Pre´cis of Doing without Concepts

Precis of Doing without Concepts

Edouard MacheryDepartment of History and Philosophy of Science 1017CL University of

Pittsburgh Pittsburgh PA 15260

Macherypittedu

wwwpittedumachery

Abstract Although cognitive scientists have learned a lot about concepts their findings have yet to be organized in a coherenttheoretical framework In addition after twenty years of controversy there is little sign that philosophers and psychologists areconverging toward an agreement about the very nature of concepts Doing without Concepts (Machery 2009) attempts to remedythis state of affairs In this article I review the main points and arguments developed at greater length in Doing without Concepts

Keywords categorization concept concept combination dual-process eliminativism exemplar induction meaning natural kindsprototype reference theory

The study of concepts is in an odd state of disarray Cog-nitive scientists working on categorization inductionand reasoning have discovered a dazzling amount ofphenomena New work on prototypes in the 1990s andearly 2000s innovative ideas on causal cognition in thefirst decade of the twenty-first century the developmentof the neo-empiricist approach that assimilates the token-ing of a concept to a multi-modal perceptual simulationand the promising growth of the neuropsychology ofconcepts have rejuvenated the field At the same time thisextraordinary amount of findings has yet to be organizedin a coherent theoretical framework The current theoriesof concepts ndash prototype theories exemplar theoriestheory theories and neo-empiricist theories ndash fail toexplain all the known phenomena and there is very littleagreement about what concepts are Doing without Con-cepts (Machery 2009) attempts to provide such a theoreti-cal framework In this article I review the main points andarguments developed at greater length in the book andI conclude that abandoning the very notion of conceptis probably required to remedy the state of disarray ofthe current psychology of concepts

1 Regimenting the use of conceptin cognitive science

Because cognitive scientists rarely spell out the notion ofconcept in detail I begin by making explicit the notionof concept that is typically used within cognitive scienceMy goal in chapters 1 and 2 is threefold To clarify thisnotion to regiment the use of the term concept and toshow that philosophers and cognitive scientists theorizeabout different things when developing theories ofconcepts

The cognitive processes that underwrite cognitive com-petences are typically assumed to access some relevantinformation or knowledge Some bodies of informationare only accessed by particular processes For instanceour implicit knowledge of the syntax of the naturallanguages we speak (eg English) is only accessed bythe processes involved in parsing and in producing

sentences When this is the case I will say that the relevantbody of information is ldquoproprietary to a particular cogni-tive processrdquo By contrast some information is ldquonon-pro-prietaryrdquo It is accessed by the cognitive processes thatunderlie several distinct cognitive competences Cognitivescientists often assume that the cognitive processes under-lying our higher cognitive competences access the samebodies of knowledge For instance the processes under-lying categorization induction and speech are hypoth-esized to access the same body of knowledge about dogswhen people classify something as a dog when theymake an induction about dogs and when they understandsentences containing the word dog This knowledge isassumed to be stored in long-term memory

These preliminary points having been made I proposeto characterize the notion of concept as follows Withincognitive science a concept of x is a body of informationabout x that is stored in long-term memory and that isused by default in the processes underlying most if not

EDOUARD MACHERY is Associate Professor of Historyand Philosophy of Science at the University of Pitts-burgh a resident fellow of the Center for Philosophyof Science (University of Pittsburgh) a member ofthe Center for the Neural Basis of Cognition (CarnegieMellon University and University of Pittsburgh) and anassociate editor of the European Journal for Philosophyof Science His research focuses on the philosophicalissues raised by psychology and cognitive neurosciencewith a special interest in concepts moral psychologythe relevance of evolutionary biology for understandingcognition modularity the nature origins and ethicalsignificance of prejudiced cognition and the methodsof psychology and neuroscience Machery has pub-lished more than 50 articles and chapters on thesetopics in venues such as Analysis The British Journalfor the Philosophy of Science Cognition Mind ampLanguage The Monist Philosophy and Phenomenolo-gical Research and Philosophy of Science He is alsoinvolved in the development of experimental philos-ophy having published several noted articles in thisfield

BEHAVIORAL AND BRAIN SCIENCES (2010) 33 195ndash244doi101017S0140525X09991531

Cambridge University Press 2010 0140-525X10 $4000 195

httpdxdoiorg101017S0140525X10000555Downloaded from httpwwwcambridgeorgcore Yale University Library on 16 Nov 2016 at 153538 subject to the Cambridge Core terms of use available at httpwwwcambridgeorgcoreterms

all higher cognitive competences when they result injudgments about x I call this characterization ldquoCrdquo

It is important to highlight some significant propertiesof concepts so understood First concepts can be aboutclasses of objects (eg DOG) events (eg GOING TO THE

DENTIST) substances (eg WATER) and individuals(eg BARACK OBAMA) Second concepts are non-proprie-tary DOG is used by the processes underlying categoriz-ation induction linguistic understanding metaphorbuilding planning and perhaps other competencesThird the elements of information that are constitutiveof a concept can vary over time and across individualsFourth it might be unclear whether a given element ofinformation about x belongs to a concept of x Finally con-cepts are used by default in the cognitive processes under-lying higher cognitive competences (I call ldquoDefaultrdquo thehypothesis that some bodies of knowledge are retrievedby default when one is categorizing reasoning drawinganalogies making inductions and so on) This entailsthat our concept(s) of say dogs is (are) only a subset ofour whole knowledge about dogs The knowledge that isconstitutive of DOG is the knowledge about dogs that isretrieved by default from long-term memory when wereason about dogs categorize things as dogs and soforth I call ldquobackground knowledgerdquo about dogs theknowledge about dogs that is not part of the concept(s)of dogs Our background knowledge about a category asubstance some kind of events and the like can becalled upon occasionally when the default body of knowl-edge is insufficient to solve a cognitive task

At this point it is useful to spell out the notion of beingused by default at greater length A default body of knowl-edge about x is the body of knowledge that is presump-tively taken to be relevant when one reasons about xwhen one categorizes things as x and so on The knowl-edge that is stored in a concept of x is preferentially avail-able when we think reason and so forth about x So tospeak it spontaneously comes to mind

The proposed characterization of the notion of conceptcaptures much of what is implicit in the use of the termconcept in cognitive science However it is also clear thatC is partly at odds with some characterizations of thenotion of concept found in the literature which are dis-cussed at length in Chapter 1 of Doing without ConceptsIn any case with the proposed characterization of thenotion of concept I do not merely aim at capturing theuse of concept in cognitive science I also want to regimentit I contend that some bodies of knowledge are retrievedby default from long-term memory when one reasonscategorizes and so forth (see section 3 for a defense ofthis claim) and I propose that the term concept shouldbe used to refer to these bodies of knowledge

2 Individuating concepts

It is certainly possible for a given individual to have severalconcepts of the same category (eg several concepts ofchairs) or of the same substance (eg several conceptsof gold) For instance one might think of chairs inseveral distinct ways each of which corresponds to adifferent concept of chairs This possibility raises thefollowing question which is examined in Chapter 3of Doing without Concepts What does it mean for twobodies of knowledge about x (eg the knowledge that

water is typically transparent and the knowledge thatwater is made of molecules of H2O) to be part of thesame concept of x (WATER) rather than of two distinctconcepts (WATER1 and WATER2)

I propose the following individuation criteria (respect-ively Connection and Coordination) When two elementsof information about x A and B fulfill either of thesecriteria they belong to two distinct concepts

1 If retrieving A (eg water is typically transparent)from long-term memory and using it in a cognitiveprocess (eg a categorization process) does not facilitatethe retrieval of B (water is made of molecules of H2O)from long-term memory and its use in some cognitiveprocess then A and B belong to two distinct concepts(WATER1 and WATER2)

2 If A and B yield conflicting judgments (eg the judg-ment that some liquid is water and the judgment that thisvery liquid is not water) and if I do not view either judgmentas defeasible in light of the other judgment (ie if I holdboth judgments to be equally authoritative) then A and Bbelong to two distinct concepts (WATER1 and WATER2)

Letrsquos clarify these two criteria a bit further Connectionunpacks the idea that the components of a given conceptmust be connected (or ldquolinkedrdquo as I say in Doing withoutConcepts) If the beliefs that water is typically transparentand that water is made of molecules of H2O are part of thesame concept WATER then using the first belief to classifysome sample as water enables me to conclude that it ismade of molecules of H2O Accessing one part of aconcept makes the other parts accessible and poised tobe used in cognitive processing (for related discussionsee Millikan 2000 Ch 10)

Coordination expresses the idea that the parts of a singleconcept should not result in conflicting judgments that areboth taken to be equally authoritative Coordination iscompatible with the parts of a given concept yielding con-flicting judgments provided that all but one are viewed asdefeasible To illustrate although the belief that grand-mothers have gray hair and the belief that grandmothersare mothers of parents would yield different judgmentsif they were used to categorize a young-looking grand-mother (eg Sarah Palin) they could still both be partof the concept GRANDMOTHER if one of the two judgments(Sarah Palin is not a grandmother) is defeated by the otherone (Sarah Palin is a grandmother) In effect the firstbelief would be treated as a heuristic that sometimesleads us astray

These two criteria merely explain what it is for a givenindividual to have one or several concepts about someclass of objects some substance and so on (within-person individuation) It does not explain what it is fortwo or more individuals to have the same concept of x(between-person individuation) Now one might expecta theoretician of concepts to provide individuation criteriafor this situation too However I doubt that these areneeded for the psychology of concepts since they seemto play no explanatory role in psychology

3 Defending the proposed notion of concept

Some cognitive scientists and philosophers of psychologyreject Default (viz the assumption that some bodies ofknowledge are retrieved by default when one is categoriz-ing reasoning drawing analogies and making inductions)

Machery Precis of Doing without Concepts

196 BEHAVIORAL AND BRAIN SCIENCES (2010) 3323


In Chapters 1 and 8 of the book I rebut the criticismsmounted against this assumption

Several findings suggest that typicality varies across con-texts Roth and Shoben (1983) have shown that dependingon the linguistic context (eg when participants arepresented with ldquoStacy volunteered to milk the animalwhenever she visited the farmrdquo or ldquoFran pleaded withher father to let her ride the animalrdquo) participants judgethat different animals are typical (cows and goats for thefirst linguistic context horses and mules for the second)Similarly Barsalou (1985) reports that judgments of typi-cality vary across contexts In Study 2 participantsrsquo typical-ity judgments about the members of two groups differedwhen these groups were conceptualized differently (asphysical education teachers and current event teacherson the one hand and as two invented types of program-mers ndash namely Q programmers and Z programmers ndashon the other hand) This study also shows that whenparticipants are familiarized with a given category in differ-ent contexts their judgments of typicality vary Barsalou(1987 1993) also reports that the typicality of objectsvaries when participants are asked to take different pointsof view on these objects For example people judge differ-ently the typicality of birds when they take the point of viewof someone from China and when they report their typical-ity evaluation from their own point of view In addition thecorrelation between typicality judgments across partici-pants is low (circa 5) and lower than expected for a givensubject on two different occasions (around 8)

Theorists have used such findings to challenge the ideathat some bodies of knowledge are retrieved by defaultwhen one categorizes draws inductions reasons and thelike Barsalou (1985) concludes that

People may not retrieve the same concept from long-termmemory every time they deal with a particular categoryInstead they may construct a diverse variety of concepts inworking memory to represent a particular category acrossdifferent situations such that the concept used to represent acategory is rarely if ever the same According to this viewlong-term memory does not contain invariant concepts (Barsalou1985 p 646)

(For similar arguments see also Barsalou 1987 Smith ampSamuelson 1997 p 170)

Theorists who reject Default have drawn two distinctconclusions As discussed in Chapter 8 of Doing withoutConcepts some theorists such as L Smith (Smith ampSamuelson 1997) contend that Default is part of the verynotion of concept and they conclude that since there areno such things as bodies of knowledge retrieved bydefault from long-term memory there are no such thingsas concepts Smith and Samuelson (1997 p 190) concludethat ldquoa successful theory of categories [ ] might requirethat we give up timeless abstractions such as conceptsrdquo

While agreeing that Default is part of the notion ofconcept typically used in cognitive science other theoristssuch as Barsalou and Prinz propose to redefine the notionof concept Concepts should be thought of as the bodies ofknowledge in working memory that are used at a giventime in a given task They are constructed on the fly todeal with the peculiarities of the task at hand and theytypically vary from time to time (Barsalou 1993 p 29Prinz 2002 Malt amp Sloman 2007)

Though Barsalou (1993) Prinz (2002) L Smith (seeSmith amp Samuelson 1997) and Malt and Sloman (2007)

take the body of evidence reviewed above to establisheither conclusion I demur on three grounds First thehypothesis that some bodies of knowledge are retrievedby default from long-term memory and used in the pro-cesses underlying the higher cognitive competences isconsistent with some variation in the bodies of knowledgethat are used at any given time This variation can have twosources When we reason about x in addition to thedefault body of knowledge about x we sometimes retrievesome specific elements from our background knowledgeabout x In addition once retrieved from memory thebody of knowledge that is retrieved by default can be tai-lored to the peculiarities of the given situation On thisview knowledge retrieval would be a two-step procedure(1) Retrieve the default body of knowledge from long-term memory (2) tailor it to the situation (Sperber ampWilson [1998] present similar ideas) Thus the mere factthat performances in experimental tasks vary from timeto time does not show that there are no bodies ofknowledge retrieved by default from long-term memoryWhat would not be consistent with Default is a verylarge variability across contexts of the knowledgebrought to bear on tasks And indeed Barsalou claimsthat there is a ldquotremendous variability in performances not only in category membership but also in typicalitydefinitions and probably most other categorizationtasksrdquo (1993 p 34 my emphasis) However as weshall now see the relevant variability in performances is infact moderate

Second the nature of the variation found by BarsalouMalt Sloman and others is either irrelevant to evaluateDefault or supportive of it Let us consider first thepieces of evidence that turn out to be irrelevant to evaluateDefault Many findings about the context-sensitivity oftypicality are misleading Roth and Shobenrsquos (1983) find-ings merely show that (unsurprisingly) people evaluatedifferently the typicality of target animals with respect tothe category of animals that get milked and the categoryof animals one uses to ride That typicality varies whenevaluated from different points of view (Barsalou 1993)does not show that peoplersquos concepts vary acrossoccasions since in effect one asks participants not touse their own concept of x to complete a task when oneasks them to make judgments about x from someoneelsersquos point of view Similarly the variability of the typical-ity judgments made by different individuals says nothingabout whether a given subject retrieves a default body ofknowledge across occasions Showing as Barsalou (1985)does that the typicality structure of a given class ofobjects can substantially vary when it is conceptualizeddifferently (viz as physical education teachers and currentevents teachers on the one hand and as Q programmersand Z programmers on the other hand) is interestingbut this finding does not show that the concept of agiven category varies across contexts and circumstancesbecause properly speaking current events teachers andZ programmers are two distinct categories although theyare composed of the same individuals Showing that thetypicality structure of a given category varies whenpeople are familiarized with this category in different situ-ations is also interesting but again it says little aboutwhether someone who is familiarized with a given categoryone way will rely on a default body of knowledge aboutthis category In addition some results that allegedly


BEHAVIORAL AND BRAIN SCIENCES (2010) 3323 197httpdxdoiorg101017S0140525X10000555Downloaded from httpwwwcambridgeorgcore Yale University Library on 16 Nov 2016 at 153538 subject to the Cambridge Core terms of use available at httpwwwcambridgeorgcoreterms

undermine Default in fact support it As noted Barsalou(1987 1993) reports that on average the test-retestreliability of typicality judgments is at least 8 It is alsohigher when participants are re-tested one hour and aday after the first test Furthermore Barsalou reportsthat the typicality of highly typical and atypical itemsdoes not change much These results are evidence thatacross occasions a default concept is retrieved fromlong-term memory

Finally a large body of evidence supports DefaultConsider linguistic understanding (Ziff 1972 discussedin Murphy amp Medin 1985) The sentence ldquoA cheetahcan outrun a manrdquo is meaningful and most peoplewould agree with it However as Murphy and Medinput it (1985 p 303) it is true only if the representedcheetah is not ldquoa 1-day old cheetah or an aged cheetahwith arthritis or a healthy cheetah with a 100-poundweight on its backrdquo But when we read ldquoA cheetah canoutrun a manrdquo these representations of cheetahs do notcome to mind This phenomenon suggests that when aspeaker utters ldquoA cheetah can outrun a manrdquo or when ahearer or a reader understands this sentence she retrievesfrom memory a default body of knowledge about cheetahsPerhaps one will object that when one reads the sentenceldquoA cheetah can outrun a manrdquo one merely constructs acontext-appropriate interpretation of ldquocheetahrdquo ratherthan retrieving a default body of knowledge about chee-tahs If this were true then people would also constructa context-appropriate representation of cheetahs if theyhad to decide whether the sentence ldquoA man can outruna cheetahrdquo is true They would for example imagine anold three-legged cheetah and the sentence ldquoA man canoutrun a cheetahrdquo would then be judged true tooHowever I predict that under time pressure peoplewould judge the sentence ldquoA man can outrun acheetahrdquo to be false This would be evidence that insuch conditions they retrieve the very default body ofknowledge they retrieve when they read ldquoA cheetah canoutrun a manrdquo Naturally with no time pressure theycould construct an interpretation of cheetah under whichthe sentence ldquoA man can outrun a cheetahrdquo is true Butthis is consistent with the existence of default bodies ofknowledge because as I have already proposed peoplecan and do retrieve some additional information (vizsome information not contained in their concepts) fromtheir background knowledge

Behavioral studies also show that some informationabout a category substance and so on is retrieved auto-matically in every context (Barsalou 1982 Whitney et al1985) Barsalou (1982) found that when people judgethat a property (eg stinks) spontaneously ldquocomes tomindrdquo when they read a given noun (eg ldquoskunkrdquo) reac-tion times in a property-verification task are similarwhen the noun is presented in a relevant linguisticcontext (ldquoThe skunk stunk up the entire neighborhoodrdquo)and when it is presented in an irrelevant linguisticcontext (ldquoThe skunk was under a large willowrdquo) By con-trast reaction times are larger in the latter condition(ldquoThe roof had been renovated prior to the rainyseasonrdquo) than in the former condition (ldquoThe roof creakedunder the weight of the repairmanrdquo) when people judgethat a property (eg can be walked upon) does not spon-taneously come to mind when they read a given noun(eg ldquoroofrdquo) Barsalou calls the first kind of property

ldquocontext-independentrdquo and the second kind ldquocontext-dependentrdquo

Cognitive neuroscience provides further evidence insupport of Default (although the relevant studies werenot developed to test this hypothesis) After havingtrained participants with novel tools Weisberg et al(2006) recorded brain activation in a perceptual task (avisual matching task) To complete this task one needsonly appeal to some structural information about theshape of the novel tools thus one would expect themedial portion of the fusiform gyrus to be activated (forreview see Martin 2007) Interestingly activation wasalso recorded in the intraparietal sulcus the premotorcortex and the medial temporal gyrus areas of the brainthat are known to store information about the typicalmovements associated with tool use It thus seems thatthe perceptual task resulted in the automatic retrieval ofinformation that was not needed to solve the task consist-ent with the idea that people have bodies of knowledgethat they retrieve by default (for similar findings see alsoJames amp Gauthier 2003 Hoenig et al 2008)

