects such as a ball on a string or an object released
from a turntableacquire circularimpetus in circular situations?This lack of in-
variancedemandsalternateand betterdescriptionsof impetusexplanations.With-
in the present perspective, it is easy to understandthat we may simply have
differentp-primsto deal with in differentcircumstances,such as the interference
of gravity (in the toss) andfalling (in the moving drop) that may explain part
of the difference between linear and circular contexts.
In general, there is a fine structureto the applicationand the drawn conclu-
sions of the impetus theory that we need to attendto. In the category of applica-
tion, how do we understand,beyond asserting that it happens, why carrying is
not viewed as the same thing as tossing, which would lead to impetus explana-
tions in either case? If we understandthe importanceof patternsof amplitudes,
the particularp-prims that depend on these, and competitorp-prims, the differ-
ences are easier to explain. Consideralso the fact that subjectsmay changetheir
minds on viewing a simulation.Whataccountsfor this differencein application?
In the category of drawn conclusions, why do we get spiral predictions rather
than strictly circular paths from circular impetus? This fine structureis below
the level of grain that McCloskey's theory can capture.
2. Principle of coverage. Althoughbalanceand overcomingare arguably(but
not, I think, convincingly) part of the impetus theory, springiness and figural
considerationsare not. The impetustheory leaves us with an embarrassingstate
of affairs. It appearsto cover some set of things that people say, but all the rest
is entirely disregarded.All the otherthings they say abouttosses, aboutC-tubes,
and aboutpendulahave no described status at all. McCloskey'sdata themselves
claim impetus-likeexplanationsin, for example, 51% of subjects'responses to
cognitive processing limitations as a major source of misconceptions, the view in that work is still
very far from my position. Certainlyproperties of human perceptualsystems must play a role in
the developmentof intuitivephysics, but neitherillusions nor processinglimitationsare likely expla-
nations for much of the content and detailed properties of the naive sense of mechanism.
200 DISESSA
the C-tube problem and only 30% of subjects' responses to the cut pendulum
(McCloskey, 1983b). (The figural and static answers to the pendulumproblem,
schematized in Figure 7, p. 164, are relevant here.) What are the rest of these
subjects saying to us about intuitive physics? My take is that they are showing
us other p-prims and other kinds of reasoningthat are just as much a legitimate
part of intuitive physics.
Is it reasonable that people would have developed a core theory of physics
that covers only an odd collection of projectile problems? Is it reasonablethat
what they say aboutevery other event in their experiencebears no relationwhat-
soever to their"intuitivetheoryof mechanics"?The impetustheorydoes not cover
the fact thata pencil may be balancedon a finger and returnsto equilibriumwhen
perturbedslightly. It does not cover squareorbits (Figure 8, p. 165). It does not
cover any of the qualitativerelationsthat correlatequantities, such as that more
effort begets a faster toss or a higher toss. It does not cover any notion of com-
bining multiple influences. Surely an intuitive physics must have something to
say aboutsituationsin which multipleforces or tendenciescome to impingesimul-
taneously.
The lack of coverage of the impetus theory is as striking in terms of form
of knowledge as it is in terms of content. Although we may remotely believe
that impetus explanations are theory-like, what concepts does the theory have
to offer to describe the many much less systematic things that people say?
Lackof coverageis not a strongin-principlecriticism.Everytheoryhas bounds.
But the impetus theory needs to articulateprinciples for its boundaries, and it
has an existing, broadercompetitorin knowledge in pieces, with which it may
be specifically compared.
3. Principle of continuity. The impetus theory offers no systematic account
of the developmentof intuitive physics. If an intuitive theory of impetusexists,
I claim knowledge in pieces already offers a better developmentalaccount of it
than is providedby McCloskey. Each of the p-primpieces is independentlymoti-
vated; learninglittle things such as p-primsis, on the face of it, easier thanlearn-
ing and becoming committedto a complete theory. Each p-primI have described
comes with plausible contexts of abstraction(principle of unproblematicgene-
sis); force as moveris an abstractedtoss; dying away is an assertionthatexplains
immediatelyevident patternsof fading amplitudein all instigatedmotions; over-
cominghappensevery time you fail once andpush harder.Ohm'sp-primcan regu-
late hundredsof personal events every day.
At the system level, a gradualsorting of prioritiesmakes much less claim on
strong mechanismsof developmentthan those, whateverthey are, that result in
a theory.
Continuityalso helps explain much of the vocabulary, in the sense of internal
descriptivecapability,thatseems to be involvedin impetus-likeexplanations.This
begins at the root notion of agency, which must evolve in early years of life and
extend graduallyto understandingthe inanimateworld in circumstancesin which

it is effective.
Continuityis similarlyleft out of the futureof intuitivephysics in McCloskey's
account. McCloskey is left with the fallbackof all misconceptionsresearch, that
these misconceived ideas must be confronted and replaced by alternatives.
It maybe useful... forphysicsinstructors

to discusswiththeirstudentstheirnaive
out
beliefs,carefullypointing what is wrongwiththesebeliefs.... In this way,
studentsmaybe inducedto give up the impetustheoryandacceptthe Newtonian
perspective.(McCloskey,1983b,p. 319)
He provides no discussion of the natureof the replacing knowledge, except

that it is different, and there is no discussion of the source of the materialsthat
form the replacement.In contrast,one of my most basic claims is thatan expert's
sense of mechanismis built on a fundamentalcontinuityin form and contentwith
intuitive physics.
4. Principle of dynamic. Finally, McCloskey makes nothing of the dynamic
thatmay be evident in a clinical interviewwhen studentsare left to thinkthrough
more than first guesses or are pushed or proposed alternatives. There is much
to learn aboutimpetus-its integrity, its components,and its competitorintuitive
notions-by looking at the trajectoryof extended interviews. My claim here is
that intuitive physics in many situationscan be fluid and adaptableas much as
it is robustand ungiving. I make only one empiricalreferencehere. In exploring
an analogto the C-tube,a sling in which a rock is spunin circles andthen released,
Globersonand I (Globerson& diSessa, 1984; also see diSessa, 1988) discovered
that, when prompted,many subjectspreferredan explanationthey could not them-
selves spontaneouslygenerate. When remindedof centrifugalforce, these sub-
jects combined the forward tendency of the rock with the centrifugaltendency
and proposeda 45 o outwardtrajectory.This is very remote from impetusin any
form. The 450 prediction combines influences. The fact that many subjects
preferredthis predictionto any of their own spontaneousideas implies a degree
of flexibility not hinted at in theory theory proposals such as McCloskey's. (In
contrast, many other suggestions were rejected by subjects out of hand; there-
fore, rigidity and flexibility depend on the nature of the perturbation.)
People are not "pre-(Newtonian)revolutionary"scientistswith systematicthe-
ories of their own about how the world works. Nor are they incompetent,pos-
sessing a bundle of powerful misconceptions from predilections or from
misreadings of the world. Instead, they possess a rich and sometimes flexible
sense of mechanism drawn from years of sorting throughdiverse experiences.
Their sense of mechanism can offer predictionsin extremely varied situations,
some of which are enough beyond normal interest and outside normal contexts
of applicationof their expertisethattheir ideas appearmisconceived. These same
202 DISESSA
ideas, however, are powerful enough to provide a basis for developing expertise
in physics.
Other comparisons. A less strongbut similarlystrikingcomparisoncan be

made to Siegler's work concerningchildren'sconceptionsof such physical situa-
tions as relative speed and durationof motions, and the balance scale (Siegler,
1978; Siegler & Richards, 1979). He views early developmentalstages in terms
of a failure to encode explicitly encodable dimensions (e.g., distance from the
pivot on the balance scale). He does not seek to understandwhether and how
children spontaneouslyassess those dimensions as relevant to the issue at hand
or how recognizable phenomenology (such as pushing perturbsa balance) can
contributeto those assessments. Siegler charts developmentin terms of the ac-
quisition of a small numberof rules, on the order of a half dozen or fewer. Evi-
dently, his estimateof the size of the evolving knowledgesystem is very different
from mine. Even a modularslice of the sense of mechanismthatdeals with balanc-
ing, equilibrium, weight, and distance must be substantiallybigger than this. If
interpretedas p-primsrelevantto a particular,narrowlydefined context, a small
numberof rules makes sense. However, such an interpretationis dubiousin view
of Siegler's aims at rigorizing Piaget'srough observationsconcerningmajor de-
velopmental changes in intellectualcompetence, such as formal operations. An
interpretationmore in line with my point of view is that Siegler chartedthe sur-
face of a much richer and broaderdevelopment. Whetherp-prims or rules are
involved is partly an issue of choice in modeling language and partly an issue
of the form and substanceof knowledge, as explained in the Cognitive Mecha-
nism section.
The reasons I characterizethis comparisonwith Siegler as less strong than
the one with McCloskeyare the differencesin age level, in broadtheoreticalframe
(developmentas opposed to learning), and, possibly, differences in problemsin-
vestigated. Thus, it is not clear there is a conflict so much as a concern for differ-
ent issues, which gives rise to answers of substantiallydifferent form. It is not
my impression that this is the case, although it is impossible to argue the issue
in this abbreviatedcontext. However, I note two studies that are relevantto the
issue of whether or not one should view Siegler's tasks and focus as different
enoughto explaindifferencesin conclusions. Kliman(1987) shows a much richer
phenomenologyand more flexibility in conceptualizationof the balance scale if
subjectsare allowed more open interaction.Metz (in press) demonstratessimilar
fine structurein local conceptualdevelopment, and she implicatesjudgments of
causal relevance, notjust encoding or acquisitionof rules, as an importantfactor
in development.
Finally, Forbusand Gentner(1986) sketchedan ambitioustheoreticalperspec-
tive on the developmentof common sense and scientific understanding.To be-
gin, there are some commonalitiesin point of view. They stressedthe importance
of simple abstractionsof events ("proto-histories"). They used "processes"(which
I interpretas behavioralp-prims) as the basis for naive physics. They noted the
importanceof qualitativeproportionalities,such as "Biggerthings are slower."
They also stressedcoherence-drivenlearning-the sort thatprobablydrives some
of the systematicities I find in the naive sense of mechanism.
The contrast between their view and the knowledge-in-pieces view is most
strikingat the earliest and latest stages of competence. According to Forbusand
Gentner, young children evolve their first senses of lawfulness out of rich and
essentially literal memories of real-world events. In contrast, my view is that
p-prims constitutethe basic encoding of the naive sense of mechanism. They are
in no way literal, and they need no such precursors. They are not individually
"rich,"in any obvious sense, althoughthey are likely to be abstractedin regions
of a child's current knowledge that are rich. P-prims constitute impressions of
lawfulness, but they must exist in large numbersat the earliest stages of compe-
tence with the physicalworld. Empiricalwork showingbabies'sensitivityto differ-
ent causal scenariosreinforcesthis last presumption.Babies seem to be surprised
at scenarios apparentlyshowing interpenetrationof solid objects or at scenarios
of causality without physical contact (Leslie, 1982, 1984; Spelke, 1991).
The very notion thatrich and literal descriptionsof relevantevents exist prior
to any sense of mechanismruns counter to what I take to be a typical develop-
mental pattern-that involving the accumulationand coordinationof p-prims to
achieve reliable (hence, retrievable)descriptions.51 If young childrenhave strong
literal-memorycapabilities,my contentionwouldbe thatthese only concernaspects
of events for which thereexist strongandelaboratedescriptivecapabilities,aspects
such as static spatial deployment. If a sense of the underlyingmechanismis in-
volved in the memorial reconstructionof an event, there is ample evidence that
childrensuffer from serious inabilityto recall events literally. One simple exam-
ple, among many in Piaget (1974/1976) and Piaget and Inhelder (1966/1971),
is that young children will draw the path of a tiddlywink through the air, im-
mediately after watchingit, in horizontaland vertical segments (see Figure 9A).
Anotherexample, to which I made reference in the section on figural primitives
(p. 165), is that children will draw the level of water in a tilted jar as in Figure
9B. These I take as strongly suggestive that a (mistaken) sense of how things
work can often predate and undermineliteral memory.
The contrastbetween Forbus and Gentnerand myself concerning expertise,
the other extreme of mechanical understanding,is also enlightening. They see
quantitativecapabilities, such as the ability to use algebraic formulationsof the
laws of nature, as the last and highest stage of achievement. I have presented
little data to argue specifically againstthis view, althoughit is clearly not in the
main line of developmentI have been chartingfor the sense of mechanism. One
bit of evidence from my MIT interviewsis, however, particularlyrelevant.None
of the studentsin my study, all A or B studentsin MIT freshmanphysics, had
S1See also Minsky's (1980) discussion of the evolution of literal memory.
204 DISESSA
(A)
(B)
FIGURE 9 (A) The path of a tiddlywinkas drawn by a child watching it. (B) Water level
in a tilted jar, drawn by a child.
the least difficulty with algebra or other mathematicsinvolved with the course.
Yet, many still had substantialdifficultieswith simple qualitativeanalysisof phys-
ical situations. They also had difficulties applying equationsthat were appropri-
ate and thatthey rememberedcorrectly. This suggests that precise mathematical
competence appears last only because it requires meaningfuluse of equations.
If, as seems likely, meaning comes from appropriatequalitativecausal concep-
tions, then, at least for this class of student, quantitativeskills in physics follow
trivially on qualitative ideas.
I mentioned an agreement with Forbus and Gentner on the importance of
processes in naive physics. Yet, even here thereare substantialdifferences. First,
reliance on processes defines a stage of development for Forbus and Gentner.
It is difficult to understandwhat this stage correspondsto in my sketch of de-
velopment, because both very early stages and later ones involve extensive use
of non-process-like p-prims (e.g., figural prims in the former case and, in the
latter case, abstract balancing and the situational, geometric causation needed
to understandNewton's third law). More deeply, Forbus and Gentner'sstages
are defined by ontologies. The second stage is defined by the causalityof a sharp
class of correlations(i.e., cause statements).The thirdstage is defined by a simi-
larly sharpclass (i.e., processes). I have arguedagainstthe sense of mechanism
being localizable in such a way, althoughI do not doubtthe utility of theoretical-
ly separatingkinds of p-prims.
Forbus and Gentner are concerned with two classes of reasoning. The first
comes from Forbus's qualitative process theory. The problem is that this set
represents reliable methods of computing consequences (albeit with important
limitson selectingunambiguous conclusions)basedon relativelycomplexand

uniformdatastructures, whichruncountertotheproposed limitations
of thenonex-
pert sense of mechanism. Processes know exactlywhen they"turnon," andone
cancomputea completeset of futurepossibilities,if not selectthe one thatwill
occur(Forbus,1984).
Thesecondclassof reasoning is analogy.AlthoughGentner wishesto seelearn-
ing from complex structure mappings fromone situation
to (memories of) another,
I doubtthesearepossibleattheimportant levelof mechanism as opposedto with-
in betterdevelopeddescriptivesystems(e.g., commonsense verbaldescriptions
of objectsandtheirspatialrelations)wheresuchproblemsas persistencedo not
exist. It is difficultto see how a sparse,dynamically limitedsystemcanmanage
to computeandassessverycomplexmaps.Further,analogymaybe seenin an
entirelydifferentlightin thecontextof p-prims.Ratherthanlearningvia assess-
ing the structureof complexmaps,simpleandabstractschemessuchas p-prims
drivethe recognitionof similaritiesbetweensituations,andtheyarealsotheba-
sis of judgmentof similarmechanismsactingin the two situations.The sense
of mechanismis shallow,particular,anddynamicallylimited.It does not draw
reliable,contextindependent conclusionsbasedon generalcriteriasuchas depth
and breadthof a map.52
EducationalImplications
A theoryof knowledgeandits developmentoughtto be significantfor educa-
tion. Naturally,I do notattempthereto presentcompellingexamplesof analysis
or instructional
design.Instead,I concludethis monograph by roughlylocating
some areasof implicationsof knowledgein pieces.
1. A targetof instruction.Traditional views of learningsciencetarget,for

example,conceptdevelopment andproblem-solving skills.Theprincipleimplicit
educationalclaimof this workhas been thatthe causalsense of individuals-
whatphenomenaare seento be self-explanatory or problematic,relevantor ir-
relevantto variouscircumstances-isanalmost-separable aspectof learningphys-
ics, if notall sciences.To be sure,thiscausalsensemustbe involvedwithconcept
development andproblemsolving,butits rolein thisandevenits veryexistence
havenotbeenacknowledged. Forexample,Reif(e.g., Labudde,Reif, & Quinn,
52Thesecriticisms of structuremappingas a basis for analogies and learning from analogies are
not the pragmaticones-"Structuredependson the goal towardwhich the analogyis aimed"-proposed
by Holyoak (Gentner, 1989). Instead, I claim that there is a distinguishedand contentful (not purely
structural)"causaljudgment"module, not dependenton local goals, that cannot be written out of
analogies-the comparisonof different situationsfor the purpose of understandingphysical mecha-
nism. AlthoughGentnermight seek to take into accountcausaljudgmentsby incorporatingthem into
her model of the "initialstructureof the domain," such judgments are frequently problematicand
will, therefore, become an ongoing part of analogical processes.
206 DISESSA
1988) characterizeslearning scientific concepts substantiallyas learningproce-