4 Developing a psychological theory of concepts

It is important to keep in mind that the notion of conceptproposed earlier (viz C) does not amount to a theory ofconcepts Rather C does two things It spells out what isimplicit in cognitive scientistsrsquo use of the term conceptand it proposes to regiment this use So what does apsychological theory of concepts consist in

As I explain in Chapter 1 of Doing without Conceptspsychological theories of concepts typically attempt toidentify the properties that are typical of concepts (ldquothegeneral properties of conceptsrdquo) Five kinds of propertiesare of interest to cognitive scientists First cognitive scien-tists are interested in the nature of the information that isconstitutive of concepts For instance cognitive scientistswant to know whether concepts consist of some statisticalinformation about the properties that are characteristic ofa class or of a substance as prototype theorists have pro-posed (eg Hampton 1979 1981 2006 2007 Smith2002) or whether they consist of causal generalizations(eg Gopnik amp Meltzoff 1997 Griffiths et al 2007Murphy amp Medin 1985 Rehder 2003a Tenenbaumet al 2007) Second cognitive scientists are interested inthe nature of the processes that use concepts Forinstance some psychologists have argued that these pro-cesses are based on similarity (eg Hampton 1993)while others disagree (eg Rips 1989) Third cognitivescientists develop hypotheses about the nature of thevehicles of concepts Thus neo-empiricists such as Barsa-lou and Prinz contend that the vehicle of concepts issimilar to the vehicle of perceptual representations (Barsa-lou 1999 2008b 2009 Machery 2006c Prinz 2002 2005)Fourth for about a decade cognitive scientists haveattempted to identify the brain areas that are involved inpossessing concepts (for reviews see eg Mahon amp Car-amazza 2009 Martin 2007 Pulvermuller 2005) Finallycognitive scientists have developed hypotheses about theprocesses of concept acquisition (eg Ashby amp Maddox2004 Gopnik 2003)

In addition to developing hypotheses about the generalproperties of concepts cognitive scientists have shown




some interest in distinguishing different types of conceptsand in identifying the properties of these types of con-cepts Medin et al (2000) have rightly insisted on theimportance of this task and on its relative neglect bycognitive scientists

Why do cognitive scientists want a theory of conceptsTheories of concepts are meant to explain the propertiesof our cognitive competences People categorize the waythey do they draw the inductions they do and so onbecause of the properties of the concepts they haveThus providing a good theory of concepts could go along way toward explaining some important higher cogni-tive competences

5 Concept in cognitive science and in philosophy

The term concept is used in philosophy particularly in thephilosophy of mind as well as in cognitive scienceChapter 2 of Doing without Concepts examines therelation between these two uses It is common among phi-losophers to assume that concept is used in the same sensein philosophy and in cognitive science and that psycholo-gistsrsquo theories of concepts aim at answering the issuesphilosophers are interested in (Edwards 2009 Fodor1998 2008 Laurence amp Margolis 1999 Margolis 19941995 Margolis amp Laurence 2006 Rey 1983 1985 2009b)In addition it is common to hold that as answers to theissues of interest in philosophy psychological theories ofconcepts are defective Thus Fodor (2003) concludes hisreview of Gregory Murphyrsquos book The Big Book ofConcepts as follows

It is part of our not knowing how the mind works that we donrsquotknow what concepts are or what it is to have one Just abouteverything that current cognitive science says about eithertopic is wrong Gregory Murphyrsquos book tells you most ofwhat there is to the psychology of concepts Read it thereforeby all means but donrsquot even consider believing it (Fodor 2003p 4)

It is also not uncommon to see some philosophical the-ories criticized for being unable to explain how we categor-ize make inductions and so on (Prinz 2002 for discussionsee Edwards 2009)

Philosophersrsquo take on psychological theories is mistakenPhilosophical and psychological theories of concepts arenot meant to answer the same questions and are thusnot competing Typically by concept philosophers referto that which allows people to have propositionalattitudes (beliefs desires etc) about the objects of theirattitudes The concept of a triangle is therefore thatwhich allows people to have propositional attitudes(beliefs desires etc) about triangles A theory of conceptsin philosophy attempts to determine the conditions underwhich people can have propositional attitudes about theobjects of their attitudes (Fodor 1998 2008 Peacocke1992 2008) but not to explain the properties of ourhigher cognitive competences By contrast psychologistsattempt to explain the properties of our categorizationsinductions and so forth but they do not attempt to deter-mine the conditions under which people are able to havepropositional attitudes about the objects of theirattitudes Furthermore psychologists do not need tohold and typically do not hold that we are able to havepropositional attitudes about the objects of our attitudes

by virtue of having specific bodies of knowledge aboutthem For instance prototype theorists do not need tohold and typically do not hold that having a prototypeis a condition for being able to have attitudes about theobjects of our attitudes In fact prototype theorists aresilent on this question

The upshot of this argument should be clear Althoughboth philosophers and cognitive scientists use the termconcept they are not talking about the same things Cog-nitive scientists are talking about a certain kind of bodiesof knowledge whereas philosophers are talking aboutthat which allows people to have propositional attitudesMany controversies between philosophers and psycholo-gists about the nature of concepts are thus vacuous

6 The heterogeneity hypothesisversus the received view

Cognitive scientists of concepts naturally acknowledgedifferences between concepts The concept of dogs isclearly different from the concept of cats More interest-ing they also acknowledge differences between kinds ofconcepts For instance there has been much work inexperimental and developmental psychology on the differ-ences between the concepts of animals and the concepts ofartifacts (eg Bloom 1996 Gelman 1988 2003 Gelman ampMarkman 1986 Malt amp Sloman 2007) But above andbeyond these differences cognitive scientists oftenassume that concepts share many properties that are scien-tifically interesting In Chapter 3 of Doing withoutConcepts I call this assumption ldquothe received viewrdquo It iswell expressed by Gregory Murphy

The psychology of concepts cannot by itself provide a fullexplanation of the concepts of all the different domains thatpsychologists are interested in The details of each ofthese must be discovered by the specific disciplines thatstudy them Nonetheless the general processes of conceptlearning and representation may well be found in each ofthese domains (Murphy 2002 pp 2ndash3)

The received view has been instrumental in the debatesthat have marked the history of the psychology of conceptssince the 1970s Cognitive scientists who are committed todifferent theories of concepts (say a particular prototypetheory and a particular exemplar theory) have attemptedto discover properties of our higher cognitive competences(eg the exemplar effect reported in Medin amp Schaffer1978) that were easily explained by the theory theyendorsed (eg the exemplar theory) but were not easilyexplicable by the competing theory (prototype theoriesdo not naturally explain the exemplar effect for discussionsee Smith amp Minda 2000) This research strategy makessense only if one supposes that a single theory of conceptsshould be able to account for all the relevant phenomenaIf contrary to the received view the class of conceptsdivides into several kinds that have little in common thedistinct theories of concepts that characterize thesekinds of concepts will account for different phenomenaand the fact that theory A but not theory B explainssome phenomenon such as the exemplar effect will notnecessarily constitute evidence against theory B

As I explain in Chapter 3 the received view stands insharp contrast with a view about concepts developed inmy own work (see also Machery 2005) the Heterogeneity



Hypothesis According to this hypothesis the class ofconcepts divides into several distinct kinds that havelittle in common ndash ldquothe fundamental kinds of conceptsrdquoBecause the class of concepts divides into distinct funda-mental kinds it is a mistake to assume that there aremany general properties of concepts and that a theoryof concepts should attempt to describe these Althoughthe heterogeneity hypothesis can be developed in severalways (Machery 2005 2006a Piccinini amp Scott 2006) Icontend that a given category (eg dogs) a given sub-stance (eg water) or a given kind of events (eg goingto the dentist) is typically represented by several distinctconcepts (eg DOG1 and DOG2) These coreferentialconcepts belong to the fundamental kinds of conceptsEach coreferential concept can be used to categorizedraw inductions understand the relevant words makeanalogies and so forth (Fig 1)

In addition the heterogeneity hypothesis contends thatthese concepts are often used in distinct processes That iswe have several categorization processes several inductionprocesses and the like each of which uses a distinct fun-damental kind of concepts (Fig 2)

If the heterogeneity hypothesis is correct the class ofconcepts is not a natural kind Natural kinds are classeswhose members share many scientifically important prop-erties in virtue of one or several causal mechanisms (Boyd1991 1999 Griffiths 1997 Machery 2005) Water anddogs are natural kinds in this sense for examplesamples of water have many properties in common invirtue of consisting of the same molecules of H2O In agiven science the scientific classificatory scheme is devel-oped to identify the natural kinds in the relevant domainbecause identifying these kinds allows scientists todiscover new generalizations Scientific classificatoryschemes are modified when they do not identify therelevant natural kinds (as happened during the chemicalrevolution in the eighteenth century) and scientificnotions are often eliminated when it is found that theyfail to pick out natural kinds (for discussion see sect

13) Because the hypothesized fundamental kinds of con-cepts have little in common the class of concepts cannotbe a natural kind if the heterogeneity hypothesis is correct

7 What kind of evidence could support theheterogeneity hypothesis

Chapter 5 of Doing without Concepts describes threekinds of evidence that can provide support for the hetero-geneity hypothesis I consider them in turn in this section

Suppose that the class of concepts divides into severalfundamental kinds and suppose also that coreferentialconcepts are often used in distinct cognitive processes(ie distinct categorization processes distinct inductionprocesses) What properties would we then expect toobserve in experimental tasks First if experimentalconditions can be designed that trigger only one of thehypothesized categorization processes or only one of thehypothesized induction processes we should expectsome experimental findings to be best explained if the con-cepts used in the relevant experimental tasks are identicalto a first fundamental kind of concepts other experimentalfindings to be best explained if the concepts used in therelevant experimental tasks are identical to a second fun-damental kind of concepts and so on For instance ifone hypothesizes that the fundamental kinds of conceptsare exemplars and prototypes then one might find categ-orization tasks where participantsrsquo categorization perform-ances are best explained if the concepts used in these tasksare prototypes and other categorization tasks where par-ticipantsrsquo categorization performances are best explainedif the concepts used in these tasks are exemplars

Second suppose that in some conditions several of thehypothesized categorization (or induction) processes aretriggered at the same time Then in some circumstancesthese processes will produce congruent outputs (eg cat-egorization judgments) while they will produce incongru-ent outputs in other circumstances When the latter

Figure 1 The heterogeneity hypothesis




happens participants will have to decide between conflict-ing judgments Participants should thus be expected tobe slower when the hypothesized processes are expectedto yield conflicting outputs than when they are notTest-retest reliability should also be expected to be lowerin the experimental conditions where it is hypothesizedthat the hypothesized categorization (induction) processeswill result in incongruent outputs than when it is hypoth-esized that they will result in congruent outputs Notice-ably this kind of evidence (particularly slower reactiontimes) has extensively been used in cognitive science toargue that a given task involves two independent cognitiveprocesses (eg Greene et al 2001)

Finally experimental and neuropsychological dis-sociations can be used to determine whether a given taskinvolves several processes The epistemology of dis-sociations is intricate (Ashby amp Ell 2002 Caramazza1986 Glymour 1994 Dunn amp Kirsner 1988 2003 Plaut1995 Shallice 1988 Van Orden et al 2001) but I maintainthat dissociations provide evidence about the number andnature of the processes underlying a given competence

8 The fundamental kinds of concepts

Now that the nature of the evidence required to supportthe heterogeneity hypothesis has been clarified it is timeto lay my cards on the table What are the fundamentalkinds of concepts And what is the evidence for their exist-ence In what follows I will briefly describe the kind ofevidence supporting the heterogeneity hypothesis butdue to limitations of space this will not amount to acomprehensive articulation of the evidence adduced inChapters 6 and 7 of Doing without Concepts

In Chapter 4 I propose that the class of concepts dividesinto at least three fundamental kinds of concepts ndashprototypes exemplars and theories These three theoreti-cal constructs are well known in the psychology ofconcepts as they correspond to the entities posited bythe main theories of concepts that have been developedsince the 1970s (for a review see Murphy 2002) Althoughthere are several distinct theories about what prototypes

exemplars and theories are these theories agree aboutthe distinctive features of each type of concept In sub-stance prototypes are bodies of statistical knowledgeabout a category a substance a type of event and so onFor example a prototype of dogs could store some statisti-cal knowledge about the properties that are typical of dogsandor the properties that are diagnostic of the class ofdogs According to prototype theories when I categorizedraw an induction make an analogy and so forth I spon-taneously bring to mind the properties that are typicaldiagnostic (etc) of the relevant category substance andso forth Prototypes are typically assumed to be used incognitive processes that compute the similarity betweena prototype and other representations such as the rep-resentations of the objects to be categorized in a linearmanner (Hampton 1979 1993 2006 2007 Rosch ampMervis 1975 Smith 2002) Exemplars are bodies of knowl-edge about individual members of a category (eg FidoRover) particular samples of a substance and particularinstances of a kind of event (eg my last visit to thedentist) For instance according to exemplar theories aconcept of dogs would consist of a set of bodies of knowl-edge about specific dogs (Rover Fido) When I categorizedraw an induction make an analogy and so on I spon-taneously bring to mind the properties of specificmembers of the relevant categories of specific samplesof the relevant substances et cetera Exemplars are typi-cally assumed to be used in cognitive processes thatcompute the similarity between a set of exemplars andother representations such as the representations of theobjects to be categorized in a nonlinear manner (Medinamp Schaffer 1978 Nosofsky 1986 1992 Nosofsky ampStanton 2005) Theories are bodies of causal functionalgeneric and nomological knowledge about categoriessubstances types of events and the like A theory ofdogs would consist of some such knowledge about dogsWhen I categorize draw an induction make an analogyand so on I spontaneously bring to mind this causal func-tional generic and nomological knowledge Recent workon causal knowledge suggests that theories might beused in cognitive processes that are similar to the algor-ithms involved in causal reasoning (Gopnik et al 2004)

Figure 2 Several processes underlying a given cognitive competence



Thus the heterogeneity hypothesis proposes that formany categories substances kinds of events we typicallyhave a prototype a set of exemplars and a theory aboutthem Thus we might have a prototype of dogs a set ofexemplars of particular dogs and a theory about dogsFurthermore prototypes exemplars and theories areoften used in distinct processes The heterogeneityhypothesis proposes that we have a prototype-based categ-orization process an exemplar-based categorizationprocess and a theory-based categorization process Notethat the hypothesis is not merely that our knowledgeabout dogs includes some knowledge about their typicalor diagnostic properties some knowledge about someparticular dogs and some causal functional and genericknowledge (as Rey [2009b] erroneously believes) Thiswould be a fairly uncontroversial claim Rather theclaim is that for most categories substances et ceterawe have several bodies of knowledge that are retrievedby default and that are often used in distinct cognitive pro-cesses (eg several distinct categorization processes)

The heterogeneity hypothesis also contends that thefundamental kinds of concepts have little in commonThis is indeed the case if these fundamental kinds reallyconsist of prototypes exemplars and theories Theyconsist of different types of knowledge they are used indifferent kinds of processes and they are probablyacquired by distinct processes Given what cognitive scien-tists working on concepts are interested in (see sect 4)they count as very different kinds of entities

One might perhaps object that prototypes exemplarsand theories do have some properties in common In par-ticular they are all bodies of knowledge they are all storedin long-term memory and they are all used in the pro-cesses underlying higher cognition This however doesnot undermine the heterogeneity hypothesis for theclaim that prototypes exemplars and theories have littlein common really states that the fundamental kinds of con-cepts have in common few properties that are scientificallyinteresting and discovered empirically Prototypes exem-plars and theories have in common numerous propertiesthat are not of interest to cognitive scientists (eg theyare all mental states) In addition far from beingdiscovered empirically the mentioned commonalitiesbetween prototypes exemplars and theories (eg theyare all bodies of knowledge they are all stored in long-term memory etc) are in fact used to identify whatconcepts are

So what is the evidence for the claim that our long-termmemory stores prototypes exemplars and theories Whenone examines 30 years of research on categorization andinduction as I do in Chapters 6 and 7 one finds out thatin both areas of research some phenomena are wellexplained if the concepts elicited by some experimentaltasks are prototypes some phenomena are well explainedif the concepts elicited by other experimental tasks areexemplars and yet other phenomena are well explainedif the concepts elicited by yet other experimental tasksare theories As already noted if one assumes that exper-imental conditions prime the reliance on one type ofconcepts (eg prototypes) instead of other types (egexemplars and theories) this provides evidence for theheterogeneity hypothesis

Letrsquos illustrate this situation with the work on categori-cal induction ndash the capacity to conclude that the

members of a category possess a property from the factthat the members of another category possess it and toevaluate the probability of this generalization (for reviewsee Feeney amp Heit 2007 Heit 2000 Murphy 2002 Ch8 Sloman amp Lagnado 2005) A large number of phenom-ena suggest that prototypes or exemplars are sometimesinvolved in induction (Osherson et al 1990 Sloman1993) Similarity-based models of induction whichassume that processes underlying induction are definedover either prototypes or exemplars explain best twowell-known findings about induction ndash the similarityeffect and the typicality effect Other phenomena arebest explained if the concepts involved in the relevantexperimental conditions are causal theories Investigatingthe judgments made by tree experts (landscapers taxono-mists and parks maintenance personnel) about thestrength of inductive conclusions about trees Proffittet al (2000) found that rather than relying on typicality(as predicted for instance by Osherson and colleaguesrsquosimilarity-coverage model) the pattern of answers andthe justifications provided suggest that experts often basetheir judgments on theories about hypothetical causalmechanisms (see also Lopez et al 1997) As explained insection 7 the fact that different properties of our inductivecompetence are best explained by theories positing differ-ent theoretical entities (viz prototypes exemplars ortheories) constitutes evidence for the existence of distinctkinds of concepts used in distinct processes Strikinglythis conclusion is consistent with the emerging consensusamong psychologists working on induction that peoplerely on several distinct induction processes (Murphy 2002Proffitt et al 2000 Rehder 2006 Sloman amp Lagnado2005)