dures of identificationand use; there is no role forjudgmentsof plausibility.Mis-
conceptionsor alternate-conceptions views of naive knowledgerecognize, in some
way, naive judgments but, in general, propose no role for them in expert think-
ing. Functions of naive elements in expert thinkingthat I have listed include be-
coming part of the infrastructure of knowing laws themselves (distributed
encoding), cuing instructed ideas in appropriate situations, and serving as
legitimizedphenomenology to mediate the gap betweenthe complexitiesandpar-
ticularities of common situations and general physical laws.
In additionto relatively direct instrumentaleffects relatedto problemsolving
and conceptual development, cultivatingthe sense of mechanismas an instruc-
tional target may have vital indirect importance. It may contributea backdrop
that allows bootstrappingfor, and provides stability to, developing concepts.
Concepts that are felt to be plausible and have rich connections to familiar
situationsare much easier for studentsto debug and extend on their own. The
sense of mechanismcan similarly provide a heuristic frameworkthat helps stu-
dents graduallyrefine their abilitiesquicklyto develop adequatescientific models
of situations.
Perhapsthe most devastatingimplicationof ignoring the sense of mechanism
in instructionis that buildingan unwarrantedwall between prior knowledge and
scientific understandingmay alienate students. I am convinced that one of the
most problematicpartsof currentinstructionis thatstudentsdo not feel that they
can really participatein physics instruction,that learning physics is a matterof
acceptingand memorizingcounterintuitive,if not meaningless,formulationsfrom
experts. One of the most strikingfindingsfrom the interviewingstudieson which
this work is based is that MIT undergraduates,when asked to comment about
their high school physics, almost universally declared they could "solve all the
problems"(and essentially all had received A's) but still felt they "reallydidn't
understandat all what was going on." My interpretationis that their naive sense
of mechanismwas not engaged and refined. This leaves studentsat sea both with
respect to solving qualitativeproblemsand also with respect to a feeling of secu-
rity thatthe physicsthey learnin the coursereallyrepresentshow the world works.
If I am correct that an appropriatesense of mechanism is an essential part of
experts'knowledge, these students'impressionsof incomprehensionare ironical-
ly more correct than their school assessments: They did not understand,even
though they could perform.
2. An account of learning difficulties. What do we say when students have
difficulties learning?This work provides suggestionsat two levels. First, it sug-
gests that learning difficulties are a system issue. In this, it conflicts with some
views that localize problems in a few misconceptionsthat can be individually
targetedand remediated(e.g., McCloskey, 1983b). Instead, I have pointed out
that sometimes broad and coordinatedchanges must be accomplishedto turn a
shallow, naive sense of mechanisminto an expert's. In addition, knowledge in
pieces preparesus to understandthe individualwork, and supportthat we might
needto give to it, thatmightbe necessaryin the diversityof personalsensesof

mechanismthatare likely. This contrastswith believingthatall studentshave
one particular taskto accomplish-unseating a competingtheoryto Newton,as
theory theories of intuitivephysicswould have it.
In addition,the hypothesizedsystemcharacteristics of the sense of mecha-
nismprovidefurtherfocus. Looselyorganizedandinarticulate knowledgesys-
temsposeparticular problemsforinstruction.Evokingandconsidering individual
be or
p-primsmight problematic impossible.
At thesecondlevel, p-primsprovidespecificaccountsof particular conceptu-
al difficulties.Somemisconceptions maybe the influenceof a particularp-prim
(e.g., force as a mover),andsomemightbe theconcertedinfluenceof a number
of p-prims(e.g., the impetustheory).Otherdifficultieslie in basicdescriptive
vocabularyinvolvedin perceivingmechanicalcausality,such as a shift from
agent-patient schematizations causality,ortheabandonment
to geometric of figural
primitivesin favorof local dynamicevolution.
3. Resourcesfor instructionaldesign. We may consider implicationsfor in-
structionaldesign,again,atgeneralandthenatmorespecificlevels.Atthegeneral
level, the contentionthatsubstantialnaiveknowledgeis involvedin expertise
meansthatstrategiesaimedat engagingthis knowledgeare necessary.Discus-
sions of principlesand proceduresthatdo not includereferenceto everyday
phenomena,for example,are muchless plausiblethanwouldotherwisebe the
case. Consideration of everydayevents,whichcaneasilybecomeveryengaging
andanimatedin classdiscussion,nowcanbe seenalsoto be centralin contribut-
ing to legitimatescientificunderstanding.
Thinkingabouteverydayphenomena
is notjustmakingananalogyor providinghelpfulscaffolding;it is invokingthe
veryresourcesoutof whichexpertiseis built,andit is alsoexercisinga compo-
nentof developingknowledgenot engagedin more schematicproblems.
At a morespecificlevel, the microstructure of p-primsrelatedto particular
developmentsmay describeconceptualresourcesandpositivecontributions of
naiveideas, in additionto locatingsome difficulties.We can understand how
realizingthe springinessin everydayobjectsmightprovidean essentialstepin
makingactionandreactionacceptableandworkableacrossa broaderrangeof
circumstances. We can understand the positiveroles of "primitive"
ideas such
as "heavierthingsmovemoreslowly"and"stifferthingsvibratefaster"in fluent
expertcomprehension. Butwe also understandhow the intuitiveplausibilityof
Ohm'slaw providedby Ohm'sp-primshouldnotbe mistakenfor understanding
the deeperphysicscausalityinvolved.
ACKNOWLEDGMENTS
I gratefullyacknowledgethatvariousstagesof this workhavebeen supported

by the Spencer Foundation(GrantNo. B-1393). Throughits many drafts, this
article has benefited from readingand critiqueby many people. I am especially
208 DISESSA
indebtedto MelindadiSessa, Susan Newman, and Jack Smith for their thorough
consideration.LaurenResnickprovidedhelpful suggestions, especially on issues
that needed more attentionthan originallygiven. Paul Duguid providededitorial
assistance.Commentsby journalreviewerswere greatlyappreciated.The intellec-
tual debt to others'work, both consonantand dissonantwith my perspective, is
broad and deep.
REFERENCES
Ackermann-Valladao,E. (1981). Statutfonctionnel de la representationdans les conduitesfinalisee

chez l'enfant[The function of representationin children'sgoal-orientedactivities]. Unpublished
doctoral thesis, University of Geneva.
Anderson, J. R. (1983). Thearchitectureof cognition. Cambridge,MA: HarvardUniversity Press.
Anderson, J. R. (1984). Spreadingactivation. In J. R. Anderson& S. M. Kosslyn (Eds.), Tutorials
in learning and memory (pp. 61-90). San Francisco: W. H. Freeman.
Brown, D. E. (1990, April). Facilitating conceptual change using mechanisticconceptual models.
Paper presented at the annual meeting of the American Research Association, Boston.
Brown, D. E., & Clement, J. (1987). MisconceptionsconcerningNewton's law of action and reac-
tion: The underestimatedimportanceof the Third Law. In J. D. Novak (Ed.), Proceedings of the
SecondInternationalSeminar:Misconceptionsand EducationalStrategiesin Scienceand Mathematics
(Vol. 3, pp. 39-53). Ithaca, NY: Cornell University.
Brown, D. E., & Clement, J. (1989). Overcomingmisconceptionsvia analogicalreasoning:Abstract
transfer versus explanatorymodel construction. InstructionalScience, 18, 237-261.
Chase, W. G., & Simon, H. A. (1973). Perception in chess. Cognitive Psychology, 4, 55-81.
Chi, M. T. H. (1992). Conceptualchange across ontological categories: Examples from learning
and discovery in science. In F. Giere (Ed.), Cognitive models of science: Minnesota studies in
the philosophy of science (pp. 129-160). Minneapolis: University of Minnesota Press.
Chi, M. T. H., Feltovich, P. J., & Glaser, R. (1981). Categorizationand representationof physics
problems by experts and novices. Cognitive Science, 5, 117-123.
Church,R. B., & Goldin-Meadow,S. (1986). The mismatchbetween gestureand speech as an index
of transitionalknowledge. Cognition, 23, 43-71.
Clement, J. (1982). Students'preconceptionsin introductorymechanics. AmericanJournal of Phys-
ics, 50, 66-70.
Clement, J. (1983). A conceptual model discussed by Galileo and used intuitively by physics stu-
dents. In D. Gentner& A. Stevens (Eds.), Mentalmodels (pp. 325-340). Hillsdale, NJ: Lawrence
ErlbaumAssociates, Inc.
Clement, J. (1986). Methods for evaluatingthe validity of hypotheses with analogies. In Proceed-
ings of the Eighth Annual Conferenceof the CognitiveScience Society (pp. 223-234). Hillsdale,
NJ: Lawrence ErlbaumAssociates, Inc.
Clement, J. (1987). Overcoming students'misconceptionsin physics: The role of anchoringintui-
tions and analogicalvalidity. In J. D. Novak (Ed.), Proceedings of the Second InternationalSemi-
nar on Misconceptionsand EducationalStrategiesin Scienceand Mathematics(pp. 84-97). Ithaca,
NY: Cornell University.
Confrey, J. (1990). A review of the research on studentconceptions in mathematics,science and
programming.In C. Cazden(Ed.), Reviewof researchin education(Vol. 16, pp. 3-56). Washington,
DC: American EducationalResearch Association.
diSessa,A. A. (1980). Momentumflow as an alternativeperspectivein elementarymechanics.American
Journal of Physics, 48, 365-369.
diSessa, A. A. (1982). UnlearningAristotelianphysics: A study of knowledge-basedlearning. Cog-

nitive Science, 6, 37-75.
diSessa, A. A. (1983). Phenomenologyand the evolution of intuition. In D. Gentner& A. Stevens
(Eds.), Mental models (pp. 15-33). Hillsdale, NJ: Lawrence ErlbaumAssociates, Inc.
diSessa, A. A. (1985). Learningabout knowing. In E. Klein (Ed.), Childrenand computers:New
directions in child development(No. 28, pp. 97-124). San Francisco: Jossey-Bass.
diSessa, A. A. (1988). Knowledge in pieces. In G. Forman& P. Pufall (Eds.), Constructivismin
the computerage (pp. 49-70). Hillsdale, NJ: Lawrence ErlbaumAssociates, Inc.
diSessa, A. A., Hammer, D. M., Sherin, B., & Kolpakowski,T. (1991). Inventinggraphing:Chil-
dren's meta-representationalexpertise. Journal of MathematicalBehavior, 10, 117-160.
Eylon, B., & Linn, M. C. (1988). Learningand instruction:An examinationof four researchper-
spectives in science education. Review of Educational Research, 58, 251-301.
Forbus, K. D. (1984). Qualitativeprocess theory. Artificial Intelligence, 24, 85-169.
Forbus, K. D., & Gentner, D. (1986). Learningphysical domains: Towarda theoreticalframework
(Tech. Rep. No. UIUCDCS-R-86-1247). Urbana-Champaign:Universityof Illinois, Department
of ComputerScience.
Gentner, D. (1983). Structuremapping: A theoretical frameworkfor analogy. Cognitive.Science,
7, 155-170.
Gentner, D. (1989). The mechanismsof analogical learning. In S. Vosniadou & A. Ortony (Eds.),
Similarityand analogical reasoning (pp. 199-241). London: CambridgeUniversity Press.
Globerson, T., & diSessa, A. A. (1984, July). The effect of age and cognitive style on children's
intuitions of motion. Paper presented at Logo '84, Cambridge, MA.
Hayes, P. J. (1979a). Naive physics I: Ontologyfor liquids (WorkingPaper35). Geneva: University
of Geneva, Center for Semantic and Cognitive Studies.
Hayes, P. J. (1979b). The naive physics manifesto. In D. Michie (Ed.), Expertsystemsin the micro-
electronic age (pp. 242-270). Edinburgh:EdinburghUniversity Press.
Henriques, G. (1984, October). Rencontresur les representationscinetiques [Meeting on kinematic
representations](TechnicalReport).Geneva:Universityof Geneva,International Centerfor Genetic
Epistemology.
Johnson, M. (1987). Thebody in the mind: Thebodily basis of meaning, imagination, and reason.
Chicago: University of Chicago Press.
Johnson-Laird,P. N. (1983). Mental models. Cambridge, MA: HarvardUniversity Press.
Kaiser, M. K., Proffitt, D. R., & Anderson, K. (1985). Judgmentsof naturaland anomaloustrajec-
toriesin the presenceandabsenceof motion.Journalof Experimental Psychology:Learning,Memory,
& Cognition, 11, 795-803.
Kaiser, M. K., Proffitt, D. R., & McCloskey, M. (1985). The developmentof beliefs about falling
objects. Perception and Psychophysics, 38, 533-539.
Klahr,D., & Chase, E. A. (1983). Structureandprocessin alphabeticretrieval.Journalof Experimental
Psychology: Learning, Memory, and Cognition, 9, 462-477.
Kliman, M. (1987). Children'slearningaboutthe balancescale. InstructionalScience, 15, 307-340.
Kuhn, T. S. (1977). The essential tension. Chicago: University of Chicago Press.
Labudde,P., Reif, F., & Quinn, L. (1988). Facilitationof scientific concept learningby interpreta-
tion procedures and diagnosis. InternationalJournal of Science Education, 10, 81-98.
Larkin, J. (1983). Problem representationin physics. In D. Gentner& A. Stevens (Eds.), Mental
models (pp. 75-98). Hillsdale, NJ: Lawrence ErlbaumAssociates, Inc.
Lehnert, W. (1981). Plot units and narrativesummarization.Cognitive Science, 5, 293-331.
Leslie, A. M. (1982). The perception of causality in infants. Perception, 11, 173-186.
Leslie, A. M. (1984). Spatiotemporalcontinuityand the perceptionof causality in infants. Percep-
tion, 13, 287-305.
Levin, I., & Simons, H. (1986). The natureof children'sand adults'concepts of time, speed, and
distanceand their sequencein development:Analysis via circularmotion. In I. Levin (Ed.), Stage
and structure:Reopening the debate (pp. 77-103). Norwood, NJ: Ablex.
210 DISESSA
Marton, F. (1981). Phenomenography- Describing conceptionsof the world aroundus. Instruction-

al Science, 10, 177-200.
McClelland,J. R., Rumelhart,D. E., & The PDP ResearchGroup.(1986). Paralleldistributed process-
ing: Explorations in the microstructureof cognition. Cambridge, MA: MIT Press.
McCloskey, M. (1983a, April). Intuitive physics. Scientific American, pp. 122-130.
McCloskey, M. (1983b). Naive theoriesof motion. In D. Gentner& A. Stevens(Eds.), Mentalmodels
(pp. 299-324). Hillsdale, NJ: Lawrence ErlbaumAssociates, Inc.
McCloskey, M. (1984, April). Cartoon physics. Psychology Today, pp. 52-58.
McCloskey, M., Caramazza,A., & Green, B. (1980). Curvilinearmotion in the absence of external
forces: Naive beliefs about the motion of objects. Science, 210, 1139-1141.
McCloskey, M., Washburn,A., & Felch, L. (1983). Intuitivephysics: The straight-downbelief and
its origin. Journal of ExperimentalPsychology: Learning, Memoryand Cognition, 9, 636-649.
Metz, K. E. (in press). Preschoolers'developingknowledge of the pan balance:From new represen-
tation to transformedproblem solving. Cognition and Instruction.
Minsky, M. (1980). K-lines: A theory of memory. Cognitive Science, 4, 117-133.
Minstrell, J. (1982). Explainingthe "atrest"conditionof an object. ThePhysics Teacher,20, 10-14.
Minstrell, J. (1989). Teaching science for understanding.In L. B. Resnick & L. E. Klopfer (Eds.),
Towardthe thinkingcurriculum:Currentcognitive research (pp. 129-149). Alexandria,VA: As-
sociation for Supervision and CurriculumDevelopment.
Papert, S. (1981). The mathematicalunconscious. In J. Wechsler (Ed.), On aesthetics in science
(pp. 104-119). Cambridge, MA: MIT Press.
Piaget, J. (1971a). Biology and knowledge:An essay on the relations between organic regulations
and cognitiveprocesses. Chicago: University of Chicago Press. (Originalwork published 1967)
Piaget, J. (197 1b). Thechild'sconceptionof movementand speed (G. E. T. Holloway & M. J. Mac-
kenzie, Trans.). New York: Ballantine Books. (Original work published 1946)
Piaget, J. (1972). Theprinciplesof genetic epistemology(W. Mays, Trans.). New York: Basic Books.
(Original work published 1970)
Piaget, J. (1976). The grasp of consciousness (S. Wedgwood, Trans.). Cambridge, MA: Harvard
University Press. (Original work published 1974)
Piaget, J., & Inhelder, B. (1971). Mental imageryin the child: A study of the developmentof imagi-
nal representation(P. A. Chilton,Trans.). New York:Basic Books. (Originalworkpublished1966)
Reif, F. (1987). Instructionaldesign, cognition,and technology:Applicationsto the teachingof science
concepts. Journal of Research in Science Teaching, 24, 304-324.
Roncato, S., & Rumiati, R. (1986). Naive statics: Currentmisconceptionson equilibrium.Journal
of ExperimentalPsychology: Learning Memoryand Cognition, 12, 361-377.
Roschelle, J. (1991). Students'constructionof qualitativephysics knowledge:Learningabout veloci-
ty and acceleration in a computermicroworld. Unpublisheddoctoral dissertation, University of
California, GraduateSchool of Education, Berkeley.
Roschelle, J., & Greeno,J. G. (1987). Mentalmodelsin expertphysics reasoning(Tech. Rep. GK-2).
Washington, DC: Office of Naval Research.
Rumelhart,D. E., McClelland,J. L., & The PDP ResearchGroup.(1986). Paralleldistributedprocess-
ing: Explorations in the microstructureof cognition (Vols. 1 & 2). Cambridge, MA: Bradford
Books/MIT Press.
Siegler, R. S. (1978). The origins of scientific thinking. In R. S. Siegler (Ed.), Children'sthinking:
Whatdevelops? (pp. 109-149). Hillsdale, NJ: Lawrence ErlbaumAssociates, Inc.
Siegler, R. S., & Richards, D. D. (1979). Development of time, speed and distance concepts. De-
velopmentalPsychology, 15, 288-298.
Smith, J. P., diSessa, A. A., & Roschelle, J. (in press). Misconceptionsreconceived: A construc-
tivist analysis of knowledge in transition.Journal of the Learning Sciences.
Spelke, E. (1991). Physical knowledge in infancy: Reflections on Piaget's theory. In S. Carey &
R. Gelman (Eds.), Epigenisis of mind (pp. 133-170). Hillsdale, NJ: Lawrence ErlbaumAssoci-
ates, Inc.
Talmy, L. (1988). Force dynamics in language. Cognitive Science, 12, 49-100.