A natural question raised by these findings concerns theconditions that prime the reliance on prototypes ratherthan exemplars and theories or on theories rather thanprototypes and exemplars (and so on) in induction (seesect 12) Because cognitive scientists have rarely fullyembraced the idea that there are several distinct kinds ofconcepts and several processes defined over them thereis little systematic work on this question (but see Rehder2006)

The research on categorization and concept learningreviewed in Chapter 6 tells an even clearer story provid-ing evidence for the existence of prototypes exemplarsand theories that are used in distinct categorization pro-cesses The research on concept combination reviewedin Chapter 7 also shows that when people produce acomplex concept they appeal to exemplars prototypesand theories However in contrast to the research oninduction and categorization it appears that a singleprocess uses prototypes exemplars and theories (insteadof several distinct combination processes each of whichuses a distinct kind of concepts)

9 Neo-empiricism

A number of cognitive scientists have recently developed anew approach to the nature of concepts (Barsalou 19992008a 2009 Barsalou et al 2003 Gallese amp Lakoff2005 Glenberg 1997 Martin 2007 Martin amp Chao2001 Prinz 2002 2005) which I have called ldquoneo-empiri-cismrdquo (Machery 2006c 2007) Although there are




differences between neo-empiricist theories they allendorse the two following theses

1 The knowledge that is stored in a concept is encodedin several perceptual and motor representational formats

2 Conceptual processing involves essentially reenact-ing some perceptual and motor states and manipulatingthese states

Thesis 1 is about the format of concepts Neo-empiri-cists claim that conceptual knowledge is encoded in per-ceptual and motor representational formats By contrastamodal theorists contend that our conceptual knowledgeis encoded in a representational format that is distinctfrom the perceptual and motor representational formats(Barsalou et al 2003 p 85) This distinct representationalformat is usually thought of as being language-likealthough importantly amodal representations need notform a language (see further on) Thesis 2 concerns thenature of the cognitive processes underlying categoriz-ation induction deduction analogy-making planning orlinguistic comprehension The central insight is thatretrieving a concept from long-term memory duringreasoning or categorization consists in tokening some per-ceptual representations a process called simulation orreenactment Cognitive processing consists in manipulat-ing these reenacted percepts (eg Barsalou 1999p 578) Following Barsalou (1999) I will use the term per-ceptual symbols to refer to concepts understood in accord-ance with Theses 1 and 2

Perceptual symbols might be one of the fundamentalkinds of concepts but I argue in Chapter 4 of Doingwithout Concepts that the evidence provided so far fallsshort of establishing this I have identified three mainshortcomings of the research in support of neo-empiricism(see also Machery 2007 for other arguments see Dove2009 Machery 2006c Mahon amp Caramazza 2008)

First what I have called ldquoAndersonrsquos problemrdquo in refer-ence to Andersonrsquos (1978) work on the controversybetween imagistic and propositional theories of thinkingNeo-empiricists typically contrast the predictions madeby amodal theories of concepts and the predictionsmade by neo-empiricist theories of concepts (egPecher et al 2004 Solomon amp Barsalou 2001 2004Yaxley amp Zwaan 2007) and they then attempt to showthat the neo-empiricist predictions but not the amodalpredictions are verified The problem is that there is nosuch thing as the amodal prediction of concepts ratherdifferent amodal theories of concepts make differentpredictions depending on what they assume about theprocesses that use amodal concepts In numerous casessome amodal theories of concepts make exactly the samepredictions as the neo-empiricist theories of conceptsdeveloped by cognitive scientists such as Barsalou (forsome examples see Machery 2007 2009 Mahon amp Cara-mazza 2008) As a result neo-empiricist findings do notdistinguish between neo-empiricism and amodal theoriesof concepts in general Rather they provide evidenceagainst specific amodal theories of concepts while beingnaturally accommodated by other amodal theories ofconcepts

The second shortcoming is what I have called ldquotheproblem from imageryrdquo Most proponents of amodal the-ories of concepts (eg Fodor 1975 Simon 1995) acknowl-edge that in some situations people rely on imagery (egvisual imagery) For instance we visualize our own home

when we are asked how many windows it has What propo-nents of amodal theories of concepts deny is that imageryis the only type of processes people have People also haveamodal concepts that are used in non-perceptual pro-cesses The fact that proponents of amodal theories ofconcepts recognize the role and importance of imageryentails that when amodal theorists expect people to relyon imagery to solve a particular task showing thatpeople use imagery in this task fails to provide evidencefor neo-empiricism and against amodal theories of concepts(for some examples see Machery 2007)

The third shortcoming is what I have called ldquothe gener-ality problemrdquo Neo-empiricists typically assume that allconcepts are perceptual symbols However it could bethat perceptual symbols constitute only a kind of con-cepts ndash a hypothesis that would naturally be consistentwith the heterogeneity hypothesis In fact researchsuggests that at least some conceptual representations ndashnamely the representations of the magnitudes of classesof objects or sequences of sounds ndash are not perceptualbut amodal (Dove 2009 Machery 2007) Although theserepresentations do not form a language and thus are differ-ent from the hypothesized representations of Fodorrsquos(1975 2008) language of thought they are not perceptualeither (Dove 2009 Machery 2006c) Dove (2009) hasdeveloped the generality problem in great detailshowing that the research in support of neo-empiricismhas typically focused on a single kind of concepts ndashnamely ldquoconcrete or highly imageable conceptsrdquo (2009p 431) ndash and that neo-empiricist findings are unlikely tobe found with concepts with low imageability such asabstract concepts

Others have identified further difficulties Reviewing arange of neuroscientific work on concepts and variousimportant behavioral studies Mahon and Caramazza(2008) grant that the perceptual and motor systems areoften activated during conceptual processing but theyinsist that this activation falls short of supporting neo-empiricism for it can be interpreted in two differentways First the interpretation preferred by neo-empiri-cists The brain areas involved in perceptual and motorprocessing or the areas near those are activated becauseconcepts are perceptual and motor representations andperceptual representations are realized in these areasSecond the amodal interpretation of these findings Theactivation of these brain areas results from the activationof other brain areas not involved in perceptual processingand from this activation spreading from the latter areas tothe former (a well-known phenomenon) Both interpret-ations account equally well for the neo-empiricist findings

Finally letrsquos say a few words about the neo-empiricistresearch in cognitive neuroscience A large number offMRI studies show that tasks meant to tap into the pro-cesses underlying higher cognition (particularly tasksinvolving the understanding of words) activate either thevery brain areas involved in perceptual and motor proces-sing or brain areas near those (see eg Barsalou 2008aKiefer et al 2007 Martin 2007 Martin amp Chao 2001Pulvermuller 2005 Simmons et al 2007 Thompson-Schill 2003) However in contrast to neo-empiricistsrsquousual interpretation of these findings I contend thatmuch of the neuroscientific research on concepts chal-lenges this approach Because neo-empiricists insist thattokening a concept means tokening some perceptual



representations they are committed to the view thatconcept retrieval should activate our perceptual areas(Simmons et al 2007) However a typical finding inneuroscience is that the brain areas activated are nearand thus not identical to the brain areas involved in per-ceptual or motor processing (a point acknowledged bySimmons et al 2007) Furthermore in much neo-empiri-cist research on concepts in neuroscience the brain areasthat are activated in the tasks meant to tap into the pro-cesses underlying higher cognition are anterior to thebrains areas activated in perceptual processing (eg Bou-lenger et al 2009 Chao amp Martin 1999 Hauk et al 2004Kable et al 2005 Martin et al 1995 Pulvermuller amp Hauk2006) A plausible interpretation is that the brain areasactivated in the tasks tapping into higher cognition areamodal representations which are distinct from theperceptual representations activated in the tasks tappinginto perceptual processes but near them To concludeit might be that perceptual symbols are a fundamentalkind of concepts but research still fails to establish itbeyond doubt

10 Hybrid theories of concepts

Several hybrid theories of concepts have been developedsince the 1970s and there is a fair amount of differencesbetween them but they all agree on several crucialpoints (Anderson amp Betz 2001 Keil 1989 Keil et al1998) Hybrid theories of concepts grant the existence ofseveral types of bodies of knowledge but deny that theseform distinct concepts rather these bodies of knowledgeare the parts of concepts Like the heterogeneity hypoth-esis hybrid theories of concepts typically concur thatthese parts store different types of information Forinstance some hybrid theories (Gelman 2004) have pro-posed that one part of a concept of x might store some stat-istical information about the xrsquos while another part storessome information about specific members of the class ofxrsquos and a third part some causal nomological or func-tional information about the xrsquos Furthermore they oftencontend that the distinct parts that compose a givenconcept are used in different processes (eg Oshersonamp Smith 1981) For example the parts that compose agiven hybrid concept might be used in distinct categoriz-ation processes distinct induction processes and so on

Although hybrid theories of concepts and the heterogen-eity hypothesis agree on several points they are far frombeing identical In section 2 I proposed two individuationcriteria that specify when two bodies of knowledge about xform two distinct concepts rather than a single conceptHybrid theories of concepts contend both that the differentcoreferential bodies of knowledge are connected andthat they are coordinated The heterogeneity hypothesisassumes that at least one of these two claims is false

Evidence tentatively suggests that prototypes set ofexemplars and theories are not coordinated Maltrsquos(1994) work on how people conceptualize water showsthat people have at least two distinct concepts of water ndasha theoretical concept of water that identifies water withany substance composed of molecules of H2O and a proto-type that identifies water with any substance that sharessome typical properties (origins use appearances) Morerecently Machery and Seppala (forthcoming) have

shown that many participants (between 20 and 80depending on the pair of sentences considered) arewilling to endorse apparently contradictory sentences ofthe following form

1 In a sense tomatoes are a fruit2 In a sense tomatoes are not a fruit3 In a sense whales are fish4 In a sense whales are not fishThat is many participants say that both (1) and (2) and

both (3) and (4) are true Although there are several poss-ible explanations of these findings a plausible explanationis that people retrieve different concepts of tomato whenthey read (1) and (2) When they retrieve a theory of toma-toes they answer that (1) is true whereas they answer that(2) is true when they retrieve a prototype of tomatoes Thissuggests that their prototype and their theory form two dis-tinct concepts rather than two parts of the same concept

11 Multi-process theories

The heterogeneity hypothesis proposes that prototypesexemplars and theories are often used in distinct cognitiveprocesses (eg distinct categorization processes) I callldquomulti-process theoriesrdquo those theories that contend thata given cognitive competence (eg categorization induc-tion or the capacity to make moral judgments) is under-written by several distinct processes Chapter 5 of Doingwithout Concepts is dedicated to examining this kind ofcognitive theory Dual-process theories which have beenembraced in social psychology are a type of multi-process theory characterized by a distinction betweentwo types of processes (slow analytic intentional pro-cesses and fast automatic processes for discussion seeeg Evans 2007 Evans and Frankish 2009 Gigerenzeramp Regier 1996 Sloman 1996) Gigerenzer and Toddrsquosfast-and-frugal-heuristics research program is anotherkind of multi-process theory (Gigerenzer et al 1999)

The default hypothesis in cognitive science is that a cog-nitive competence is underwritten by a single cognitiveprocess and the burden of proof typically is on thosewho hold a multi-process theory for some cognitive com-petence In light of the recent work on a range of cognitivecompetences this state of affairs should be revised Evi-dence suggests that cognitive competences are commonlyunderwritten by several distinct processes

Postulating that a given cognitive competence is under-written by distinct processes raises a host of questions thathave rarely been explicitly confronted by proponents ofmulti-process theories (but should be) The two mostimportant issues are the following onesA In what conditions are the cognitive processes under-

lying a given cognitive competence triggered Arethey all always triggered Are they rather triggered indistinct circumstances Or perhaps in overlapping cir-cumstances If they are not all always triggered whatcues or processes determine their triggering Is theirtriggering under intentional control

B If the cognitive processes that underlie a given cognitivecompetence are triggered in the same conditions howdoes the mind choose between their outputs or inte-grate themIt is fair to say that current multi-process theories such

as the dual-process theories have typically failed to give




clear answers to these questions This limits their capacityto genuinely predict experimental outcomes

What about the prototype-based exemplar-based andtheory-based cognitive processes In what conditions arethey triggered And if they are triggered simultaneouslyhow does the mind choose between their outputs Thereis no systematic work on these issues in fact I hope thatthis book will invite cognitive scientists to systematicallyinvestigate them

What is known can be presented briefly It appears thatthe categorization processes can be triggered simul-taneously (eg Allen amp Brooks 1991 Smith et al 1998)but that some circumstances prime reliance on one ofthe categorization processes Reasoning out loud seemsto prime people to rely on a theory-based process of categ-orization (Smith amp Sloman 1994) Categorizing objectsinto a class with which one has little acquaintance seemsto prime people to rely on exemplars (Smith amp Minda1998) The same is true of these classes whose membersappear to share few properties in common (Minda ampSmith 2001 Murphy 2002 Smith amp Minda 2000) Verylittle is known about the induction processes except forthe fact that expertise seems to prime people to rely ontheoretical knowledge about the classes involved (Lopezet al 1997 Proffitt et al 2000)

12 Open questions

One of the virtues of the heterogeneity hypothesis is tobring to the fore a range of questions that have not beensystematically examined by cognitive scientists I nowsummarize some of these issues

First psychologists should investigate the factors thatdetermine whether an element of knowledge about x ispart of the concept of x rather than being part of the back-ground knowledge about x Frequency of use is the onlyfactor that has been systematically investigated (Barsalou1982) Other factors should be considered ndash includingattention and explicit teaching

Second there are several prototype theories severalexemplar theories and several theory theories Althoughevidence indicates that we have prototypes exemplarsand theories it remains unclear however which proto-type theory exemplar theory or theory theories is correctThat is the exact nature of prototypes exemplars and the-ories remains to be investigated Cognitive scientists havetypically attempted to show that concepts are prototypesor that concepts are exemplars or that concepts are the-ories but they have paid little attention to investigatingthe nature of prototypes exemplars and theories ingreat detail Similarly it is important to determine whichprototype-based model of categorization (inductionetc) which exemplar-based model of categorization(induction etc) and which theory-based model of categ-orization (induction etc) is correct (instead of comparingsay a specific exemplar-based and a specific prototype-based model of categorization as has usually beendone) Recently some psychologists have taken up theimportant task of comparing the models of categorizationand of induction developed by prototype theories(Sloman amp Lagnado 2005) as well as the models of categ-orization and of induction developed by theory theorists(Rehder amp Kim 2006) Such efforts should be systemati-cally pursued

Third multi-process theories are also another importantresearch area that requires systematic attention I havesketched a framework for developing multi-processtheories of the higher cognitive competences identifyingseveral key questions that need to be answered by propo-nents of these theories Multi-process theories need todeal with some important issues that have not been fullysolved Of particular importance is the kind of evidencethat can support multi-process theories Among thethree kinds of evidence I have distinguished the legiti-macy of dissociations remains controversial and shouldbe investigated further It is also plausible that otherkinds of evidence can support multi-process theoriesWhile contemporary psychologists often endorse dual the-ories of cognition that distinguish System 1 and System 2processes there are numerous other types of multi-process theory Furthermore existing multi-processtheories such as dual-process theories do not specify inwhich conditions the hypothesized processes are triggeredand how their outputs are integrated and as a result theyare unable to yield clear predictions instead of mere posthoc accommodations

Psychologists should also develop detailed multi-processtheories of those cognitive competences that are thebest candidates for being realized by several distinctprocesses ndash namely categorization and induction (seesect 11) So far we know very little about how the distinctcognitive processes that realize competences such as categ-orization and induction are organized We do not reallyknow whether outside the lab the categorization (or induc-tion) processes are triggered simultaneously or in distinctconditions We do not really know what determines theirtriggering And we do not know what happens to theoutputs of the categorization (or induction) processeswhen these processes are simultaneously triggered

13 Concept eliminativism

Let us take stock We have seen that the class of conceptsdivides into several distinct kinds of concepts namelyprototypes exemplars and theories which have little incommon Categories substances events are often rep-resented by several coreferential concepts (a prototype aset of exemplars and a theory) These are not parts ofconcepts but are rather bona fide concepts Prototypesexemplars and theories are also typically used in distinctcognitive processes for example in distinct categorizationprocesses although little is known about the organizationof these processes It is rarely the case that a given cogni-tive competence is underwritten by a single cognitiveprocess rather the mind usually includes several distinctprocesses that do the same thing Finally the heterogen-eity hypothesis focuses attention on a range of empiricalquestions for which systematic empirical informationis missing partly because cognitive scientists have notconsidered the heterogeneity hypothesis seriously

To conclude this article I want to discuss a radical pro-posal inspired by the views about concepts presented inDoing without Concepts and in this precis article Cogni-tive scientists might be better off renouncing the verynotion of concept Rather they should use theoreticalterms introduced to refer to the fundamental kinds ofconcepts ndash namely prototype exemplar and theory The



heterogeneity hypothesis contends that the class ofconcepts is not a natural kind It does not support manycausally grounded generalizations because the class ofconcepts divides into several fundamental kinds thathave little in common Furthermore theoretical termsare often rejected when it is found that they fail to pickout natural kinds To illustrate some philosophers(Murphy amp Stich [1999] building on Griffiths [1997])have proposed to eliminate the term emotion from thetheoretical vocabulary of psychology on precisely thesegrounds The proposal here is that concept should beeliminated from the vocabulary of cognitive science forthe same reason

Chapter 8 of Doing without Concepts examines theintricate and controversial logic of eliminativist arguments(see also Mallon et al 2009 Murphy amp Stich 1999 Stich1996) Many eliminativist arguments attempt to concludethat there are no xrsquos (eg no beliefs see Churchland[1981] and Stich [1983] or no races for discussion seeeg Mallon [2006]) from the fact that the definition of xis not satisfied For instance if nothing satisfies the defi-nition of belief or concept it is concluded that there areno beliefs of concepts However such eliminativist argu-ments are bound to be unsuccessful because they areenmeshed with controversial issues concerning howwords such as belief or concept refer (Mallon et al2009) What we need is another kind of eliminativist argu-ment which clarifies when it is legitimate to eliminate ascientific term from a scientific classificatory scheme

In a nutshell I propose that scientific terms should beeliminated on pragmatic grounds (this is what I havecalled ldquoscientific eliminativismrdquo) To determine whetherx has a legitimate place in the vocabulary of a givenscience or whether it should be eliminated one shouldexamine whether using x helps to fulfill the goals of thisscience ndash particularly whether it helps its classificatorypurposes Picking out natural kinds is the primary functionof theoretical terms in many sciences (Quine 1969 butperhaps not in all sciences Russell 1948) Thus when itis found out that a scientific term fails to pick out anatural kind there is a presumption that it should be elimi-nated from the relevant science However one still needsto consider and weigh the costs and benefits of eliminatingthis term Perhaps keeping this term might simplifycommunication between scientists On the other handkeeping this term might prevent the development of amore appropriate classificatory scheme (a common situ-ation I suspect) If the relevant term does not pick out anatural kind and if the benefits of keeping it do notclearly overweigh the costs then one should eliminate it

Because concept does not pick out a natural kind if theheterogeneity hypothesis is correct there is a presumptionthat it should be eliminated from the theoretical vocabu-lary of psychology Furthermore the continued use ofconcept in cognitive science might invite cognitive scien-tists to look for commonalities shared by all concepts orto develop another theory that would encompass all thephenomena known about the processes underlyinghigher cognition If the heterogeneity hypothesis iscorrect these efforts would be wasted By contrast repla-cing concept with prototype exemplar and theory wouldbring to the fore the urgent open questions discussed insection 12 For instance speaking of a prototype-basedcategorization process an exemplar-based categorization

process and a theory-based categorization processmakes it clear that there are several categorization pro-cesses and brings to the fore the questions of the organiz-ation of these categorization processes

Now one might worry that eliminating the wordconcept would make communication among cognitivescientists cumbersome To some extent this is likely tobe true as is suggested by the frequent use of this termin this article But I doubt that the elimination ofconcept would make communication too cumbersomeafter all when required cognitive scientists can alwaysappeal to the description ldquobodies of knowledge used inhigher cognitionrdquo It seems likely that using such a descrip-tion will not invite cognitive scientists (or at least not to thesame extent) to look for commonalities among all bodies ofknowledge used in higher cognition If this is correct thenthe costs resulting from the elimination of concept arelimited and cognitive scientists should eliminate thenotion of concept from their theoretical apparatus

ACKNOWLEDGMENTSI would like to thank Guy Dove Kevan Edwards and Paul Bloomfor helpful comments on previous drafts

Open Peer Commentary

Default knowledge time pressure and thetheory-theory of concepts

doi101017S0140525X10000324

Thomas BlanchardDepartment of Philosophy Rutgers University New Brunswick NJ

08901-1107

tblanchardphilosophyrutgersedu

Abstract I raise two issues for Macheryrsquos discussion and interpretationof the theory-theory First I raise an objection against Macheryrsquos claimthat theory-theorists take theories to be default bodies of knowledgeSecond I argue that theory-theoristsrsquo experimental results do notsupport Macheryrsquos contention that default bodies of knowledge includetheories used in their own proprietary kind of categorization process

Edouard Machery (in Doing without Concepts Machery 2009p 12) claims that psychologists (including theory-theorists) takeconcepts to be ldquobodies of knowledge that are used by default inthe processes underlying the higher cognitive competencesrdquo (p11) According to Macheryrsquos gloss default knowledge about forexample dogs is the knowledge that is ldquopreferentially availablerdquo(p 11) and ldquospontaneously comes to mindrdquo (p 12) in mostcontexts in which we make judgments about dogs Peoplersquosknowledge about a category is not exhausted by their defaultknowledge but their non-default (background) knowledge isless easily retrievable and is used only when default knowledgeis insufficient for the task at hand Machery suggests thatwhether or not a subjectrsquos judgment about x provides evidenceabout the content of her default body of knowledge relative tox depends (at least partly) on whether or not she made the judg-ment under time pressure Thus he notes in his Precis (targetarticle sect 3 para 8) the fact that under no time pressurepeople may retrieve a representation of a cheetah that makesthe sentence ldquoA man can outrun a cheetahrdquo true is no evidencethat this representation belongs to their default knowledge

CommentaryMachery Precis of Doing without Concepts



rather than to their background knowledge about cheetahs Insituations with no time pressure people can and do retrieveinformation from their background knowledge

This turns out to be crucial when we consider the nature ofthe theory-theoristsrsquo experiments There is a stark contrastbetween the kinds of experimental tasks relied upon by say pro-totype-theorists on the one hand and theory-theorists on theother hand Prototype theorists have usually relied upon exper-imental tasks in which subjects were instructed for example tolist properties associated with a category in a short time periodor make categorization judgments under explicit time pressure(see eg Hampton 1979 Rosch amp Mervis 1975) By contrastmany of the experimental tasks designed by theory-theorists donot involve any element of time pressure Consider for instanceRipsrsquos (1989) famous pizza experiment As Smith and Sloman(1994) have rightly noted ldquothere was no mention of speed inRipsrsquo instructionsrdquo (p 380) Or consider Keilrsquos (1989) transform-ation experiments in which subjects were asked to make judg-ments about the biological membership of an animal that hasundergone unusual transformations Nothing in Keilrsquos text indi-cates that subjects had to make their categorization judgmentsunder any particular time pressure

This raises two issues for Macheryrsquos discussion of the theory-theory First because Ripsrsquos and Keilrsquos experimental tasks didnot involve any element of time pressure they do not show thatthe knowledge retrieved by subjects to solve these tasks belongsto their ldquodefault knowledgerdquo in Macheryrsquos sense Subjects mighthave retrieved information from their background knowledge tomake their categorization judgments (Given the unusual natureof the tasks they were asked to solve this would not be surprising)However theory-theorists take Ripsrsquos and Keilrsquos experiments toprovide crucial insights into the nature and structure of conceptsIf theory-theorists assume concepts to be default bodies of knowl-edge then they have misinterpreted Ripsrsquos and Keilrsquos results Amore charitable hypothesis is that pace Machery theory-theoristsdo not consider that being used by default in higher-level cognitivecompetences is a necessary condition for a piece of information tocount as conceptual knowledge This hypothesis is supported bysome of the theory-theoristsrsquo own claims For example bothMurphy and Medin (1985 p 296) and Gelman (2003 p 244)take theories to be concepts but allow that in many contextssubjects rely by default on prototypical features to make routineand quick categorization judgments and retrieve theoretical infor-mation only when prototypical information is insufficient for thetask at hand

Here is the second issue Machery (2009 pp 52 119) con-tends that for most categories our default knowledge includesa theory of that category and that theory-like default bodies ofknowledge are associated with a distinct kind of categorizationprocess He mentions Ripsrsquos and Keilrsquos results in support ofthis claim (pp 183ndash87) However these experiments do notsupport Macheryrsquos contention since they are consistent withthe claim that theories belong to background rather than todefault knowledge Moreover the literature generated byRipsrsquos results partly disconfirms Macheryrsquos claim Thus Smithand Sloman (1994 pp 379ndash80) failed to replicate Ripsrsquosresults when they instructed subjects to make their categorizationjudgments quickly This suggests that the theoretical informationon which subjects relied in Ripsrsquos original experiment did notbelong to their default knowledge

Now some experimental results mentioned by Machery dosupport the idea that default bodies of knowledge include theor-etical information Thus Luhmann et al (2006 cited in Machery2009 p 186) have shown that subjects can use causal knowledgeto make categorization judgments even under time pressure Inthe learning phase of the experiment subjects learned aboutthe causal relations between various features of fictionalanimals In a subsequent categorization phase subjects reliedupon this causal knowledge to make categorization judgmentseven when they were instructed to respond as quickly as possible

This is evidence that their default knowledge about those fic-tional animals included causal (hence theoretical) knowledgeHowever as Luhmann et al argue their results are consistentwith (and according to them even support) the idea that theirsubjectsrsquo theory-driven categorization process was not ldquoqualitat-ively different from traditional similarity-based processes basedon weighted feature matchingrdquo (2006 p 1104) Subjects maysimply have assigned strong weights to causal features whenthey first learned the category and later used those featuresweights in a simple similarity computation when they madecategorization judgments Thus these results do not supportMacheryrsquos claim (2009 p 119) that theory-like default bodiesof knowledge are used in a categorization process similar toscientific reasoning and different in kind from the categorizationprocesses associated with for example prototypes

ACKNOWLEDGMENTI thank Alvin Goldman for helpful advice and comments

Carving nature at its joints using a knifecalled concepts

doi101017S0140525X10000336

Justin J Couchman Joseph BoomerMariana V C Coutinho and J David SmithDepartment of Psychology University at Buffalo State University

of New York Buffalo NY 14260

jjc38buffaloedu jboomerbuffaloedu

mvc5buffaloedu psysmithbuffaloedu

Abstract That humans can categorize in different ways does not implythat there are qualitatively distinct underlying natural kinds or thatthe field of concepts splinters Rather it implies that the unitary goalof forming concepts is important enough that it receives redundantexpression in cognition Categorization science focuses on commonalitiesinvolved in concept learning Eliminating ldquoconceptrdquo makes this moredifficult

Categorization researchers hypothesize that minds groupobjects to reflect natural groupings in the world For exampletheorists often point to the family-resemblance relationshipsembodied by biological taxonomies Poodles resemble eachother more than they do Labradors dogs resemble each othermore than they do bears and so forth These similarities pre-sumably hold in the world and in our mental lives Of coursethere is a fundamental separation between categories in theworld and concepts in the mind Fortunately though categor-ization science finds its grounding and coherence at both theecological (world) and psychological (mind) levels of analysisUnfortunately this often leads to the misconception thatldquocategoryrdquo and ldquoconceptrdquo are equivalent or interchangeableterms ndash as Machery correctly points out in Doing withoutConcepts (Machery 2009 pp 8ndash14)

The world contains coherent assemblages of objects that aresometimes called natural kinds family-resemblance categoriesor consequential stimulus regions All organisms share the pro-foundly important goal of learning the concepts that summarizethese assemblages These concepts are behavioral and psycho-logical equivalence classes ndash they let creatures behave similarlytoward similar things and understand similar things similarlyConcepts are the mindrsquos tool for representing and respondingto natural kinds

The conceptual goal of living organisms is unitary ndash to developpsychological representational schemes that preserve theseassemblages in acting and understanding The fitness conse-quence of not doing so is unitary The assemblages of things inthe world ndash the natural kinds ndash may have a unitary structuretoo One sees that categorization science is motivated by this



unitariness That concepts are a staff of life for living organisms isa profound truth that would be lost by breaking the fieldinto process splinters This is a principal reason for preservingthe coherence of the field

Now it is true that minds lack direct access to the worldrsquos thingsin themselves Therefore minds must create mental conceptsthat estimate natural-kind categories Given this necessarydistinction between natural-kinds in the world and our mentalestimates of them it is not clear why concepts prototypesexemplars theories or any other mental representation wouldbe considered a natural kind Indeed if there are any groupsthat are not natural kinds mental representations are likelyamong them

Machery (2009 p 241) contends that because it does not pickout a unified natural kind ldquoconceptrdquo is a useless or perhaps evenharmful term We disagree It does not shatter the field ofconcepts if it turns out ndash as it does ndash that concept estimationproduces a marvelous variety of concepts and learningprocesses ndash including prototypes exemplar memories categoryrules decision bounds and theories This only means that thephenomenon of concepts is rich and diverse It also means thatthe unitary goal of estimating concepts is central enough to sur-vival that it deserves and receives redundant expression withincognitive systems To us this centrality emphasizes the needfor a coherent field of concepts and categorization

Of course the field sometimes seems less than coherentbecause researchers tailor category tasks to elicit differentprocesses Prototype theorists use large stimulus sets thatreduce stimulus repetition and defeat exemplar memorizationThese categories also have family-resemblance organization thatencourages prototype formation Exemplar theorists often usecategories that share little or no family resemblance Perhaps themost iconic of these are ldquo5ndash4rdquo categories As Machery (2009pp 175ndash77 see also Smith amp Minda 2000) correctly notesthese categories share few features preventing prototype for-mation and the stimulus sets are small Participants see manyrepetitions of each stimulus and naturally memorize themRule-based tasks are clearly simply unidimensional in characterencouraging hypothesis testing reasoning and solution by ruleformation

However the progress in this area shows the method in themadness Researchers have found that the dominant categoriz-ation process is strongly affected by the size of categories theirperceptual coherence the dimensionality of the taskrsquos sufficientsolution the stage of category learning the default tendenciesof the categorizing organism and so forth (Ashby amp Maddox2005 Blair amp Homa 2003 Couchman et al in press Homaet al 1981 Murphy 2002 Smith amp Minda 1998 2000) More-over it is clear that these generalizations are not just isolatedfindings It is insufficient to claim that the effect of stimuli on cat-egorization is captured completely by a description of prototypesexemplars or theories (as Machery [2009 sect 832] requires)Rather organisms tune sensitively to the affordances of their cat-egory experience in order to choose the most cognitively econ-omical and adaptive learning solution they can Viewed from abroader perspective these different strategies are instances oforganisms navigating upon a multi-dimensional fitness surfaceBut the goal the navigation and the surface all embody aunitary fitness potential that must not be overlooked and thatreally is not divisible

This broader perspective also raises many important theoreti-cal questions What species have which categorization potenti-alities and which constraints What were the antecedents ofhumansrsquo categorization system in the vertebrates or primatesWhat are the affordances of language and symbolic reasoningand how do they change the nature of concept formationWhat are the developmental stages by which humans acquiretheir mature concepts and adaptive action patterns in theworld What are the implications of cognitive developmentcognitive aging and neurological impairments for category

learning both regarding constraints and regarding preservedor spared capacities on which remediation could capitalizeto maximize education and training Replacing ldquoconceptrdquo withldquoprototyperdquo ldquoexemplarrdquo or ldquotheoryrdquo as Machery (2009 p 242)suggests would unwarrantedly deemphasize these importantquestions In our view answering all these questions is alsobest served by understanding the unitary nature of the task athand

That task is to carve nature at its joints using the psychologicalknife called concepts It is true it is profoundly important toknow and it is all right for the progress of science that theknife is Swiss-Army issue with multiple blades

Not different kinds just special cases

doi101017S0140525X1000052X

David DanksCarnegie Mellon University and Institute for Human amp Machine Cognition

Department of Philosophy Pittsburgh PA 15213

ddankscmuedu

httpwwwhsscmueduphilosophyfaculty-danksphp

Abstract Macheryrsquos Heterogeneity Hypothesis depends on his argumentthat no theory of concepts can account for all the extant reliablecategorization data I argue that a single theoretical framework basedon graphical models can explain all of the behavioral data to which thisargument refers These different theories of concepts thus (arguably)correspond to different special cases not different kinds

One of Macheryrsquos central arguments for his HeterogeneityHypothesis in Doing without Concepts (Machery 2009) is thatfor each different theory of concepts there are reliable data-sets ndash behavioral neuroscientific and dissociation ndash that arebest (or only) explained by that theory That is prototype-based exemplar-based and theory-based kinds of concepts areall required in order to explain all of the data Thus Macheryconcludes we have heterogeneity for many of the cognitiveitems thought to be single concepts people have cognitivelydistinct prototype-like exemplar-like and theory-like kinds ofconcepts

This argument requires that these three theorieskinds ofconcepts be qualitatively different not special cases of a moregeneral framework If there were a unifying account that encom-passed all three theories of concepts then that theory would (byhypothesis) be able to explain all the same data as the heterogen-eity hypothesis that Machery proposes in his book Hybridtheories of concepts are the most prominent instances of sucha unifying account and Machery provides numerous argumentsagainst them precisely because they have the potential to under-mine his data-based argument for the heterogeneity hypothesisHybrid theories are not however the only way to unify thethree theories of concepts

The behavioral data used in Macheryrsquos argument can all Icontend be explained by a single theory of concepts based ongraphical models At a (very) high level a graphical modelencodes a set of relationships whether informationalprobabilis-tic causal communication taxonomic or other More formally agraphical model has two components (1) a graph composed ofnodesvertices and edges (directed andor undirected) thatencode the qualitative relationships and (2) a representation ofthe quantitative relationships The precise quantitative com-ponent depends on model-specific features joint probabilitydistributions and structurallinear equations are typical Bayesiannetworks (causal or probabilistic) structural equation models andMarkov random fieldsnetworks are probably the most commongraphical models though there are many other types (Lauritzen[1996] provides a comprehensive formal overview of graphicalmodels)




One instance of the theory-theory (causal model theory) isalready explicitly formulated in the language of graphicalmodels (specifically causal Bayesian networks) A deeper con-nection is suggested by the fact that all three theories ofconcepts ultimately understand concepts as structured relationsamong features components causes and effects and so onthough there are between-theory differences in the nature ofthe objects and relations

The suggested deep connection exists All three theories ofconcepts can provably be represented in the graphical modelsframework (Danks 2004 2007) More precisely for each particu-lar prototype-based exemplar-based or theory-based1 theory ofconcepts there is a corresponding class of graphical models suchthat (A) there is a one-to-one mapping between (i) a concept inthe ldquotraditionalrdquo psychological theory and (ii) a particular graphi-cal model in that class and (B) inferencereasoning using thatgraphical model is behaviorally indistinguishable from usingthe corresponding prototype-based exemplar-based or theory-based concept As a concrete example multiplicative prototypeconcepts with second-order features are isomorphic to Markovrandom fields with restricted clique potential functions (Danks[2004 2007] proved this result for categorization the resultshave since been extended to all of the activities that Macherydiscusses) These three ldquokindsrdquo of concepts are (at least formally)each representable as a different set of restrictions within theshared graphical models framework specifically the frameworkof so-called chain graphs

These formal results open the door for homogeneity toreemerge as a live possibility Of course this homogeneityoccurs at the level of graphical models rather than exemplarsversus prototypes versus theories An individual can have whatappear to be both prototype-based concepts and exemplar-based concepts simply by having different specific graphicalmodels The force of the argument for heterogeneity is thusblunted There is a single theory that can account for all of the(behavioral) data Importantly this proposed theory is not ahybrid theory It does not hold that a single concept is composedof different graphical models corresponding to the differenttypes of concepts Rather the proposal is that any particularconcept corresponds to only one graphical model and diversityin our graphical models leads to apparent diversity in types ofconcepts

This account focuses on the formalcomputational structure ofthe different theories of concepts and one might object that itignores other more ldquometaphysicalrdquo claims made by their propo-nents This concern goes to the hard question of how to define orcharacterize a theory At one extreme we could define a theory(of concepts) by its behavioral predictions and ignore all otherassertions made about the theory At the other extreme wecould say that a theory is given by a maximal set of consistentclaims made by proponents of that theory regardless of thenature of those claims The graphical models account takes theintermediate view that a theory should be interpreted relativelyminimally as the set of claims necessary to explain the phenom-ena that purportedly fall within its reach (similar in spirit to theapproach in Strevens 2000) I contend that (though I do nothave room to argue for) the proper ldquominimalrdquo understanding oftheories of concepts is in terms of the formalcomputationalstructure that they attribute to particular concepts Althoughthere is a clear rhetorical difference between saying a conceptis based on a ldquoprototyperdquo versus an ldquoexemplarrdquo I contend thatthe difference matters only when it leads to a difference in theconceptrsquos formalcomputational structure