Viennot, L. (1979). La raisonnementspontane en dynamiqueelementaire [Spontaneousreasoning
in elementary mechanics]. Paris: Hermann.
Vosniadou, S. (1989). On the natureof children'snaive knowledge in astronomy. In S. M. Glynn,
R. H. Yeany, & B. K. Britton (Eds.), Thepsychology of learning science (pp. 149-177). Hills-
dale, NJ: Lawrence ErlbaumAssociates, Inc.
Wason, P. C., & Johnson-Laird,P. N. (1972). Psychologyof reasoning:Structureand content.Cam-
bridge, MA: HarvardUniversity Press.
White, B. Y. (1981, February).Designing computergames to facilitate learning (Tech. Rep. No.
AI-TR-619). Cambridge, MA: MIT Artificial Intelligence Laboratory.
APPENDIXA: CASE STUDIESOF P-PRIMANALYSIS
Ohm's P-Prim
Ohm'sp-primmakesa goodcase study.It is amongthe betterdevelopedcases,
andit showsmanyof the heuristicp-primidentification principles(Methodsec-
tion, p. 120andfollowing)in action.I referencetheseprinciplesparenthetically
as theyapplyhere.Ratherthanpresentinganabstractcaseforthep-prim,I pro-
vide a narrativefor easierandequally(if not more)informativereading.The
narrative doesdoubleduty,showingdatausedin thedevelopment of thetheoret-
ical frameas well as theiruse in discovering,refining,andvalidatingdescrip-
tionsof p-prims.Of course,thenarrative is a reconstruction,
andthechronologies
shouldnot be takentoo seriously.
The discoveryof the Ohm'sp-primbeganby consideringa class of simple
everydayevents.Howdo peopleknowto pushharderon biggerthingsto make
themmove (principleof obviousness)?Initially,it seemedplausibleto me that
no p-primmightexisthere,onlyinarticulate musclecontrolschemata.However,
I anticipatedthis wouldbe an important p-primclass, becauseit is so common
andcriticalto choosingactionsin orderto accomplisheverydaytasks(principle
of functionality).It seemedevidentthata largeclassof situationswasessentially
isomorphic at this level of description,althoughsuchintuitionswarrantskepti-
cism (principleof diversity).I was at the timeconsideringdifferentversionsof
force as a mover, continuousforce, force as a spinner, and so on. These all could
use the sameprincipleto modulateeffortandresult,provideda schematization
suitableto all of themcouldbe found(principleof coverage,principleof ab-
straction).It seemedevidentthatanyof thesephenomenacouldbe a contextof
abstractionfor sucha p-prim(principleof unproblematic genesis).
Manyyearsbefore,I hadhypothesized a centralclassof causalrelationsthat
I called"causalsyntax."It involvesa trioof elements:anagentor causalsource,
a legitimizedcausalconnection,anda patient.It madesenseto trysucha tripar-
tite relation,withemphasison the amplitudeof effortor intensityin the causal
source,the amplitudeof its resultin the patient,andon somemodulating effect
in the causalconnection.Fromevidenceof intuitiveequivalentsof F = ma
212 DISESSA
a push),I beganto settleonsomeformof resistance

(a heavymass"resists" as
themediating
property,buttheexactcharacterizationof thiswasuncertain.
It was helpful that I knew even very young childrenhad the ancillaryknowl-
edgeto makesucha p-primuseful.Reasoningon systemsinvolvingproportion-
ality, inverse proportionality,and compensationhad been studiedby Piaget and
of functionality,
others(principle principleof scavenging
data,principleof
continuity).
Out of curiosity, I began to search for situationsthatviolated this p-prim. Af-
ter all, a principlethatis universallytrueshouldhavemorethanp-primstatus.
An early attemptinvolvedlookingoutsideof the domainof mechanics.I considered
electricity.Insteadof a counterexample,I ranacrossa stunningexample,Ohm's
law: Current flow in a is to thevoltageandinverselyrelated
circuit proportional
to the resistance. To that point, I do not recall having made the connection be-
in anelectricalcircuit.(Having
tweenanobject'sresistingmotionandtheresistance
wasto me a thoroughly
beenanelectronichobbyist,"resistance" deadmetaphor.)
It made a great deal of sense that if a p-prim such as Ohm'sp-prim existed,
it would be appliedby the discoverer of Ohm'slaw and frozen into the technical
vocabularyused to describe it. It also helped that the British and Germanwords
for voltage are evocativeof effort or latentimpetus,tension. Plausiblythese might
even have arisen from bodily tension (principle of the body), althoughI never
discovered how to follow up on the idea. Granted,essentially any semanticthe-
ory could rationalizethe applicationof words such as resistance and tension to
such a situation, but the use of both resistance and tension together was at least
suggestive (principleof diverse evidence). I also noted with satisfactionthatcur-
rent is an almost ideal "result"in that it is motion of a sort that would seem to
requirea cause. The descriptionof Ohm'sp-prim as involving an effort or level
of impetus (located in a generalizedagent), a resistance, and a measureof result
seemed to be succeeding.
It was also importantthat I knew Ohm'slaw is very intuitiveand easy, in the
first instance, to teach (principleof scavenging data, this time from instructional
experience). Because my orientationhad all along been that intuitive ideas are
responsible for some of the learnabilityor nonlearnabilityof various technical
concepts, now having an explanationfor thatparticularfact with regardto Ohm's
law was encouraging. This was also the first instance of an importantclass of
functions I have identified for p-prims in more expert knowledge, serving as
qualitativeapproximationsto the technical concepts.
I looked throughseveral texts to see if I could find examples of explicit invo-
cation of Ohm'sp-prim. It seemed to me that sensitive instructorswould make
use of it. Not only did I find such uses, but one of the texts I checked also con-
tainedan explicit warningthatthe schematizationof Ohm'slaw as a resultdriven
by an impetus, modulatedby a resistance, is only metaphorical,as, indeed, it
is. It is only a gloss on an expert'scausal view of Ohm'slaw. (See relateddiscus-
sion in diSessa, 1988.)
Soon, however, I had a clear counterexampleto Ohm'sp-prim. Vacuumclean-

ers speed up when theirnozzle is covered, whereas, intuitively,the coveringhand
would appearto providea clear resistanceor interference.Thus, I expectedto find
thatpeople would predictthe pitch would go down when the nozzle was covered.
A few of my early interviews were successful in this regard, but there were
complications. First, many people had strange ideas about how pitch is deter-
minedin this situation.They believedsome sortof musicalor resonancephenome-
non was responsible. In the end, I salvaged some such subjectsby simply telling
them that the pitch is essentially the speed of the motor, so that they could pro-
ceed with the core causal analysis in which I was interested.
Anothercomplicationwas that many people simply rememberedthat a vacu-
um cleaner's pitch increases. Many of these also could be salvaged by asking
them why such an increase occurred. Some subset of these gave unanticipated
corroborationof the applicationof Ohm'sp-prim to the situation. They asserted
that an increased resistance or interferencewould cause the pitch to go up be-
cause the motor would have to work harderto "try"to compensate. Thus, an in-
telligent impetus (thatunderstandsOhm'sp-prim) could attemptto make up for
an increasedresistanceby increasingits own efforts. A few subjectseven retracted
their assertionthat the pitch would go up when I questioned, and they admitted
as questionablethe implied assertionthata motorcould know to try harder(prin-
ciple of dynamic). These short episodes underscore, if it needs emphasis, that
answers per se should only with great care be takento relate directly to p-prims,
"misconceptions,"or any encoded knowledge.
"Workingharder"to some extent threatenedmy Ohm'sp-prim interpretation
of the vacuumcleaner. Some subjectsnever appearedto unpackit into its (I con-
sidered) Ohm'sp-primjustification. But in the end, I considered working hard-
er's existence to be a confirmationthat motors are viewed as generalizedagents.
Thus, agent-likephenomenaother than Ohm'sp-prim should be on tap in situa-
tions involving motors and the like (principle of ready availability). The more
I discovered the diversity of p-prims that studentsused, the less I was disturbed
by variations, and the less I expected to see p-primsunpackedalways into, what
seemed to me to be, their evidentjustificationor even their intuitiveroots (prin-
ciple of diversity, principle of impenetrability).
I puzzled for an extended time over two facts. The first was that, in order
to cover all the instances I found of it, resistanceor interferencehad to be very
general categories (principleof invariance).Partly, I became less worried about
this as I accumulatedmore and more p-prims that seemed to have comparably
wide scope. Partly, as it became more evident that Ohm'sp-prim is a centralele-
ment, it seemed to me more likely and less importantthatthere mightbe disjunc-
tive cues to the inversely modulating quantity. Whether it is interpretedas
interferenceor resistance,all the cued proportionaland inverse proportionalcon-
nections among "effort,""resistanceor interference,"and "result"can do the es-
sential work of Ohm'sp-prim.
214 DISESSA
The second puzzling fact was that the impetus level of the motor was appar-
ently confounded with an indicatorof effect. In the workingharder interpreta-
tion, pitch indicates effort on the part of the agent. In the "thevacuum cleaner
pitch slows down for increasedresistance"interpretation,the pitch is an indica-
tion of decreased result. Several things lessened my concern. First, it became
clear that reliably connecting p-prim slots to localizable features in a situation
is a difficultaccomplishment,implyingmore structureto p-primsthanone should
ordinarilyexpect. One should expect slipping and sliding in attribution.Also,
it seemedto me agency is frequentlyambiguousin this way. Resultsare frequently
taken as measures of (less perceptible)effort. Finally, it seemed to me that the
vacuumcleaner situationcontainedsuch a good exemplarof interference,a simi-
larly good exemplarof impetus(a motor), thatexactly how one madethe attribu-
tions to the particularsof the situationwas irrelevantto the fact that all of the
aspects of Ohm'sp-prim evidently existed, so that it would be seen to regulate
the situation in any case.
More generally, I came to accept that p-prim analyses may not involve any
single, clearly drawnmodel of the situation. For example, in instancesof appli-
cationof Ohm'sp-prim, I ceased to expect to find a clear demarcationof (a) agent,
(b) location and form of impetus("effort"or tension) in the agent, (c) an articula-
ble causal(perhapseven spatial)topologyconnectingthe agentthrougha resistance
or interferenceto (d) an evident patient, which was (e) the locus of a resultmeas-
urement clearly independentof the impetus. Rough but reliable cues to Ohm's
p-prim, and ready features to plug into relations of effort and result, were all
that I felt were needed.
For the most part, althoughit entailed a numberof complicationsthat had to
be sorted out, I came to consider the vacuumcleaner good verificationof Ohm's
p-prim at the same time that it refined my expectationsabout what exactly con-
stitutedthis knowledgeelement (principleof discrepancy,principleof invariance
in the form of predicting a "misconception").
Subsequently,I became awarethat Ohm'sp-primmightbe implicatedin inter-
personal interactions, such as influencing and convincing. In his work on force
dynamicsin language, Talmy (1988) made similarobservationsthatfurtherrein-
forcedthe case for the realityof one of the startingpointsfor Ohm'sp-prim, causal
syntax.Talmyused termsratherdifferentfromagent andpatient, butthe schemati-
zation is the same. Talmy also heightened my appreciationfor the special role
of result. In effect, result defines patient as the locus of effect of the (generally
unaffected) agent.
Several other pieces of research in recent years have indirectly (via "strong
vocabulary")addedmoredatato the case for Ohm'sp-prim,particularlywithregard
to the almost anthropomorphicattributionsOhm'sp-prim makes with regardto
agency of inanimatethings. Work by Minstrell (1989) and Brown and Clement
(1987) continuedto indicate that agency in various forms is a central attribute
in intuitive physics. Minstrell also developed data showing that very young
childrencan reason specifically aboutthe effect of mass inforce-as-mover situa-