Machery provides an admirable survey of the vast behavioralpsychological literature on concepts and his conclusion ndash thereare at least three distinct kinds of concepts ndash is reasonable ifthere is no unifying framework for those ldquokindsrdquo But all of thebehavioral data that Machery discusses can in fact be explainedas bodies of knowledge encoded as graphical models The differ-ences in behavior can be explained as different graphical

structures ndash that is different special cases ndash rather than trulydistinct kinds

ACKNOWLEDGMENTThe author is partially supported by a James S McDonnell FoundationScholar Award

NOTE1 Statements about ldquotherdquo theory-theory of concepts apply

only to the few computationally precise instances of that idea(eg causal model theory)

An additional heterogeneity hypothesis

doi101017S0140525X10000348

Guy DoveDepartment of Philosophy University of Louisville Louisville KY 40297

guydovelouisvilleedu

Abstract In this commentary I make three points concerning Macheryrsquosresponse to neo-empiricism First his methodological critique fails toremove the threat that neo-empiricism poses to his conceptualeliminativism Second evidence suggests that there are multiplesemantic codes some of which are not perceptually based Third thisrepresentational heterogeneity thwarts neo-empiricism but also raisesquestions with respect to how we should ldquodo without conceptsrdquo

Machery (2009) provides a substantial and convincing argumentthat prototypes exemplars and theories form distinct mentalkinds The real question is whether or not this heterogeneityshould cause us to eliminate concepts as a mental kind Neo-empiricism (eg Barsalou 1999 Glenberg 1997 Prinz 2002)poses a direct threat to this proposal because it provides aunified account of concepts As Machery notes neo-empiricisminvolves two core claims that conceptual knowledge is encodedin sensorimotor representations and that conceptual processinginvolves some kind of perceptual simulation If these claimsapply to prototypes exemplars and theories then Macheryrsquoscase for eliminativism falls apart This is not simply a hypotheticalargument at least one prominent neo-empiricist (Prinz 2002)endorses both the independence of prototypes exemplars andtheories and the theoretical unity of concepts

In Doing without Concepts Machery claims that neo-empiri-cism faces two main problems The first is that certain amodalmodels (ie ones containing non-perceptual representations)are compatible with the empirical findings cited in support ofneo-empiricism The second is that many amodal theoristsacknowledge that perceptual imagery is important to some cogni-tive processes Machery (2009 p 116) concludes that ldquothere is nostrong evidence that concepts (or some concepts) are in factsimilar to perceptual representationsrdquo This argument is incon-clusive though because it fails to provide strong evidenceagainst perceptually based conceptual representations orindeed for amodal ones

In his Precis Machery identifies an additional problem forneo-empiricism which he refers to as the generality problemThis problem arises because there are robust bodies of evidencesuggesting that some specialized conceptual representations areamodal Although the generality problem clearly underminesstrong forms of neo-empiricism it is compatible with weakerforms (Machery 2007 Weiskopf 2007)

I suggest that neo-empiricism faces a more serious problemThis problem is not merely a lack of generality but rather thepresence of heterogeneity Evidence from a number of sourcessuggests that conceptual knowledge is encoded in multiple rep-resentational formats some of which are not perceptual (Dove2009) In other words it supports what might be called therepresentational heterogeneity hypothesis (RHH)



The case for the RHH can be seen most clearly in recent neu-ropsychological and neuroimaging research on imageability Tra-ditionally cognitive scientists examined imageability in terms ofprocessing advantages for high-imageable concepts over low-imageable ones in several cognitive tasks (Paivio 1987 Watten-maker amp Shoben 1987) Two major theories emerged in responseto the behavioral evidence the dual-code theory (Paivio 1987)and the context-availability theory (Schwanenflugel amp Shoben1983) Recent evidence from cognitive neuroscience supportsboth to some degree (Dove 2009)

Because my concern is the RHH I will focus on the evidencefor distinct representational formats First consider neuropsy-chological case studies Several research teams describeaphasic patients with significant left hemisphere damage whoexhibit a selective semantic impairment for high-imageablewords (Berndt et al 2002 Bird et al 2003 Crepaldi et al2006) Patients with a selective semantic impairment for low-imageable words have also been found (Marshall et al 1996)

Second a number of event-related potential (ERP) exper-iments support a neuroanatomical distinction between conceptsof high and low imageability For example Holcomb et al(1999) created a task that involved manipulations of bothcontext and concreteness ERP recordings were time-locked tosentence final words in a word-by-word reading task in whichparticipants made semantic congruency judgments (eg Armedrobbery implies that the thief used a weapon vs Armedrobbery implies that the thief used a rose) They found that sen-tence-final concrete words generated a larger and more anteriorN400 than sentence-final abstract words in both contextsFurther studies have found context-independent topographiceffects associated with imageability in single-word presentations(Kellenbach et al 2002 Swaab et al 2002) Thus ERP studiesemploying diverse tasks support the notion that different cogni-tive systems are associated with the semantic processing ofhigh- and low-imageable words

Third the idea that neural activity is modulated by imageabil-ity is generally supported by neuroimaging data A number ofstudies find that high-imageable words elicit greater activationthan low-imageable words in superior regions of the left temporallobe (Binder et al 2005 Giesbrecht et al 2004 Mellet et al 1998Noppeney amp Price 2004 Sabsevitz et al 2005) and inferiorregions of the left prefrontal cortex (Binder et al 2005 Gies-brecht et al 2004 Goldberg et al 2006 Noppeney amp Price2004 Sabsevitz et al 2005) Giesbrecht et al (2004) forexample manipulated both imageability and semantic priming(a measure of the influence of context) in an event-relatedfMRI study Participants were presented with prime word fol-lowed by a target word The words were either semanticallyrelated (bread and butter) or unrelated (wheat and slipper)Half of the pairs consisted of two high-imageable words andhalf of the pairs consisted of two low-imageable words Inkeeping with the general finding that context effects are distinctfrom imageability effects each of these manipulations modulatedactivity in anatomically distinct areas of the left hemisphere

The RHH seems to be grist for Macheryrsquos mill It certainlyundermines the threat posed by neo-empiricism to unifyconcepts In keeping with this Piccinini and Scott (2006) haveargued that the divide between cognitive processes that requirelanguage and those that do not provides a better case for concep-tual eliminativism than the heterogeneity identified by Machery

However the RHH also raises an interesting problemBecause there is every reason to suppose that prototypes exem-plars and theories have high- and low-imageable variants thesekinds are likely to be representationally heterogeneous Follow-ing Macheryrsquos reasoning this should threaten their status astheoretical kinds It seems reasonable to resist this inferencebut then the challenge is to provide an argument for resistingeliminativism in the context of prototypes exemplars and the-ories that does not apply to the larger context of conceptsWhether this can be accomplished or not remains to be seen

Unity amidst heterogeneity in theoriesof concepts

doi101017S0140525X10000543

Kevan EdwardsPhilosophy Department Syracuse University Syracuse NY 13210

kedwar02syredu

httpphilosophysyreduFacEdwardshtm

Abstract This commentary raises two concerns with Macheryrsquosapproach in Doing without Concepts The first concern is that it maybe possible to preserve a unified theory of concepts by distinguishingfacts about concept individuation from facts about cognitive structuresand processes The second concern questions the sharpness of thedistinction Machery draws between psychological and philosophicalconceptions of concepts

I think Machery is entirely right in Doing without Concepts(2009) to draw attention to the phenomenon that he refers toas the heterogeneity of concepts (as the term ldquoconceptrdquo is usedin cognitive psychology) I also agree that this stands in the wayof any unified account of concepts as exemplars prototypes orconstituents of theories Nevertheless I take issue with Mach-eryrsquos eliminativist conclusion The logical heart of my worry isthat heterogeneity across exemplars prototypes theories andso on precludes a unified theory of concepts only insofar asone assumes that such a theory needs to be built out of thekinds of structures and processes described by advocates ofthese approaches I think this assumption is worth questioningand I want to suggest an alternative approach that rejects it

The alternative begins with the idea that concepts are individ-uated by considerations more abstract than their role in cognitivestructures and processes The specific version of this idea that Ifavor individuates concept types in terms of the individualskinds or properties they represent or refer to (see Edwards2009 2010) Before saying more about this approach I want tomention another issue that might partially explain whyMachery is relatively blind to ndash or perhaps it is more charitableto say disinterested in ndash this kind of alternative

When it comes down to how to construe the starting point for atheory of concepts Machery (2009 pp 32ndash51) is quick to sidewith psychologists as opposed to philosophers Macherychooses sides in part to undermine various philosophicalconcerns on the grounds that philosophers are engaged in afundamentally different research project I think Machery maybe overlooking some potentially relevant issues here Just to beclear I agree that philosophers typically begin from a differentstarting point than psychologists and tend to emphasize differentconsiderations However this is always the case with topicslocated at the intersection of academic disciplines Admittedlythis realization does not show that philosophers and psychologistsare converging on the same theoretical entity in this particularcase Nevertheless I think one ought to be very cautious in con-cluding that different apparent starting points amount to a funda-mental difference in subject matter Given space constraints Icanrsquot properly address Macheryrsquos claims on this subject but Irsquollbriefly mention several issues regarding which philosophers maybe able to contribute to how psychologists think about concepts

One topic to highlight ndash not surprisingly ndash is the notion ofrepresentation It seems to me this is a place where philosophersand psychologists tend to take recognizably different approachesPsychologists even those who clearly are working within theframework of a Representational Theory of Mind often appearto presume something in the vicinity of representationalcontent without making this explicit Consider for examplehow natural it is to talk about exemplars or prototypes being ofor about a category In contrast philosophers in particular phi-losophers of mind have been something close to obsessed witheither explaining such talk in naturalistically respectable termsor explaining how to do without it Examples of the former




include Dretske (1981) Fodor (1987 1992) Millikan (19841993) examples of the latter include Field (2001) Horwich(1998a 1998b) I think Macheryrsquos neglect of these discussionspartially explains why a representation-based view of conceptsisnrsquot on his radar I should note that I am here echoing concernsraised by Rey (2009b)

Another topic much discussed in recent philosophy can beused to shed light on the potential relationship between rep-resentation and more blatantly psychological considerations Ihave in mind the debate between so-called reductionists andnon-reductionists especially insofar as this has involved ques-tions about multiple-realization (see in particular the debatebetween Fodor (1974 1997) and Kim (1992 1998) Speakingvery roughly the upshot of this debate has been that both anti-reductionists and their opponents have confronted the fact thatreality exists (or is properly describableconceptualized) atdifferent levels of abstraction At a minimum this forces one tobe careful about the level at which a particular theoreticalentity is construed The cost of picking too low a level of analysisis that one wrongly identifies the target entity with a disjunctionof lower-level entities each of which is better described as arealization (or instance) of the target To repeat a now hackneyedexample it is a mistake to identify doorstops with the arrange-ments of physical stuff (slices of wood hinged pieces of metalbricks old printers etc) out of which particular doorstops aremade The now standard functionalist line is that something isa doorstop by virtue of playing a characteristic role in a largersystem for example a system involving doors people and soon Someone who holds a representational view of conceptscan make a similar suggestion Type-individuating concepts interms of their representational properties frees up the possibilityfor different instances or realizations of a concept type to show upin substantially different cognitive structures and processes

The alternative approach to which I have gestured brings manyquestions in its wake I have not for example supplied argu-ments for the conclusion that a representationalist view of con-cepts should supplant rather than supplement accounts thatappeal to mental structures and processes Similarly I have notsaid anything about whether embracing a representationalistalternative forces a substantive (rather than merely terminologi-cal) shift in work done by advocates of exemplar prototype andtheory-based approaches Moreover there are familiar reasons toworry about a representation-based approach The goal of thepresent commentary is to make two relatively modest points(1) There is an admittedly radical approach to concepts thatneeds to be undermined before one opts for eliminativism (2)Various issues that have featured prominently in recent philos-ophy have the potential to impinge upon discussions of conceptsin psychology and to do so in potentially important ways

Two uneliminated uses for ldquoconceptsrdquoHybrids and guides for inquiry

doi101017S0140525X1000035X

Chad Gonnerman and Jonathan M WeinbergDepartment of Philosophy Indiana University Bloomington Indiana 47405

cgonnermindianaedu jmweinbeindianaedu

httpwwwindianaedueel

Abstract Macheryrsquos case against hybrids rests on a principle that is toostrong even by his own lights And there are likely importantgeneralizations to be made about hybrids if they do exist Moreovereven if there were no important generalizations about conceptsthemselves the term picks out an important class of entities and shouldbe retained to help guide inquiry

We concur with Macheryrsquos broad assessment in Doing withoutConcepts (Machery 2009) that the science of concepts has

revealed matters to be rather a mess But we strongly disagreewith what Machery takes as the philosophical upshot of that mes-siness as expressed in his bookrsquos title that psychologists shouldscrub ldquoconceptrdquo from their lexicon First the hybrid option isstill live and if concepts are hybrids they may well be naturalkinds Second even if ldquoconceptrdquo fails to pick out a natural kindit may yet be scientifically useful even on Macheryrsquos own terms

Machery correctly observes that hybrid theorists owe anaccount of when two bodies of knowledge are parts of thesame concept He proposes that psychologists are committed tothe principle that they are genuine parts only if they are coordi-nated (Machery 2009 p 64) that is if they generate consistentevaluations of cases But he adduces some evidence that coordi-nation often fails across parts of putative hybrids for exampleone representation says tomatoes are vegetables a bit later theother says they are not (p 72)

But this coordination condition is too strong as it would break upbodies of knowledge that psychologists treat as conceptual wholesnamely sets of exemplars Contextual shifts such as those thatadjust the perspective of the subject (eg from artist to biologist)can within selective-retrieval models shift which exemplar isretrieved even when both exemplars are stored as being aboutthe same thing ndash and thereby can shift for particular caseswhether they will be counted as in the target category (see Braisby2005) Sets of similarly stored exemplars would thus fail Macheryrsquoscoordination condition ndash forcing their dissolution likely down toindividual exemplars But thatrsquos not how psychologists treat suchexemplars His argument against hybrids proves too much

That the coordination condition is too strong does not removethe need for hybrid theorists to provide an account of conceptualwholes Here is one proposal building on machinery used muchearlier in Macheryrsquos book when he attempts to distinguishconceptual from background knowledge (pp 11ndash12) Therehe suggests that a body of knowledge about x is in the conceptof x just in case it is ldquopreferentially available when we thinkreason and so on about xrdquo (p 11) There is nothing unique toa hybrid that keeps its fans from saying such things about its sub-conceptual parts A hybrid theorist could therefore propose thattwo coreferential representations are parts of the same concept ofx just in case each is preferentially available in such a way whenwe think reason and so on about x (We note that Macheryrsquosaccount of conceptualization may require some technical tweakfor Frege cases we expect that whatever would do that for hisaccount would also work for this proposal) But a failure ofcoordination is irrelevant to this weaker condition

Whether this proposal succeeds is an open question and that issufficient for our purposes here in contending that hybrids arestill a live option For if ldquoconceptsrdquo picks out hybrids psychologyshould probably keep the term as there will likely be informativegeneralizations to be made As Machery argues empiricalresearch suggests that for most categories we have a prototypea body of exemplars and a theory but not a definition So ahybrid theorist would be able to theorize that most conceptspartially consist of a prototype and so on but not a definitionBut we cannot read such generalizations off of Macheryrsquos refer-ence-fixing description (p 12) they are discoveries So if con-cepts are hybrids ldquoconceptrdquo probably picks out a class thatsupports many informative useful generalizations

Even if the above is incorrect about hybrids ldquoconceptrdquo wouldstill have important work to do As Machery suggests scientistsshould keep a term if it plays a useful role (p 239) ldquoConceptrdquodoes that by efficiently marking out a class that scientists wantto make claims or ask questions about even if it should turnthat there are vanishingly few generalizations to be made aboutsomething merely in virtue of its being a concept There aremany terms in good scientific standing across various disciplinesthat play such roles including we would suggest ldquoalgorithmrdquoldquosub-atomic particlerdquo or ldquonutrientrdquo (and even some specificnutrients such as ldquovitamin Brdquo see Elder 1994 p 259) Noneof these categories seems to support many generalizations about



their members qua members yet each is very useful in organizingestablished knowledge and continuing inquiry in their respectivedomains And the same holds for ldquoconceptrdquo For example theneo-empiricist friendly to Macheryrsquos general take on our concep-tual systems might want to defend the substantive claim thatperceptual symbols are a kind of concept Such a claim would besubstantive to the effect that the delineated class contains someadditional sorts of entities Indeed Machery himself wonderswhether there are other kinds of concepts (p 249) For examplehe writes ldquoEvidence shows that people have some knowledgeabout ideals What is now needed is to determine whether thesebodies of knowledge qualify as conceptsrdquo (p 249) This strikes usas a meaningful and important question and one for which theterm ldquoconceptrdquo is obviously useful in asking

So independent of the question of hybrids psychologists shouldkeep the term Even if Machery is right and concepts are not anatural kind the potential dangers here would be better addressedthrough reformation instead of elimination The practical advice totake away from Macheryrsquos arguments may be not that scientistsshould get rid of ldquoconceptrdquo but that they should be more carefulin understanding that this term likely fails to pick out a very tidysort of natural kind Doing so should allow them to steer aroundthe sorts of dangers that Machery (2009) hypothesizes about(eg pp 242ndash243) without sinking an otherwise fruitful vehicleof inquiry The psychology of categorization inference and soon may be much messier than philosophers and psychologistshave hoped But ldquoconceptrdquo is still likely to be a vitally importantword for theorizing about that mess

Concept talk cannot be avoided

doi101017S0140525X10000361

James A HamptonDepartment of Psychology City University London London EC1V OHB

United Kingdom

hamptoncityacuk

wwwstaffcityacukhampton

Abstract Distinct systems for representing concepts as prototypesexemplars and theories are closely integrated in the mind and thenotion of concept is required as a framework for exploring thisintegration Eliminating the term ldquoconceptrdquo from our theories willhinder rather than promote scientific progress

While most people interested in concepts will find much to agreewith in this book (Doing without Concepts Machery 2009) it is theeliminativist thesis that will find most resistance Machery providesanalogical cases in psychology such as ldquoemotionrdquo and ldquomemoryrdquoEmotion and memory it is argued may prove to be terms referringto a varied set of phenomena without any identifiable single associ-ated brain system Similar cases can be found in other sciences ndashfor example ldquospeciesrdquo and ldquoplanetrdquo The concept of species is pro-blematic because there is not always a clear criterion for differen-tiating one species from another instead biological laws describethe distribution of genes over populations of individuals (Mayr1982) While problems of definition mean that ldquospeciesrdquo is not awell-defined term in biology it would however be hard toimagine biological discourse without it There are just too manygeneral truths that need to be expressed Similarly astronomersran into trouble with the designation of Pluto as a planet giventhe discovery of other large orbiting bodies that had beenlabeled as asteroids But the term still has a referential meaningScience needs more loosely defined general referring expressionsin addition to the carefully defined terms that figure in theoriesI argue that cognitive science still needs the notion of ldquoconceptrdquoeven if it proves multifaceted and hard to define satisfactorily