tions (personalcommunication,July 1987). Again and again in my own analyses,
agency has been implicated. For example, see the discussion of action and reac-
tion (p. 151 and following). In addition,agency is a featureof children'sthinking
that Piaget emphasized(principleof continuity),althoughinterpreteddifferently
and in a very differentoverall theoreticalcontext. Finally, some of my own work
with small children (Globerson& diSessa, 1984) showed some extreme exam-
ples of anthropomorphicand animistic thinking. With such evidence for crude
precursorsof the more refined agency shown in Ohm'sp-prim, I became much
more comfortablewith refined versions of agency and effort showing up in an
adult p-prim.
Because it seems like a powerful methodologicalmove, I mention two other
examples comparableto the vacuum cleaner'srole in the case for Ohm'sp-prim,
predictingmisconceptions.First, the railroad-car-in-spaceproblem(see the first
subsectionof Systematicity)was inventedto satisfy two criteria. (a) It is appar-
ently an occasion for apt applicationof impetus ideas to mechanical situations
if such well-integrated conceptions exist and dominate intuitive physics
(McCloskey, 1983a, 1983b). (b) I believed a different (p-prim) analysis would
better capturesubjects'answers. I was confident figural and springinessp-prims
would dominate impetus in this context.
Similarly,if my p-primdecompositionof the impetustheoryis correct,I should
be in a good position to predict nuances in available data. So, I expected, for
example, that simply releasing a body to fall would not elicit impetus explana-
tions nearly as much as tossing it up into the air. I was able to confirm these
predictions,roughly, by checking with researcherswho had researcheddrop and
toss problems(J. Clement,personnelcommunication,July 1986, referencedbriefly
in Clement, 1983). Both the railroad-car-in-spaceand dropping (as opposed to
tossing) problems are good examples of the power adequatedescriptionshould
provide via the principle of invariance.
The Bell
My first encounterwith the Montessoribell conundrum(see subsectionon bells
in the Developmentsection), thatthickerbells of the same otherdimensionshave
a higherpitch, was instigatedby a colleaguewho was puzzledby the fact. It seemed
immediatelyevidentto me thata p-primmust be involved, becausepeople'sreac-
tions were so swift and sure (principle of obviousness). In fact, in some ways
this behaviorseemed exceptionalin that, given such a counterintuitiveresult,very
few people seemed to have any other way at all to think about the problem (in
contrast to the principle of diversity). As my analysis progressed, however, I
did find significant diversity in and across individuals.
My first take on the featuresinvolved was thatpeople were reactingto mass.
A greatermass in a harmonicoscillator leads to reduced frequency. I soon dis-
216 DISESSA
covered, however, that this was only true for a relatively small percentage of
the populationI interviewedaboutthe problem. Most seemed directly to impute
size as the controlling influence. This made sense in retrospect;it is more evi-
dent than mass (principleof ready availability). Mass would be more expected
from physics-instructedindividuals.
I was struckby how limited many people's penetrationinto the problemwas.
I discovered that some subjectshad evidently abstractedthe p-prim from direct
phenomenologyof size and pitch; a (informaladult) subjector two were entirely
unaware that pitch had to do with frequency and vibration, yet were as secure
as others in the belief that a bigger object should have a lower pitch (principle
of impenetrability).Others did mention weight as probablythe controlling fac-
tor. As with resistanceand interference,I have not consideredit critical thattwo
versions of the bell p-primmight exist: one involving size, one involving weight.
I suspectedthat, for those who respondedto the weight connection, it would be
consideredmore fundamental.Some informaldata led me to believe that weight
did have a higher reliabilitythan size for most who had respondedto both attri-
butes (principleof dynamic), althoughit is not entirely evident why this should
be so. After all, taking wavelength to be the determinerof pitch is physically
as appropriateas takingmass. Perhapshigh school physics does not teach waves
as well as it teaches F = ma. Nonetheless, the mass connection seemed to have
greater reliability priority.
Although most subjects were ready with analogies-church bells compared
withjingle bells, xylophones, musical instrumentsof various sizes- I was struck
that some initially could not produceany example of the phenomenonthey iden-
tified to be at the root of the situation.This, along with the rapidityand expressed
certaintyof responses, heightenedmy confidence that a p-prim (or several) was
at stake ratherthan analogy (principle of content over form).
The p-prim richness of the situationbecame more evident when, in later in-
terviews, I pressed people for plausibleexplanationsfor the fact thatthickerbells
do have higher pitch. A few reactions are instructive. Some people responded
thatthe bell would "notvibrateas much"(apparentlyreferringto amplitude)be-
cause it was thicker. I did not pursuethe issue of whetherthis was becausethick-
er bells are more massive and so resist much motion or whetherthis was a result
of stiffness. Because the bell did not move far, it could do so in less time. At
first, these reactionspuzzled me; I did not understandthe logic. If it was being
restrainedfrom moving, surely it would move more slowly as well as not mov-
ing as far. RereadingPiaget on time, distance, and speed problemswas instruc-
tive (principleof continuity, scavenging data). Young children seem separately
to encode very simple heuristics, such as if an object is ahead, it musthave gone
faster, or, more relevantly, if an object covers a reduceddistance, it takes less
time. (This p-priminterpretationis, of course, dramaticallydifferentfrom Piaget's
interpretation.)At least young children systematicallyignore the contingencies
that make these conclusions justifiable, such as the presumptionof equal speed
that would seem to have to underliethe logic of less distance implies less time.
Adults may also invoke these simple p-prims occasionally, when pressed.
The apparentuse of less movementimplies less time also underwrotethe rela-
tive impenetrabilityof p-prims and the fact thatthey may frequentlybe used in-
dividually ratherthan in a reliable package of caveats and contingency checks.
This also underlinedthe contrastto experts' applicationof scientific concepts.
I also became less concerned about the fact that attributingvery schematic p-
prims to adults made their causality seem primitive. A sense of mechanism is
elastic. It grows by considering possibilities that are plausible only in virtue of
relativelyunfilteredand abstractphenomenology.Low reliabilitypriorityp-prims
that help make everyday experiences immediately familiar may sometimes be
pressed into use in the place of more highly reliable p-prims.
APPENDIXB: LISTOF P-PRIMS
Ohm'sp-prim
"*Schematization:An agent or causal impetusacts througha resistance
or interferenceto producea result. It cues andjustifies a set of propor-
tionalities, such as "increasedeffort or intensity of impetus leads to
more result";"increasedresistanceleads to less result."These effects
can compensateeach other;for example, increasedeffort andincreased
resistance may leave the result unchanged.
"*Key attributes: Resistance or interference, agency.
"*Prototypical circumstances: Pushing a box with variable effort on
different surfaces.
"*Relation to schooledphysics: Reused in Ohm'slaw. Glosses F = ma,
with the force representingthe causal impetus, m the resistance, and
a the result.
"*Comments:Centraland very broadlyapplicable, from many physical
to interpersonalrelations such as influencing.
Force as mover
"*Schematization:A directedimpetusacts in a burston an object. Result
is displacementand/or speed in the same direction.
"*Attributes:Violence.
"*Circumstances:A throw.
"*Relationto schooledphysics: Glosses F = ma, but only from the state
of rest. Responsible for "thingsgo in the direction they are pushed"
misconception.
* Comments:InvolvesOhm'sp-primin reasoningabouteffect of impetus.
218 DISESSA
Force as deflector (cf. force as a mover)

"*Schematization:A shove may act in concert with prior motion (mo-
mentum) to produce a compromise result, directionallybetween the
two.
"*Relation to schooled physics: May be a relatively low-priority p-
prim "encouraged"by instructionbecause it is more compatiblewith
F = ma.
"*Comment:Frequently, subjects explicitly justified this, the evident
deflection (after the fact), as a "compromise"in dynaturtlesituations
(diSessa, 1982). As many "combinedeffects" ideas, this seems to de-
velop later and to have lower priority than categorical ideas ("the
stronger influence gets its way").
Continuousforce
"*Schematization:As force as mover, but involving constant effort.
"*Attributes:Steady effort.
"*Circumstances:A car engine propels a car.
"*Relation to schooled physics: May gloss F = ma. But when the result
is takento be speed (the early-on case) ratherthan acceleration(more
sophisticated), it accounts for misconception of "motion requires a
force."
Force as a spinner
"*Schematization:Off-center pushes create spinning.
"*Circumstances:Especially salient in cases of circular symmetry.
"*Relation to schooled physics: Glosses torquelaws but also undermine
plausibility of linear F = ma in such circumstances. Studentsthink
forces that create spin cannot simultaneouslycreate linear motion or
have a reduced effect in creating translation. This latter idea seems
to involve a kind of principle of conservation of effect.
Intrinsic or spontaneous resistance (see force as a mover)
* Schematization:Especially heavy or large things resist motion.
Interference
"*Schematization:Influencesthatdo not directlyaid or conflict may still
interfere.
"*Circumstances:For example, gravity interferes with horizontal mo-
tion (may explain dying away in such circumstances).
"*Relation to schooled physics: This constitutesan impediment,but ap-
parently not a great one, for independenceof orthogonal forces.
"*Comment:Interferencemay be causallyevident(e.g., a handon a rotat-
ing drill chuck) or imputed (e.g., gravity interferingwith horizontal
motion).
Dying away
"*Schematization:All motion, especiallyimpulsivelyor violentlycaused,
gradually dies away.
"*Attributes:Fading amplitude.
"*Relation to schooled physics: Implicatedin impetus misconceptions.
It underminesthe Newtonian principle of constant motion in the ab-
sence of force in the same way that continuous force does.
Workingharder
"*Schematization:More effort or cues to more effort may be interpreted
as if in an effort to compensate for more resistance.
"*Circumstance:Attributionto higherpitch, loudernoise from a clogged
vacuum cleaner.
"*Comment:This seems to be a relativelyprimitiveanthropomorphicas-
sociation, but I have observed it in many adults'reactionsto the vacu-
um cleaner problem.
Change takes time ("warmingup")

"*Schematization:Changes take time to "blossom."
"*Attributes:Crescendo.
"*Circumstances:Acceleration from cannon shot continues after shell
escapes the barrel.
"*Relationto schooledphysics: Underminesinstantaneouscausality, for
example, in F = ma.
"*Comments:Probablyrelatesto a collection of "gradualness" p-prims-
thatrapidchangesrequiresevere or violent intervention.Subjectsreact
to rapid change especially in dynamic visual presentations,such as a
simulation.May be less salientin staticpresentation,for example,draw-
ings of angled trajectories.
Vacuumsimpel
"*Schematization:Emptiness requires filling.
"*Circumstances:Sucking.
"*Relation to schooled physics: This p-prim must defer to forceful ex-
planation;an outside influence mustpush things into evacuatedspace.
"*Comment:May be culturalto some extent. It is obviously sanctioned
by "Natureabhors a vacuum."Consider an extension: How do chil-
dren explain the fact that sand fills in scooped-out space?
The ConstraintCluster
The following four p-primshave been studiedin less detail; descriptionsare less
certain. Every member of this class must be underminedin instructionbecause
forces must come to explain all these circumstances.
220 DISESSA
Bouncing
"*Schematization:An objectcomes into impingementwith a big or other-
wise immobile other object, and the impinger recoils.
"*Relation to schooled physics: Bouncing must cease to be primitive,
come to be seen as macro-phenomenoninvolving springinessand (in-
tuitive versions of) F = ma.
Supporting
"*Schematization:"Strong"or stableunderlyingobject keeps overlaying
and touching object in place.
"*Attributes:Strictlytopological. No force implications.Supportingob-
jects are not agentive.
"*Relation to schooled physics: Centrallyimplicatedin "bookon the ta-
ble" misconceptionthat tables do not supportby pushing objects up.
This substitutesfor the Newtonian explanation, which may involve
springiness and must involve upward forces.
"*Comments:The weight of the supportedobject is usually seen to be
transferredinto and throughthe supporter.Hence, scales may "weigh"
objects, although,in the most primitivecases, only contactcounts;ob-
jects weigh the same even in an accelerating elevator.
Guiding
"*Schematization:A determinedpath directly causes an object to move
along it.
"*Attributes:Influenced by symmetry, other figural considerations.
"*Circumstances:Railroadcar moving along a track;ball follows a tube.
"*Relation to schooled physics: Intuitively, the motion of a ball follow-
ing a tube needs no explanation. In extreme cases, the ball may be
seen to follow in the center line of the tube, needing no contact or
forces. This mustdefer to force explanationsin physics class; the sides
of the tube must push to the inside of a turnto cause the ball to follow
along.
"*Comments:Generally of relatively low-reliabilitypriority. Defers to
blockingor impenetrabilityexplanations."Squareorbit"(p. 165) seems
to be a related figural manifestation.
Clamping
"*Schematization:An object "clamped"by opposite forces (also when
pulled simultaneouslyand equally in opposite directions) is held sta-
bly in place.
"*Circumstance:A vice.
* Relation to schooled physics: Equal and opposite forces not only do
not mandaterest, but also have nothingto do with stabilityunderper-
turbation.Clamping does not seem to be problematicin instruction,

because dynamicsseldom involves analysisof clampedsituations.Dy-
namic balance provides a productive alternativeto clamping.
Rigidity
"*Schematization:A cluster of phenomenarelating to the presumption
thatmost objectsare effectively infinitelyrigid. Typically,this involves
lack of "give"and coordinatedmotion of all parts.
"*Attributes:Solidity.
"*Relationto schooledphysics: Must defer to springiness;rigidityis less
compatiblewith Newtonianphysics than"stiffnessseen as increasing-
ly firm springiness."
"*Comments:Rigidity may have perceptualorigin in immediatevisual
perception of coordinatedmotion.
Springiness (spring scale p-prim)

"*Schematization:Objectsgive understressingforce. The amountof give
is proportionalto force.
"*Circumstance:Clay or couch pillow under pressure.
"*Relation to schooled physics: Becomes much more fundamentalthan
rigidity, but it only glosses more detailed analyses.
"*Comments:Initially,springinessis associatedwith semistaticphenome-
na and situations:little connection, for example, to oscillation, which
would be a naturalphysicist association.
Equilibrium
"*Schematization:A system with multiple influences has a naturaldo-
main of stability within some range of parametersof the influences.
"*Attributes:Stability, nonaligned influences.
"*Circumstances:An orbit may be viewed as stable confluence of cen-
trifugal, gravitational,and other forces. Equilibriumis like balanc-
ing, as in dynamic balance, where conflict may not be salient.
"*Relation to schooled physics: Must come to defer to mechanisms of
stability that are much more specific and complex than simple
equilibrium.
"*Comments:This is a powerful, centralp-primthatgeneralizesdynam-
ic balance. There are frequently figural considerations.
Generalizedspringiness
* Schematization:Disruptiveinfluenceon equilibriumcreatesa displace-
ment from equilibriumproportionalto strength of the influence.
* Circumstances:Pushing a pan balance "away from equilibrium."
222 DISESSA
"*Relationto schooledphysics: No useful work in early learningbut does

not seem to be disruptiveof basic dynamicalconcepts. Must come to
defer to specific, forceful mechanisms.
"*Comments:This is a perfectly reasonablepresumptionof linearity. It
is like springiness but without mediating deformation.
Dynamic balance
"*Schematization:A pair of forces or directedinfluences are in conflict
and happen to balance each other.
"*Attributes:Conflict, equality, steady state.
"*Circumstances:Two people push against one another.
"*Relation to schooled physics: Dynamic balance is generally com-
patible with physics instruction. It may be used to gloss "canceling
forces."
"*Comment:This phenomenonpreparesfor (cues) overcoming, should
one of the forces involved increase or decrease.
Overcoming
"*Schematization:One force or influence overpowers another.
"*Attributes:Changingrelative strength.Acceleratingeffect of success-
ful influence.
"*Circumstances:A resistingforce gives way; an animateagentincreases
effort.
"*Relation to schooled physics: Generallythis seems innocuous but not
very helpful, either.
Abstract balance
"*Schematization:Some quantitiesmust balance-an imperativeform of
dynamic balance.
"*Attributes:Frequentlythere are figural contributions.
"*Circumstances:The monkey balancing a weight problem (p. 137).
"*Relation to schooled physics: May be a useful gloss on algebraiccon-
straints of various sorts.
"*Comment:Differs from equilibriumin thatchangesin one quantityare
necessarily followed by changes in the balancedquantity.This p-prim
probablyrequiresspecific "reasons"to assumequantitiesbalance, such
as figural considerationsor convertibility ("worth").
Canceling
"*Schematization:An influencemay be undoneby an oppositeinfluence.
Generally involves sequential acts that result in no net effect.
"*Attributes:Conflict. Comparablebut opposite influences.
* Circumstances:Interpretsdynaturtlekick (to move) and antikick (to
stop).
* Relation to schooledphysics: Becomes a mathematicalscheme, inter-