Macheryrsquos argument rests on there being three distinct formsof knowledge that are recruited by default by cognitive processes

namely prototypes (P) exemplars (E) and theories (T) Thedanger of eliminating the notion of concept is that the impor-tance of the relations between these forms of knowledge risksbeing underplayed First there is the obvious point that the PE and T representations (letrsquos call them PET) of dog all referto the same class ndash they are broadly co-referential (give or takesome differences in categorization resulting from exceptionalcontexts) What makes them co-referential is the fact that theyrepresent the same concept Without a notion of concept it ishard to explain why they co-refer

More importantly the term ldquoconceptrdquo is needed as part of anaccount of the many situations in which the PET systems inter-act How does one discuss concept combination including theformation of composite prototypes the importing of exemplarknowledge and the coherence checking of the result throughbackground theory if one cannot have the integrative termldquoconceptrdquo to specify just what it is that is being combined Thecombination occurs at the concept level and the description ofthe processes involved then requires elaboration in terms ofthe PET systems Similarly in concept learning we need an over-arching notion of concept in order to describe how PET systemsinteract Experiential concepts like DOG or CUP may first belearned by a child through interacting with individuals encoun-tered in everyday life When a variety of individuals are knownand it is necessary to learn to use the words ldquodogrdquo and ldquocuprdquo cor-rectly then prototypes may be formed enabling generalizationto other individuals discrimination of other classes and theaccumulation of generic knowledge As the child then developswider knowledge the prototype notion of DOG may be sup-plemented by theoretically driven concepts like mammal orspecies and by essentialist ideas about the causal properties ofbiological kinds or the need to defer to expert opinion aboutcorrect classification

Far from aiding scientific advance treating the PET systems aslargely independent of each other may impede investigation of theimportant ways in which information is transferred betweenthem It can also be argued that the three systems are not aseasily distinguished as Machery would require Consider proto-types and exemplars Machery agrees that much of the researchand debate concerning prototypes and exemplars has beendirected at a very restricted form of behavior namely learningto classify simple geometrical shapes in a laboratory settingwhere the categories to be learned are not easily distinguishedwithout extensive training Even in this arcane area of psychologythere is considerable evidence that under different conditionspeople will either learn individual exemplars or will abstract pro-totypes (Smith amp Minda 1998) If we move to the more ldquoconcep-tualrdquo domain of natural language terms then the question ofprototype versus exemplar models hardly arises For exampleStorms et al (2000) have investigated whether typicality in super-ordinate categories like FISH FRUIT or FURNITURE is bestpredicted by similarity to the category prototype or by similarityto ldquoexemplarsrdquo But in this case the exemplars are simply proto-types defined at a more specific level (eg CHAIR andTABLE) So the question is not which of two distinct systems isdriving the behavior but rather which level of abstraction isinvolved within a single representational system Some conceptsdo have genuine exemplars ndash the concept of ldquoBeethoven Symph-onyrdquo to a musician will be heavily dependent on knowledge of thenine exemplars But there will be a close link between knowledgeof the exemplars and generalized knowledge about the typicalstructure and expressive vocabulary found in the works

Likewise there has been a rapprochement between prototypeand theory-based elements of concepts In discussing the notionof prototype (Hampton 1998) I have proposed that the dis-tinguishing feature of prototype representations is that they rep-resent the center of a class and not its boundary It is this fact thatgives rise to category vagueness typicality gradients the lack ofexplicit definitions and the preponderance of generic (ratherthan necessary) features in peoplersquos accounts of the content of




the concept The notion of prototype as a form of schema is there-fore free to be supplemented by causal connections within therepresentation resulting in a structured frame representation(Barsalou amp Hale 1993) Mutability and centrality of propertiesmodal judgments of necessity and dissociations between simi-larity-based typicality and theory-based categorization can all beaccommodated within this single representational system

In short it is too soon to be counseling despair about integrat-ing prototype exemplar and theory-based representations into acoherent account of the concept of concept

Eliminating the ldquoconceptrdquo concept

doi101017S0140525X10000567

Stevan HarnadInstitut des Sciences Cognitives Universite du Quebec a Montreal Montreal

Canada H3C 3P8 School of Electronics and Computer Science University of

Southampton SO17 1BJ Southampton United Kingdom

harnadugamca

httpusersecssotonacukharnad

Abstract Machery suggests that the concept of ldquoconceptrdquo is tooheterogeneous to serve as a ldquonatural kindrdquo for scientific explanation socognitive science should do without concepts I second the suggestionand propose substituting in place of concepts inborn and acquiredsensorimotor category-detectors and category-names combined intopropositions that define and describe further categories

Whatever a ldquoconceptrdquo is we have at least one for every thing wecan recognize act on name or describe including not only thethings denoted by all the dictionary words we understand butalso everything we know what to do with (Harnad 2007) evenif we donrsquot know its name or it has none ndash perhaps becauselike ldquothings that are bigger than a breadboxrdquo no one has everbothered to name it

ldquoThingsrdquo can be individual objects (nonliving or living) kindsevents actions properties or states We have ldquoconceptsrdquo ofcountless such things and having the concept simply meansbeing able to do something with respect to those things anaction that has a right and wrong about it ndash anything fromapproachingavoiding the thing to interacting with or manipulat-ing it in some way identifying it (correctly) by name saying truethings about it imagining it and thinking and reasoning about it

In Doing without Concepts Machery (2009) suggests thatalthough there is no ldquonatural kindrdquo corresponding to the intersec-tion of prototypes examples theories and sensorimotor rep-resentations each may still turn out to be a legitimate naturalkind of its own I will sketch an alternative that scraps both theuse and the mention of ldquoconceptrdquo altogether

Consider conceptrsquos twin ldquoperceptrdquo If a concept is roughlyan ldquoideardquo then a ldquoperceptrdquo is an ldquoimagerdquo Should we ban talkof percepts too Pylyshyn (1973) suggested banning talk ofldquoimagesrdquo ndash as unobservable unmeasurable homuncular andmost important nonexplanatory ndash to be replaced by propositionsand eventually computations which are genuinely explanatoryin that they can generate the capacity that the images or ldquoperceptsrdquohad been meant to explain (Harnad 2006)

With findings on mental rotation (Shepard amp Cooper 1982)however ldquoperceptrdquo has made a comeback in the form of internalanalog structures and processes that have some of the propertiesof images but can do the internal generative work with nohomunculus sometimes more efficiently than computation(Digital computation can always approximate analog dynamicsas closely as we like A picture is always worth more than 1000words but 10000 words come closer It cannot however bewords all the way down Harnad 1990)

Apart from their sensory shapes objects have sensorimotorldquoaffordancesrdquo things that objects are amenable to having donewith them (by our bodies and their shapes) A chair (but not a

pyramid or a pincushion) affords sittability-upon a doornailbut not a doormat affords grasping and turning But is an affor-dance-detector a ldquorepresentationrdquo

We need to be able to recognize birds for example before wecan start doing anything with them including talking and think-ing about them No machine vision program could perform any-where near human level using prototype-matching to recognizebirds raw example-storage would do even worse And withoutthose verbal theories could not even get off the ground(because it canrsquot be words all the way down)

So what we need first is not bird representations but bird-detectors For most of us visuomotor contact is our first introduc-tion to birds but it is not ldquowerdquo who pick up the affordances weare no more aware of the tuning of our internal category detec-tors than subjects in mental-rotation experiments are aware ofrotating their inner images Internal mechanisms do this ldquoneo-empiricalrdquo work for us (Barsalou 1999 Glenberg amp Robertson2000) The work of cognitive science is to discover those mechan-isms That done it no longer matters whether we call them con-cepts ideas notions representations beliefs or meanings

Cognitive science has not yet done this job though Turing(1950) set the agenda long ago Scale up to a model capable ofdoing everything we can do (Harnad 2008) The first hurdle issensorimotor category detection the mechanism for learning cat-egories from sensorimotor interactions with the world guided byerror-correcting feedback We share this capability with mostother species learning to detect and act upon sensorimotor affor-dances To categorize is to do the right thing with the right kind ofthing (Harnad 2005)

Some categories are innate We recognize and know what to dowith them because natural selection already did the ldquolearningrdquo bygenetically pretuning our ancestorsrsquo brains But most categorieswe have to learn within our lifetimes including everythingnamed and described in our dictionaries plus many thingsactions events properties and states we never bother to nameWe learn to do the right thing with them and perhaps describethem on the fly How did we get those names and descriptionsOur species is the only one that has them

According to our account so far we only have the categories forwhich we have learned through direct experience what to do withtheir members One of the most adaptive things our species alonedoes with many of our categories is to name them For withlanguage evolved our capacity to produce and understand stringsof category names that encode truth-valued propositions predicat-ing something about something This allowed us to acquire newcategories not only by sensorimotor induction but also by verbalinstruction For once we have a set of categories ldquogroundedrdquodirectly in our sensorimotor capacity to detect their membersand nonmembers we can also assign each category an agreedarbitrary name (Harnad 1990) and then we can define anddescribe new categories conveying them to those who do notyet have them by combining and recombining the names of ouralready grounded categories (Cangelosi amp Harnad 2001) in prop-ositions Then and only then does the ldquotheory-theoryrdquo come in forverbal definitions and descriptions are higher-order category-detectors too as long as all their component terms are grounded(Blondin-Masse et al 2008) Here we are right to call them ldquorep-resentationsrdquo for they are descriptions of categories and can begiven to and received from others without every individualrsquoshaving to learn the categories directly from experience ndash as longas the category-names used in those descriptions are ultimatelygrounded in direct sensorimotor categories

There is much ongoing research on the mechanisms of sensor-imotor category learning in computers neural nets robots and thebrain as well as on the origins and mechanisms of natural languageprocessing It is nowhere near Turing-scale but this sketchrearranges the cognitive landscape a bit to preview how we canas Machery suggests do ldquowithout conceptsrdquo What takes theirplace is innate and mostly learned sensorimotor category-detectors(for which the learning mechanisms are still not known but



neither prototypes nor exemplars are likely to play much of a rolein them) progressively supplemented by verbal category represen-tations composed of grounded category names describing furthercategories through propositions The real challenge is getting thisto work Turing-scale Alongside that momentous and substantivetask which of the landmarks we elect to dub ldquoconceptsrdquo or ldquoideasrdquoseems pretty much a matter of taste [A fuller version of this com-mentary entitled ldquoConcepts The Very Ideardquo is available at httpeprintsecssotonacuk18029]

Defending the concept of ldquoconceptsrdquo

doi101017S0140525X10000373

Brett K Hayes and Lauren KearneySchool of Psychology University of New South Wales Sydney NSW 2052

Australia

bhayesunsweduau

lkearneypsyunsweduau

Abstract We critically review key lines of evidence and theoreticalargument relevant to Macheryrsquos ldquoheterogeneity hypothesisrdquo Theseinclude interactions between different kinds of concept representationsunified approaches to explaining contextual effects on concept retrievaland a critique of empirical dissociations as evidence for conceptheterogeneity We suggest there are good grounds for retaining theconcept construct in human cognition

The past decade has seen prolific growth in research on humanconcepts both in terms of the phenomena studied and thegeneration of explanatory models With research proceeding onso many fronts it is tempting to see the field as becoming fragmen-ted with little prospect of a unified theory However it is hard tosee how the field would be advanced by assuming that individualobjects and events are represented by ldquoseveral conceptsrdquo (Doingwithout Concepts Machery 2009 p 52) especially when the prop-erties of these alternate concepts and the demarcation betweenthem have not been specified Moreover we argue that thereare good reasons for rejecting the ldquoheterogeneity hypothesisrdquoHere we identify three problems with this view

1 Are objects really represented in multiple ways It is truethat a variety of theoretical models (prototype exemplartheory-based) have been proposed to explain how concepts arerepresented and used However Macheryrsquos implication thateach has equal explanatory power and therefore warrants statusas a separate kind of concept is incorrect While there are cer-tainly limits to the things that exemplar models can explain (seeMurphy 2002) there is also little doubt that exemplar modelsdo a better job of explaining laboratory data on category learning(see Kruschke [2005] for a review) and other key phenomena(eg the effects of category-irrelevant features on classificationas in Allen amp Brooks 1991) than models which only assumestorage of prototypical features

The ldquotheoryrdquo approach was originally proposed to explainaspects of conceptual experience that lie outside the purview ofldquodata-drivenrdquo approaches including sensitivity to conceptualcoherence and the causal basis of categories (Murphy amp Medin1985) However there have been significant advances towardsintegration of data-driven and theory-based approaches Heitrsquos(1994a) integration model incorporates prior knowledge into anexemplar model to successfully predict patterns of categorylearning in knowledge-rich domains (Carmichael amp Hayes2001) In Rehder and Murphyrsquos (2003) KRES model backgroundknowledge is represented in a connectionist network to explainhow data-driven category learning is accelerated in the presenceof knowledge and how ambiguous features are reinterpreted inthe light of feedback Similar efforts to integrate data-drivenand knowledge-based approaches have met with considerablesuccess in explaining category-based induction (eg Kemp ampTenenbaum 2009) Note that these are not ldquohybrid modelsrdquo in

the sense used by Machery They do not assume separate ldquoexem-plarrdquo and ldquoknowledgerdquo modules Instead prior knowledge is rep-resented in a format similar to that used for learning newexemplars (eg the integration model instantiates prior knowl-edge as retrieval of relevant exemplars from memory KRESdoes it via feature to feature associations and similaritybetween known and novel concepts) These models show thatthe conceptual knowledge associated with ldquotheoryrdquo and ldquoexem-plarrdquo approaches need not be thought of as independent andthat their interaction can be accounted for within a unified theor-etical framework

2 Different representations for different tasks One of the keysources of evidence cited in support of the heterogeneity hypoth-esis is that different aspects of object knowledge are activated inthe service of different goals or tasks Classification learning forexample focuses attention on the differences between cat-egories whereas inductive predictions rely on knowledge ofwithin-category structure (Markman amp Ross 2003) The firstthing to note is that this kind of flexibility is not limited to con-cepts It is a ubiquitous feature of human cognition Forexample changes in retrieval conditions lead to importantchanges in the kinds of memories that are activated (Roediger2008) but this does not mean that we should abandon thenotion of object ldquomemoryrdquo Second it is important to recognisethat empirical findings pointing to the ldquocontext dependencyrdquo ofconcepts can be oversold A consistent empirical finding is thatalthough context is important a core conceptual representationof a given object is often retrieved when the object is used in avariety of conceptual tasks (Markman amp Ross 2003) Forexample taxonomic features appear to be activated by default ina variety of tasks involving biological concepts whereas the acti-vation of causal relations is dependent on task goals and domainexperience (Shafto et al 2007)

Finally and most critically it is possible to develop models thathave the flexibility to explain context dependent changes inconcept retrieval ndash SUSTAIN (Love et al 2004) is a goodexample Such models achieve flexibility by incorporating well-established psychological processes such as selective attentionand discrimination learning and the use of a range of similaritymetrics However they do so while retaining an assumption of aconceptual ldquocorerdquo in object representation As well as explainingdata that Machery claims are problematic for standard modelsthese approaches may be extended to examine how conceptualknowledge influences performance in domains such as problem-solving (Markman amp Ross 2003) and recognition memory (Heitamp Hayes 2008) which have traditionally been divorced from thestudy of concepts

3 What kind of evidence is needed to test the heterogeneity

hypothesis Machery suggests that the heterogeneity hypothesisis supported to the extent that we can identify experimental orneuropsychological dissociations across different conceptual tasksAlthough similar views are frequently espoused in the psychologi-cal literature (eg Ashby amp Maddox 2005) they should be treatedwith considerable caution Deciding whether a given data patternsupports models that posit a single causal process (eg exemplarmemory) or multiple independent processes is a complex andtricky business Patterns of single or double dissociations in categ-orization performance across tasks or patient populations can beproduced by single-process models (eg Newell amp Dunn 2008)Rather than simply assuming that patterns of dissociation pointto a particular kind of cognitive architecture future progress inmapping conceptual representations will need to apply rigorousanalytical techniques such as state-trace analysis (Dunn 2008)and a careful comparison of well-articulated single- and multiple-process models

The empirical work summarised by Machery suggests that ourconcept representations are complex and multifaceted Howeverwe argue that the balance of empirical and modelling work showsthat the notion of ldquoa conceptrdquo remains a useful heuristic inpsychological explanation




The faux fake forged false fabricated andphony Problems for the independence ofsimilarity-based theories of concepts

doi101017S0140525X10000385

Anne J JacobsonDepartment of Philosophy University of Houston Houston TX 77204-3004

ajjacobsonuhedu

wwwclassuheduphiljacobson

Abstract Some things in our environment are not what they seem andthey provide a challenge to theories of concepts that emphasize similaritySection 1 of my commentary explores a dilemma this situation creates forMachery Section 2 describes a more general problem for prototype andexemplar theories Section 3 locates a place for similarity-based conceptsand indicates an alternative to Macheryrsquos thesis

1 A dilemma for Machery The large number of words we havefor potentially deceptive things attests to the fact that they label aphenomenon that is important to us Nonetheless the knowledgeoften essential to distinguishing between the real thing and some-thing else ndash causal knowledge ndash is left out of the prototype andexemplar accounts Thus Machery (2009) says ldquoneither proto-types nor exemplars store causal knowledgerdquo and so tasks thatrequire subjects ldquobring some causal knowledge to bearrdquo (Doingwithout Concepts p 187) are evidence for a third theory of con-cepts the theory or knowledge account As a consequence verygood mimics and imitations whose similarity to the real thing isoften undetectable to the casual observer answer to the criteriongiven by prototype and exemplar theories Many red pandas looka great deal like raccoons but they are not raccoons (Flynn et al2000 Sato et al 2009) Seeming gold rings may be made ofcopper cubic zirconias fool many people who take them to bediamonds and gopher snakes are close indeed in appearanceto rattlers though they are not poisonous

Machery argues that typically we have at least three indepen-dent but co-referential concepts for each kind of thing a proto-typical concept an exemplar concept and a knowledgetheoryconcept (see Murphy [2002] for the ldquoknowledgerdquo label) Butthere is a problem Cases of good fakes (eg cubic zirconias vsdiamonds) either fit a correct use of the prototypical or exemplarconcept or they do not Suppose they do if so then the prototy-pical or exemplar concept of a diamond will apply to things thatthe knowledgetheory account concept does not and the two willnot be co-referential

Alternatively suppose that cubic zirconias that fit the prototy-pical concept are not correctly called diamonds In this casetheir independence is threatened The criteria given by theknowledgetheory account are able to overrule those of theother two Machery resists the idea that the knowledgetheoryconcept can so dominate (Ch 3 sect 33) However given hisclaim that the sentence ldquoTina Turner is a grandmotherrdquo(Machery 2009 p 72) is true under one interpretation and falseunder another the example shows that if one type does not dom-inate different uses of ldquograndmotherrdquo will vary in reference

The argumentative context in which Machery considers fakesleads him to restrict his attention to sentences Further his intui-tion does not clearly say that ldquoFake dollars are dollarsrdquo is false(2009 p 72) But as Machery insists the range of the use of con-cepts covers far more and intuitions about sentences do notsettle whether criteria from the knowledgetheory account willoverrule in practice

2 A more general problem The failure to encode causal knowl-edge creates another problem for prototype and exemplar con-cepts and by implication Macheryrsquos thesis Experimental workin Macheryrsquos discussion of prototype and exemplar theories islargely represented by descriptions of testing subjects on listssentences pictures and drawings including patterns of dotsIn contrast Murphy (2002 p 60) remarks that his knowledgetheory account rejects the idea that we ldquolearn concepts in

isolation from everything else (as is the case in many psychologyexperiments)rdquo But our classifications of objects need to work toidentify and track them in a dynamical environment That sort oftask carries quite different demands among other things classi-fication needs to yield some clues about how our environmentwill unfold

Infants have a capacity relevant to identifying kinds in a chan-ging environment one that Carey and Xu (2001) claim controver-sially does not appear until 12 months That is at some pointinfants expect kinds of objects to persist and not change intoanother kind Before then mere spatial-temporal continuitydominates a duck moving behind a screen and an emergingrabbit do not have to be two distinct objects Knowledge thatkinds persist and do not turn into one another looks like causalknowledge of how things work and Carey and Xu maintain it ispreceded by other causal knowledge Further the presumptionthat kinds persist facilitates the acquisition of knowledge abouthow objects of some kind interact with their environmentSuch knowledge is important because it enables one to anticipatesome of what will happen

From this perspective there is at least a tension betweensaying that concepts are important in classifying and sayingthat they do not encode causal knowledge It is unclear whyusing non-causal concepts would have any survival advantagesexcept in situations where the tasks to be performed haverestricted success conditions We look at such a task in the nextsection

3 Conclusion Prototypes and exemplar theories may stillcharacterize what is needed to fulfill some of the tasks visionneeds to undertake One is that of perceptual organization Ourvision at one stage is a succession of saccades the successive-sac-cades stage leads to our experience of a scene of stable objectsThis important result appears to be driven by gestalt groupingsbut it can be facilitated by top-down categorizing that may inmany cases be triggered by similarity to a prototype or exemplar(Ogmen 2007 Ogmen et al 2006) Success in this sort of task issimply getting some organization It can be achieved even if oneis being taken in by fakes and the faux

The preceding material suggests that there are layers to ouruses of concepts ones distinguishable in terms of the complexityof the knowledge needed In creatures made for action percep-tual organization is required for almost any vision-guided actionIts success conditions may be much less demanding than those ofanother task anticipating how onersquos environment is going tounfold Finally human cognitive life and public communicationamong other things have a considerable interest in getting itright and not being taken in by potential deception These differ-ent stages can be seen as posing less and more demanding tasksfor our uses of concepts Machery has instead posited multipleindependent processes but we have seen reasons for questioningthe independence That said one should nonetheless expect thatbold hypotheses from a subtle mind like Macheryrsquos will provemore resilient than one might first expect

ACKNOWLEDGMENTSI am indebted to Josh Brown and Josh Weisberg for discussions of thismaterial

Hybrid vigor and conceptual structure

doi101017S0140525X10000555

Frank KeilDepartment of Psychology Yale University New Haven CT 06520

frankkeilYaleedu

Abstract Machery rightly points out a diverse set of phenomenaassociated with concepts that create challenges for many traditional



views of their nature It may be premature however to give up suchviews completely Here I defend the possibility of hybrid models ofconcept structure

In Doing without Concepts Machery (2009) provides a service tous all by reminding us of the challenges of specifying what con-cepts are and how they are mentally represented Moreover bymoving to the radical position that we should do away with theconcepts altogether he forces all of us to think more deeplyabout why we might want to preserve such notions Finally heis correct in pointing out the ways in which philosophical andpsychological approaches to concepts often seem to be askingand answering different questions Despite all this it seems tooextreme to assume that more traditional notions of conceptsare bankrupt or that philosophers and psychologists are alwaystalking past each other

Here I focus on one alternative to Macheryrsquos proposal that heseems to dismiss too lightly ndash the idea that concepts might have ahybrid structure I am not yet sure about the extent to which thepotentially hybrid facets to concepts are actually parts of the con-cepts per se or whether they should instead be considered aslinked to concepts that are themselves much simpler atoms inthe manner described by Fodor (1998) Cognitive science stillhas not fully answered Fodorrsquos reasons for doubting that manycognitive phenomena associated with concepts reflect internalstructural features of concepts as opposed to aspects of how weuse and work with concepts but to accept Fodorrsquos argumentsis to discount Macheryrsquos view as well so let us assume herethat we do want to assume internal structures to representationsof concepts and that those structures help explain many psycho-logical phenomena associated with concepts such as inductioncategorization and conceptual change

Fodor (1998) has characterized concepts as ldquothe smallest unitsof thoughtrdquo and in this respect many psychologists and philoso-phers agree Does such a characterization compel us to Mach-eryrsquos heterogeneity hypothesis namely that we must haveseveral distinct concepts of water because depending oncontext we seem to use the concept in different ways It is diffi-cult to see why Machery discounts the hybrid alternative byarguing that people will endorse conflicting statements aboutkinds such as tomatoes whales and the like and those conflictscan only be explained by assuming that they are drawing ondifferent concepts These different uses are supposedly notldquocoordinatedrdquo and therefore people cannot be referring to thesame concepts But this coordination problem does not seem tobe so lethal for hybrid views If a given concept has a hybridstructure consisting for example of typicality-based informationcausal schema functional relations and logical entailments itmight well be the case that different contexts cause people toweight those properties very differently and respond in differentmanners across tasks Ever since Lakoffrsquos (1972) demonstrationsthat different ldquohedgesrdquo such as ldquotechnically speakingrdquo andldquoloosely speakingrdquo can cause us to categorize kinds such aswhales differently it has been known that some hallmark wayswe use concepts such as categorization can show strong vari-ations as a function of situational and sentential concepts Butif hybrid models are right they seem more than adequate fordealing with such phenomena Machery would need to providea detailed model of internal hybrid representations of conceptsthat showed how they were intrinsically unable to computation-ally model such effects and he has not yet done so

Machery suggests that the parts of hybrid concepts must beldquocoordinatedrdquo such that this cannot create inconsistencies suchas categorization judgments that whales both are and are notfish This coordination property is seen as an essential part ofhybrid models and hybrid models are described as incoherentor empty without it This was not an obvious conclusion Con-sider for example contexts in which we might describe aperson as ldquoshortrdquo and then ldquotallrdquo If we see a 2-meter person prac-ticing with Olympic gymnasts we might well call him tall butwhen observing him practice with an Olympic basketball team

we would call him short We may well know his true heightand the true heights of the other players but the contexts callfor different ways of assigning thresholds on the vertical dimen-sion that we would then use to consider someone tall or short Isthis to be taken as evidence for multiple concepts of tall andshort Are there then an indefinitely large number of such con-cepts that are depending on all the micro-contexts that couldshift the thresholds to tiny degrees There is a strong tendencyto resist such a route and it seems that for similar reasons weshould resist claims that hybrid structures are undermined byconflicting categorization judgments in different contexts Categ-orization inconsistencies do not pose a problem if there are stillsystematic ways that categorization judgments can be shown tovary across contexts as a function of a description of their internalhybrid structure Machery would be correct in pointing out thatsuch systematic accounts are not yet fully worked out but thereare no obvious reasons why they might not be in the long run

Hybrid approaches also have other appeals They can forexample provide continuity and coherence to models of concep-tual change over time as for example when the causal or rule-based aspects of a concept become more differentiated as achild grows older and come to be weighted more and more rela-tive to the associative components (Keil 1989 Keil amp Newman2010) Macheryrsquos heterogeneity alternative sees the child as pro-gressing through a series of unconnected concepts that somehowmagically tend to unfold in the same way across children Hybridmodels can also help explain how concepts differ across broadcategories such as natural kinds and artifacts where differentcomponents of the hybrid might be present to differentdegrees and accordingly assigned different weights as well asbeing processed in different ways (Hampton et al 2009)

Machery has done us all a great service He raises a host ofinteresting troubles for many accounts of concepts and he is tobe commended for trying to build a larger common ground ofinquiry between philosophers and psychologists His book is arefreshing new perspective that prods all of us to furtherdevelop our own theories of concepts

The function and representation of concepts

doi101017S0140525X10000397

Sangeet S Khemlania and Geoffrey Goodwinb

aDepartment of Psychology Princeton University Princeton NJ 08540bDepartment of Psychology University of Pennsylvania Philadelphia PA

19104

khemlaniprincetonedu

ggoodwinpsychupennedu

httpwwwprincetonedukhemlani

httpwwwpsychupennedupeopleggoodwin

Abstract Machery has usefully organized the vast heterogeneity inconceptual representation However we believe his argument is toonarrow in tacitly assuming that concepts are comprised of onlyprototypes exemplars and theories and also that its eliminative aspectis too strong We examine two exceptions to Macheryrsquos representationaltaxonomy before considering whether doing without concepts is a goodidea

In Doing without Concepts (Machery 2009 hereafter DwC)Machery proposes that heterogeneity in the mental represen-tation of ldquoconceptsrdquo is sufficient to render that term useless Ashe argues the term can refer to exemplars prototypes and the-ories However it can also refer to defaults (Connolly et alFodor et al 2007) aspects (Prasada amp Dillingham 2009)Boolean concepts (Feldman 2000 Goodwin amp Johnson-Laird2010 submitted Shepard et al 1961) and connections yet tobe discovered Thus in our view Macheryrsquos taxonomy is toonarrow and it underestimates the degree of heterogeneity that




exists in the representation of concepts It excludes a variety ofconceptual phenomena that do not fall within the purview of pro-totypes exemplars and theories We turn next to describe twoexamples of such conceptual phenomena ndash Boolean conceptsand connections expressed by generics

Boolean concepts ndash those that are composed out of negation(not) disjunction (or) and conjunction (and) ndash are an importantkind of everyday concept They occur frequently in the form oflaws rules or procedures Indeed Macheryrsquos criteria for individ-uating concepts are themselves Boolean concepts (see sect 2 ofMacheryrsquos Precis of DwC in this issue) A concept does not needto be entirely Boolean in order for it to contain relevant Booleanstructure however How individuals learn Boolean concepts isstill not resolved but many of the current leading contendersare not based on exemplars prototypes or theories (Feldman2006 Vigo 2009) We have recently proposed an alternativetheory based on mental models which analyzes the complexityof concepts in terms of the number of distinct possibilities thata concept can be compressed into (Goodwin amp Johnson-Lairdsubmitted) This theory predicts the acquisition of Boolean con-cepts as well as if not better than the other leading contendersand it too is not based on the representational mechanisms thatMachery assumes to be exhaustive in explaining conceptualknowledge

Recently we published a paper documenting the occurrenceof ldquoconceptual illusionsrdquo in which people think that particularinstances of a Boolean concept are possible when in fact theyare not and vice versa (Goodwin amp Johnson-Laird 2010) Atypical example is the concept green and large or else greenin a context in which all possible objects are either green ornot and either large or small (the ldquoor elserdquo here represents exclu-sive disjunction) A large percentage of individuals think that anobject that is both green and large is possible given this descrip-tion which is in fact erroneous The exclusive disjunctionbetween the two clauses means that the only possible object isone that is both green and small This error as well as otherslike it is predicted by the mental model theoryrsquos principle of con-ceptual truth And as far as we can tell accounts based on proto-types exemplars or theories have no way to explain these errors

Other examples of conceptual phenomena unaccounted for bythe taxonomy described in DwC include the connections andrelations that link concepts together Such connections can beconcepts unto themselves and are revealed by generic assertionssuch as ldquotigers are stripedrdquo ldquobarns are redrdquo or ldquoticks carry Lymediseaserdquo which express generalizations about kinds of things(Carlson amp Pelletier 1995 Gelman 2003 Lawler 1973) Allthree assertions are true for different reasons and as such gen-erics provide a means for studying the types of connections werepresent between concepts of kinds and properties We havefound that for statements such as ldquotigers are stripedrdquo the relationbetween the kind (ldquotigersrdquo) and the predicate (ldquoare stripedrdquo) canbe distinguished from logical statistical and causal connections(Khemlani et al submitted Prasada et al submitted) Thesedistinctions could account for phenomena in concept learningand conceptual development without importing any assumptionsmade by other theories of concepts Thus by studying genericsit is possible to discover the conceptual structure of generaliz-ations without assuming the representational structure of theconcepts to which they refer Machery proposes that futureresearch should examine differences between generics andparticularly how they differ from quantified assertions (DwCp 200) and we agree wholeheartedly Unfortunately the pro-posals in DwC do not leave room to explore such advances in con-ceptual organization as they encourage researchers to couch theirwork as falling within the domain of three fundamental classes ofconceptual representation

We do not think these phenomena which point to even greaterheterogeneity in the mental representation of concepts thanMachery suggests strengthen Macheryrsquos eliminativist argumentto do away with concepts The elimination of the term

ldquoconceptrdquo in favor of greater specificity tacitly endorses theassumption that prototypes exemplars and theories (and what-ever else) are all that comprise concepts It thus presupposesthat contingent facts about the mental representation of conceptsare the sole criterion for deciding whether ldquoconceptrdquo ought to bepreserved But this presupposition ignores the common functionthat diverse sorts of concept play in representing knowledge andin communication Concepts represent and convey systematicbodies of information and they would retain this function regard-less of how they are mentally represented In other words wethink that the question of what counts as a concept needs answer-ing at the computational level not at the algorithmic one (cfMarr 1982)

Thus Macheryrsquos eliminativist argument is too powerful Itgives no grounds for thinking that the term ldquoconceptrdquo is in anespecially precarious position In much the same way that theterm organizes a wide array of representational processes sotoo do terms like ldquothinkingrdquo ldquoattentionrdquo and ldquomemoryrdquo WouldMachery have us do away with these terms as well given hetero-geneity in the cognitive processes to which they refer Perhapsbut we think this is going too far

In sum the key functions of concepts are to represent andcommunicate knowledge and this general functional propertyis what argues in favor of preserving the term ldquoconceptrdquo Webelieve that heterogeneity at the level of mental representationis no obstacle to the further empirical investigation of concepts

Concepts are a functional kind

doi101017S0140525X10000403

Elisabetta LalumeraDipartimento di Psicologia Universita di Milano-Bicocca 20126 Milano Italy

and Cogito Research Centre in Philosophy Universita di Bologna 40100

Bologna Italy

elisabettalalumeraunimibit

httpsitesgooglecomsiteelisabettalalumera

Abstract This commentary focuses on Macheryrsquos eliminativist claimthat ldquoconceptrdquo ought to be eliminated from the theoretical vocabularyof psychology because it fails to denote a natural kind I argue for themore traditional view that concepts are a functional kind whichprovides the simplest account of the empirical evidence discussed byMachery

The novelty of Macheryrsquos proposal in Doing without Concepts(Machery 2009) is the claim that the term ldquoconceptrdquo ought tobe eliminated from the theoretical vocabulary of psychologybecause it fails to denote a natural kind I will not dispute theclaim that concepts are not a natural kind There is a growingconsensus among psychologists that the structure of conceptsmay vary along many dimensions depending on expertisedomain of objects categorized and conceptual task involvedMuch of this evidence is reviewed by Machery himself as wellas in other recent studies in the philosophy of psychology (Picci-nini amp Scott 2006 Weiskopf 2009b) My point here is that onphilosophical grounds this evidence is perfectly compatiblewith the much less revisionary claim that concepts are a func-tional kind Something is a concept by virtue of the function itperforms within a cognitive system and something is theconcept of a certain category C (at least partially) by virtue ofthe further specific function of representing it It is a furtherquestion whether or not the functional kind ldquoconceptrdquo is realizedby natural kinds (Weiskopf 2009b) Functional kinds can be indi-viduated and described independently of their realizers Thishowever does not deprive them of a central role in experimentalpsychology

1 Concepts are a functional kind It is disputable that a fullcharacterization of the psychological usage of ldquoconceptsrdquo is



ldquothose bodies of knowledge that are used by default in theprocesses underlying the higher cognitive competencesrdquo(Machery 2009 p 7) Concepts are also normatively character-ized as those bodies of knowledge that can perform a doublefunction ndash namely abstraction and projection of knowledge ndashand that are able to be recombined almost freely in order toform more complex concepts and thoughts Let us focus on thefirst two functions as the third one is arguably not specific to con-cepts only Abstraction is the ldquobottom-uprdquo process of extractinginformation from a single encounter with an object or property-instance and generalizing such information to all encounterswith that object or property The experience of tasting rhubarbonce and finding it bitter would be of no use if I could not storeit as information about rhubarb independently of the specificepisode of tasting it by means of a general representation ndash aconcept (Bloom 2000) Category induction is the complementaryldquotop-downrdquo process of projecting such knowledge to new encoun-ters When you tell me you like rhubarb pie I form the expectationthat you will also like other bitter-tasting foods This is an appli-cation of my concept of rhubarb Thus the function of a conceptis that of a ldquomental gluerdquo which connects onersquos past experiencewith the present (Millikan 1998 Murphy 2002) It is becausethey perform this complex function that concepts are used bydefault in our higher cognitive capacities and not vice versaPsychologists are interested in discovering how such function isperformed ndash by what structures and mechanisms The functionalproperties of concepts provide constraints on what may count asan adequate concept-realizer This is the first sense in whichconcepts as functional kinds are not dispensable in psychology