preted by numerical or algebraic cancellation.
Equilibration
"*Schematization:A returnto equilibriumis the naturalresultof remov-
ing a disequilibratinginfluence. It needs no further explanation.
"*Attributes:Disruption resolved.
"*Circumstances:If a disequilibratingweightis takenout of a panbalance,
it "returnsto equilibrium."
"*Relation to schooled physics: Must defer to specific mechanismsthat
force returnto some other configurationthat is otherwise judged in-
tuitively as simply "more natural."
"*Comments:This is a powerful self-explanatoryprinciple of change
without intervention. A typical assumptionabout the patternof mo-
tion in returningto equilibriumis "gradualslowing"; less typical is
diminished bobbing or sloshing.
Recoil
"*Schematization:Released tension, as in dynamic balance, results in
generation of opposite impetus.
"*Attributes:Tension, release, violence in the recoil itself.
"*Circumstances: Rope breaks in tug of war; pullers are "thrown"
backward.
"*Comments:Relatively unsophisticatedand low-priority p-prim.
Released object falls (straightdown)
"*Relationto schooledphysics: Implicatedin some impetusdatathatcar-
ried objects do not have impetus and fall straightdown on release.
"*Comments:This is an everyday phenomenonbut with low reliability,
even in the naive view. It defers to reasoningon the basis of forceful
interventionof gravity or naturaltendency as a better explanation.
Wobbling
"*Schematization:Slow movement(especially of small objects) is prone
to irregularity.
"*Attributes:Unusual slowness, irregularity.
"*Comments:Considergenesis possibilitiessuch as a marbleslowing and
thus moving irregularlyon a kitchen floor, or try to walk very slowly
(resulting in imbalance). This may be a good example of a common
but low-priority p-prim.
Bigger means lower pitch (or slower)

* Circumstances:Bells, musical instruments.
* Relation to schooled physics: Comes to gloss a fundamentalrelation
224 DISESSA
in the simpleharmonicoscillator-using slower (frequency)ratherthan

pitch.
Stiffer means faster
"*Circumstance:Vibration of objects.
"*Comment:Comparableto bigger means lowerpitch but naively much
lower priority if it exists at all.
FiguralPrimitives
I do not providean element-by-elementanalysisof figuralgestaltsandtheircausal
implications,if such an analysisis possible. Instead,I note, as remarkedin several
of the p-prim descriptions, that the spatially evident form of a situationcan in-
fluence judgment of naturalnessof motion. In general, people behave as if con-
siderationssuch as symmetry,similarity,and continuityof forms of various sorts
have dynamical implicationsin situationswhere physicists do not. See the dis-
cussion of squareorbit in the section on figuralprimitives(p. 165 and following)
and the railroad car in space (p. 158).
Children'sP-Prims
We take an abbreviatedlook at younger children'sp-prims. Some of these come
from interviewswith children(Globerson& diSessa, 1984), and some come from
p-prim analyses of others' data.
The set of time, speed, and distancep-primsthatfollows helps explain Piaget-
ian results in the presenttheoreticalframework.These expectationsare general-
ly true, all things being equal, but children apply them without caveats. They
are used by adults at low priority or in more elaboratecombinationsto achieve
more reliable use. See the discussion in the Bell subsectionof Development. To
understandchildren'sperception,faster may be betterparaphrasedas harder, in-
volving more intensity or more effort (see Piaget, 1946/1971b, p. 175).
Being ahead implies having gone faster. (May ignore relative startingpo-
sition.)
Gettingto a goalfirst (or completingan actfirst) meanshavinggonefaster.
(May ignore relative startingposition.) Filling a cup may constitutereach-
ing a goal, independentof considerationsof how tall or wide the cup is.
Thus, goal reachingmay supersededistanceconsiderationssuch as the ver-
tical height gained in filling the cup or the final height achieved.
Passing (overtaking)means going faster.
Smaller objects naturally go faster. (Bigger things are slower.)
Lighter things go faster. (Heavier things go slower.)

Less distance covered means less time. This is stable only later than some
of the previous ones, as distance covered (distance conservation) is a
problematicconcept in the beginning.
Windingup ahead implies stopping later (independentof "observed"or-
der). Judgmentsof relativetimingmay be distortedby net accomplishment.
Goingfaster (more intensity, more effort, "goingharder")means goingfor
more time. This is obviously judged to be incorrectby adults, but it appar-
ently exists enough to be articulatedby childrenat relatively low priority.
Perhapsit may be abstractedfrom the fact that faster things generally go
farther, but, as well, one must go for a longer time to go farther.
Generalizedmomentum
"*Schematization:Things generally continue to go as they have been
going.
"*Comments:This seems a very early recognitionof momentum,but it
is not restrictedto linear or even circularcases. Some childrenarticu-
late the principle and may believe that even fairly arbitrarypatterns
can be "trained"and will be extended after release. A shook object
may continue shaking. An object run in a squaremay continue doing
so after release.
Things move in the direction they are facing

* Comments:Childrenappearto anthropomorphizeinanimateobjectsin
this subtle way. Young childrenwill explain their reactionsto the cut-
stringcircularmotionby saying the object merely went the way it was
facing, whetherit was facingdirectlyoutward(probablyfiguralreason-
ing) or, more consonantwith momentum,in the directionit was going.

Jurnal

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Jurnal

Загружено:

Авторское право:

Доступные форматы

Toward an Epistemology of Physics

Author(s): Andrea A. diSessa

The aim of this work is twofold: to understandthe intuitive sense of mechanism

Requestsfor reprintsshouldbe sent to AndreaA. diSessa,GraduateSchool of Education,University

Thelatersectionsof themonograph havethreegoals:(a)to abstractandclarify

The Physical Sense of Mechanism

1. Assess the likelihoodof variousevents based on generalizationsaboutwhat

there must be many such aspects of muscularcontrol that are inarticulate.I am

treatedas self-explanatory.Everyoneknows thata book can rest on a tablewithout

Guideto the Monograph

My programfor studyinghumans'senseof physicalmechanismcenterson iden-

1. Elements: P-primsarerathersmallknowledgestructures, typicallyinvolving

words,it describespotentialfeedbackthatcanreinforceor undothe initialacti-

flock-of- tense- panic-

FIGURE 1 A structuredpriorities network.

mediaries, in turningtense-and-nervousoff. Note that if you knew your mother

to Freudianviews that take a single memorableevent to organize a pano-

3This is really a developmentalprincipleratherthan a systematicityper se. However, the result-

circumstancesor why unpredictedphenomenaactuallyoccur. Frictionand

Data Analysis Overview

The formaldatacollection phase of this work consisted of open-endedclinical

Difficultiesin EmpiricalInvestigationof P-Prims

Micro-events. P-primsare, in general, rathersmallandparticularknowledge

Nonbehavioral. Problemstrategies,goals, andplansorganizesubstantial

specifiable in advance, for example, by abstracttask analysis, and reasoningthat

Inarticulate. P-prims are not strongly related to dictionarylexicon. Much

Unfamiliarvocabulary. Ratherthanrelatingto linguisticallysanctionedat-

A sociological argument. The diversity of empirical claims relating to

diversity of views on intuitiveconceptionscalibratesthejob before us. My judg-

p-prim descriptions and propose strategies to uncover other p-prims or better

Principleof obviousness. Thefamiliarity

Principleof impenetrability.P-primsarerelativelyprimitivein anexplana-

Principleof diversity. I believetherearemanyp-prims.So, heuristically,

Principleof coverage. Thebreadthof commonexperiencemustbe covered

planet (p. 165) initiatedconsiderationof figural primitives as a class-those that

Principleof strong vocabulary. P-primsprobablyclusterin areasof strong

Principleof unproblematicgenesis. Generally,thereshouldexist com-

Principleof the body.4 Thisis a specializedprincipleof strongvocabulary

4This heuristic is too close to a programof studyingmeaning advancedby MarkJohnson(1987)

abstractedin internallyevidentterms, especiallyearly in development.Thus, agen-

Principleof functionality. Thisprincipleemergesfromthepresumption

Principleof readyavailability. Oneshouldbe ableto understand

Principleof continuity. A fundamental

Principle of dynamic. The evolution of subjects'explorationsin an inter-

Principleof invariance. Thisis a generalprinciplethat,if onegetsa descrip-

Principleof diverse evidence. This is relatedto, but moregeneralthan,

Principleof scavenging data. The humansense of physicalmechanism

Principleof discrepancy. Evidentlywhenpeoplegive nonphysicsexpla-

Theprincipleof content over form. P-primsare content-based

The Basic Forceand MotionCluster

section on constraints,these two subsectionsconcernthe implicationsof the naive

(imposed resistance), which leads some subjectsto predicta reducedresult (cor-

Resistance. Spontaneousresistance to forces and influences is a phenome-

to a tug on a string. The argumentfor this conclusion is based, in part, on the

Force as a mover. Pushingan objectfrom restcauses it to move in the direc-

FIGURE 2 Alternativemodels of a shove's effect on a moving body.

Force as a spinner. Young childrenoften know thatpushingan object off-

Continuous push. A p-primrelatedtoforce as a moveris continuouspush,

that characteristicallyinvolve certainpatternsof amplitude.It is more than that

Clamping(holdingfixed) shows plainlythe need for complexreasoningin order

plausibility)even though such intuitionsmay not be instructedor even have any

Dynamic balance. Dynamic balancing is likely abstractedfrom situations

One frequentuse of dynamic balancing at novice stages of physics problem