2 Concepts as a functional kind are multiply realizable in a

broad sense Vehicles and food are familiar non-mental func-tional kinds Traditionally the essence of a mental functionalkind is to bring about some outcome or to exercise some capacity(Kim 1992 Putnam 1967) With ldquobroad senserdquo here I mean thatthey are realizable by items characterized by different structuralproperties Assuming that abstraction and category induction arethe fundamental capacity characterizing the functional kindldquoconceptrdquo there are plausibly many ways in which a humanmind can organize itself in order to exercise them Capacitiesmay well be individuated by their ends not just by their means(Millikan 1998) So for example a chemistrsquos capacity to abstractand project information about water may involve means that I donot possess But so long as we are both able to abstract accumu-late and project information about water we share the samecapacity that is we both possess a concept of water The variabil-ity of structure of the functional kind ldquoconceptrdquo is just as compa-tible with the empirical evidence discussed by Machery as hisown proposal is It has the advantage of leaving it openwhether prototypes exemplars and theories exhaust the possibleconcept-realizers or more plausibly not

3 Concepts as kinds are multiply realizable in a narrow

sense The narrow sense is the traditional sense associatedwith the phrase ldquomultiple realizabilityrdquo in philosophy of mindIt is the idea that a token (mental) functional kind can be realizedby different physical substances The classical example is a stateof pain multiply realized by silicon chips or C-fibers To denythat concepts are multiply realizable in this narrow sense isvery demanding It requires a commitment to a strong form ofphysicalism Contemporary neurophysiological research on con-cepts aims at individuating which areas of the human brain areinvolved in specifically conceptual tasks and explaining how thisis done It is not committed however to the further metaphysicalclaim that no other kind of matter appropriately organized couldbring about the same capacities If neurophysiology is not surelypsychology need not be committed to a robust reductionistagenda That is there is no reason why psychology should dis-pense with the functional kind ldquoconceptrdquo qua multiply realizableThe neurophysiological evidence quoted by Machery on thevariability of concepts is compatible with the anti-eliminativistfunctional view

4 The functional kind ldquoconceptrdquo has a role in

psychology Machery claims that there are pragmatic reasonsfor the elimination of ldquoconceptrdquo if intended as a natural kindterm This is not so if it is taken as a functional kind termConsider the task of accounting for the fact that onersquos generalrepresentation of a category ndash say dogs ndash changes over timeInitially it is constituted by a bunch of exemplars then it developsinto a summary prototypical representation and later it becomesa theory In order to describe the three structures as stages of adiachronic process of change one needs to make reference towhat is common to them namely realizing the capacity of repre-senting dogs generally or being instances of the functional kindldquoconcept of dogsrdquo This intrapersonal explanation properlybelongs to the psychologistrsquos agenda and so do interpersonalaccounts of variability

From conceptual representations toexplanatory relations

doi101017S0140525X10000415

Tania LombrozoDepartment of Psychology University of California Berkeley CA 94720

lombrozoberkeleyedu

httpcognitionberkeleyedu

Abstract Machery emphasizes the centrality of explanation for theory-based approaches to concepts I endorse Macheryrsquos emphasis onexplanation and consider recent advances in psychology that point tothe ldquoheterogeneityrdquo of explanation with consequences for Macheryrsquosheterogeneity hypothesis about concepts

The ldquotheoryrdquo approach to concepts one of three that Macheryendorses in Doing without Concepts (2009) reflects a widespreadview in psychology that is typically taken to be promising but inneed of further development (eg Carey 1985 Gopnik amp Meltzoff1997) According to this approach concepts constitute or areembedded within intuitive theories about the world Howeverboth philosophers and psychologists have been quick to pointout that the notion of an intuitive theory is underspecified andthat appeals to scientific theories run the risk of substituting onemystery for another (Laurence amp Margolis 1999) Machery isaware of these concerns but he identifies the locus of the theoryviewrsquos commitments not in the notion of an intuitive theory or inappeals to scientific theory but in explanation For an advocateof the theory approach concepts store ldquoknowledge that canexplain the properties of the category membersrdquo so ldquomuchhangs on the notion of explanationrdquo (Machery 2009 p 101)

Machery is right to focus on the central role of explanation fortheory-based approaches to concepts In a set of influentialpapers that kick-started theory theorizing theories weredefined as ldquoany of a host of mental explanationsrdquo (Murphy ampMedin 1985 p 290) and characterized in terms of ldquolaws andother explanatory mechanismsrdquo (Carey 1985 p 201) Categoriz-ation and category learning were described as ldquospecial cases ofinference to the best explanationrdquo (Rips 1989 p 53) with aconcept invoked ldquowhen it has a sufficient explanatory relationto an objectrdquo (Murphy amp Medin 1985 p 295) But suchappeals to explanation were not proposed against a backdrop ofpsychological theories of explanation Rather they predomi-nantly appealed to what Machery calls a ldquofolk understanding ofexplanationrdquo (2009 p 102) contributing to the concern thattheory-based approaches to concepts are underspecified

What to do One option would be to develop psychologicaltheories of explanation with our folk understanding as a guideA second option would be to ground psychologistsrsquo appeals toexplanation in philosophical theories of scientific explanationGiven the analogy between science and cognition that motivates




many proponents of the theory approach this seems like anatural move But it is not the one Machery advocates In factMachery cautions that ldquophilosophical accounts of scientific expla-nation would probably be useless for spelling out the psychologi-cal notion of theoryrdquo (2009 p 102)

I want to suggest that Machery is wrong to dismiss the psycho-logical value of theories of explanation from the philosophy ofscience and to neglect recent advances that move the psychologyof explanation beyond ldquofolk understandingrdquo Scientists are afterall psychological creatures and there is every reason to expectthe aspects of human cognition that shape everyday explanationsto play a role in science Scientists and everyday cognizers alsoface similar problems and have similar goals They confrontlimited data and from this they must construct a representationof the world that supports relevant predictions and interventions

But there is another reason to expect a close correspondencebetween philosophical and psychological accounts of explanationone that stems from the philosophical methods typicallyemployed Here in uncharitable caricature is how theory devel-opment often proceeds Philosopher P1 proposes theory T1 ofexplanation philosopher P2 quickly generates putative counter-example C a specific case in which T1 makes one predictionabout what is explanatory but philosopher P2rsquos intuitiondemurs The philosophical community pronounces one way orthe other based largely on shared intuitions about C so T1

stands (for now) or gives way to a new theory This is not themost efficient way to collect data and it would not pass musterfor an experimental psychologist but to the extent philosophersare like everyday folk one would expect convergence betweenphilosophical theories and descriptively adequate accounts ofeveryday intuitions

In fact a growing body of experimental work suggests thattheories of explanation from philosophy can usefully inform thepsychology of explanation and bear a close correspondence toeveryday judgments (for reviews see Keil 2006 Lombrozo 2006)While there is no consensus on a theory of explanation in philos-ophy different strands of theorizing seem to capture differentaspects of the psychology of explanation For example somestudies on the role of explanation in category learning have drawnon subsumption and unification accounts of explanation (egWilliams amp Lombrozo in press) while others on categorizationand inference are consistent with causal theories (eg Rehder2003b 2006) Empirical research on the cognitive significanceand consequences of different kinds of explanations ndash specificallyfunctional versus mechanistic explanations (Kelemen 1999Lombrozo 2009 under review Lombrozo amp Carey 2006Lombrozo et al 2007) ndash has its roots in Aristotle but can trace apath to contemporary philosophers such as Daniel Dennett

One reason to appreciate this richer philosophically informedpsychology of explanation is because it has implications forMacheryrsquos heterogeneity hypothesis In particular the two distinctsummary representations that Machery recognizes ndash theories andprototypes ndash can be understood as embodying different kinds of(potentially) explanatory structure Machery recognizes thispoint and in fact rejects philosophical accounts of explanationsuch as Salmonrsquos statistical relevance model in part becauseallowing statistical relationships to play a role in explanationwould ldquoblur the distinctionrdquo between theories and prototypes(Machery 2009 p 102) But perhaps the fact that explanationsare sensitive to causal and statistical relationships is a reason toendorse such accounts Evidence suggests that explanations aresensitive to multiple kinds of knowledge ndash about causal structureand functional relationships (Lombrozo amp Carey 2006) aboutstatistical regularities (Hilton amp Slugoski 1986) and about prin-cipled generic knowledge (Prasada amp Dillingham 2006) Theseare precisely the kinds of knowledge that Machery suggestsconcepts contain

Recognizing the ldquoheterogeneityrdquo of explanatory structure doesnot eliminate the heterogeneity of concepts but it does suggest apath to unifying concepts by appeal to explanation It also pushes

back Macheryrsquos concerns about natural kinds and eliminationfrom concepts to explanation What are the distinct kinds ofexplanatory relations and do they as a class support relevantgeneralizations that suggest ldquoexplanationrdquo is a natural kind anda valuable theoretical term for a mature psychology Perhapsthese are the questions we should be asking

Concepts and theoretical unification1

doi101017S0140525X10000427

Eric Margolisa and Stephen Laurenceb

aDepartment of Philosophy University of British Columbia Vancouver BC

V6T 1Z1 Canada bDepartment of Philosophy University of Sheffield Sheffield

S3 7QB United Kingdom

margolisinterchangeubcca slaurenceshefacuk

httpwebmaccomericmargolisprimary_sitehomehtml

httpwwwshefacukphilosophystaffprofilesslaurencehtml

Abstract Concepts are mental symbols that have semantic structure andprocessing structure This approach (1) allows for different disciplinesto converge on a common subject matter (2) it promotes theoreticalunification and (3) it accommodates the varied processes thatpreoccupy Machery It also avoids problems that go with hiseliminativism including the explanation of how fundamentally differenttypes of concepts can be co-referential

In Doing without Concepts Machery (2009) claims that philoso-phers and psychologists are not talking about the same thingwhen they use the term concept and that this is a consequenceof their having differing explanatory interests But there arereasons to reject Macheryrsquos division between philosophical andpsychological subject matters regarding concepts First weshould recognize the significant influence that philosophicaland psychological theorizing have had on each other Forinstance prototype theorists have been inspired by philosophicalcritiques of definitions theory-theorists have drawn upon philo-sophical accounts of natural kind terms and developmental psy-chologists have prioritized addressing the philosophical challengeof explaining how learning enriches a conceptual system Like-wise philosophers have been deeply influenced by psychologicalwork on typicality effects essentialist thinking and conceptualchange in childhood (to name just a few examples) Secondeven where philosophers and psychologists do have differingexplanatory agendas the same can be said of just about anytwo fields in cognitive science and in science generally Linguistsand psychologists have differing explanatory aims too as docognitive psychologists and neuro-psychologists ndash not unlike biol-ogists and chemists This hardly shows that theorists in these fieldsarenrsquot talking about the same thing (eg NPs conditioning orDNA) Third there is a payoff to identifying a single subjectmatter that underlies the efforts in philosophy psychology andother areas of cognitive science The result is greater theoreticalunification ndash a prized explanatory virtue

Now concept is a term of art But we would suggest thatMachery gets off on the wrong foot by characterizing conceptsas ldquobodies of informationrdquo Instead concepts should be takento be mental symbols that have semantic structure (which fixesthe propositional content of thoughts via a compositional seman-tics) and processing structure (which explains how conceptsfigure in various mental processes) Rather than saying thatprototypes exemplars and theories constitute fundamentallydifferent types of concepts it is better to locate such bodies ofinformation in a conceptrsquos processing structure On thisapproach the concept DOG is akin to a word in a sentenceand its structure includes a prototype a theory and so forth(Laurence amp Margolis 1999) The principal advantage toviewing concepts in this way is that it makes sense of how philo-sophers and psychologists can be talking about the same thing



while illuminating the fertile cross-disciplinary interactions thatthe study of concepts enjoys And though we (two philosophers)are promoting the idea that concepts are mental symbols this isnot an exclusively philosophical viewpoint Versions of it areendorsed by many cognitive scientists (eg Carey 2009 Jackend-off 2002 Pinker 1997 Pylyshyn 2007 Sperber amp Wilson 1995)

Is our account of concepts a hybrid theory Yes and no It doesbring together prototypes exemplars and theories by saying thatthey are bound to the same mental symbols The concept DOGfor example sometimes activates a prototype sometimes exem-plars and sometimes a theory But a concept need not haveeach type of processing structure and the activation of onepart does not require activating other parts Machery arguesthat the heterogeneity hypothesis has the explanatory advantageof accounting for the diverse psychological processes that areassociated with higher cognitive capacities But a theory thatunites diverse processing structure through links to a commonmental symbol can handle this diversity just as well

Machery asks why theorists who reject the heterogeneityhypothesis do not concede that the various bodies of information(the prototype theory etc) amount to distinct yet co-referentialconcepts of fundamentally different types ndash his own view (2009p 64) But what justifies Macheryrsquos claim that on his account adog prototype a dog exemplar and a dog theory are co-referen-tial To the extent that these representational structures havereferents the referents are hardly likely to be the same Forexample a dog-prototype would refer to things that are similarto the central tendency that the prototype describes while adog-theory would cover things that are at odds with the centraltendency (eg the offspring of two dogs that doesnrsquot look any-thing like typical dogs) By contrast on our account the issueof concept identity is easily handled in terms of the type identityof the mental symbol that unifies these various knowledge struc-tures This symbolrsquos identity is a matter of what it refers to plusfeatures of the representationrsquos vehicle for distinguishing amongco-referential concepts with differing cognitive significance(Margolis amp Laurence 2007)

As realists about concepts we also do not find Macheryrsquos casefor eliminativism compelling For one thing we would argue thatconcepts as we understand them do constitute a natural kind byMacheryrsquos criteria But also Macheryrsquos standard for the reality ofpsychological kinds is too high If his standard were enforced ndash ifa kind has to play an important role in many scientific generaliz-ation beyond those that characterize it ndash wersquod have to give up onmany core psychological constructs such as module compu-tation and representation But while these high-level kindsmay not satisfy Macheryrsquos criteria for realism they play a keyrole in describing the mindrsquos operations and helping scientiststo empirically investigate its overall organization MoreoverMacheryrsquos standard probably cannot even be maintained for hisfundamentally different types of concepts for example numer-ous distinct types of structures tend to get lumped togetherunder the heading of a theory And though we lack the spaceto press the point here Macheryrsquos approach to eliminationwould also have dire consequences outside of psychology Argu-ably we would have to give up most high-level kinds includingcell vertebrate and chemical element

In sum a realist account of concepts as mental symbols withboth semantic and processing structure addresses the explana-tory concerns that Machery raises while avoiding the problemsconnected to his eliminativism Taking psychological and philoso-phical theories of concepts to be about a single subject matterallows for far greater theoretical unification placing concepts atthe center of a broad investigation into the nature of cognitiveprocesses cognitive development meaning justification andthe mindrsquos relation to the world

ACKNOWLEDGMENTEric Margolis would like to thank Canadarsquos Social Sciences andHumanities Research Council for supporting this research

NOTE1 This article was fully collaborative the order of the authorsrsquo

names is arbitrary

Where are naturersquos joints Finding themechanisms underlying categorization

doi101017S0140525X10000439

Arthur B MarkmanDepartment of Psychology University of Texas Austin TX 78712

markmanpsyutexasedu

httpwwwpsyutexasedupsyFACULTYMarkmanindexhtml

Abstract Machery argues that concepts are too heterogeneous to bea natural kind I argue that the book does not go far enough Theoriesof concepts assume that the task of categorizing warrants a unique setof cognitive constructs Instead cognitive science must look acrosstasks to find a fundamental set of cognitive mechanisms

There is a persistent worry that cognitive scientists may not becarving nature at its joints This fear underlies debates overwhether computational representations or dynamical systemsbest explain cognitive processing (eg Markman amp Dietrich2000 Spivey 2007) It lies at the heart of critiques of the useof brain imaging to understand cognitive function (Uttal 2001)This issue is also central to the target book Doing withoutConcepts (Machery 2009)

This question is important because cognitive scientists typicallyorganize theories around tasks Memory is explored by havingpeople study items and then probing their memory for thoseitems at some later time Decision-making research involves pre-senting people with a set of options and having them select oneAs Machery points out categorization research often involvesspecific tasks such as classification and category-based inductionTheories then aim to explain performance in these tasks

Machery takes the structure of the cognitive science literatureon concepts as a given and then suggests that the notion of aconcept is misleading On his view there are (at least) three dis-tinct types of concepts prototypes exemplars and theoriesUsing a single term ndash concepts ndash to refer to all of these isdangerous because it fails to carve nature at its joints

I suggest that the problem is even worse than Machery makesit out to be Fundamentally the set of tasks that we study involvesa series of cross-cutting cognitive mechanisms At present cogni-tive science assumes that tasks like classification and category-based induction require an explanation that involves some setof representations and processes that are shared (to somedegree) across different kinds of categorization tasks but arerelatively distinct from the kinds of representations and processesthat are involved in decision-making memory or attention

Ultimately we need to reorient our theories to find the com-monalities across tasks that are typically thought of as differentIn the study of concepts there are already some hints in theliterature that this reorientation is starting to take place

The most prominent shift in research on categorization comesfrom work relating categorization to memory For example theresearch by Ashby Maddox and colleagues draws parallelsbetween behavioral and neurobiological research on categoriza-tion and memory (Ashby et al 1998 Maddox amp Ashby 2004)This work incorporates research from neural systems involvedin memory to make predictions for performance in categorylearning experiments Research on the kinds of categoriesthat amnesics can learn is also inspired by the desire to createparallels between memory and categorization (Knowlton et al1994)

Research on the influences of learning tasks on category learn-ing also forms parallels between categorization and memory




(Markman amp Ross 2003) A growing body of research suggeststhat the tasks people perform while learning categories influ-ences what people learn about those categories For exampleYamauchi and Markman (Yamauchi amp Markman 1998 Yamauchiet al 2002) found that people tend to learn about features thatdistinguish one category from another when learning by classify-ing examplars In contrast people tend to acquire category pro-totypes when learning by making predictive inferences aboutnew category members that are missing some features (see alsoChin-Parker amp Ross 2002) Markman and Ross (2003) drew onthe memory literature and argued that this type of categoryacquisition involves transfer appropriate processing which isalso observed in memory (eg Morris et al 1977)

The relationship between categorization and memory is alsobeing driven forward by work on the influences of communi-cation on category acquisition Garrod and colleagues (Garrodamp Anderson 1987 Garrod amp Doherty 1994) found that peopletend to arrive at a common method for referring to informationin the environment over the course of conversing with othersabout those objects Markman and Makin (1998) observed thatthe similarity of categories across people is made more uniformby communicating with others The act of establishing acommon reference influences memory by ensuring that categorylabels refer to a common set of properties

These communicative factors have a significant influence onthe ability to extract information about categories frommemory For example Malt et al (2003) found that it is hardto predict the labels people give to a common set of objects(such as jars bottles boxes and containers) based solely on simi-larities among the items Furthermore there is no clear relation-ship between the labels given to objects in different languagesInstead the labels given to objects reflect communicativehistory of those objects in the language The labels peoplelearn for objects then are determined in part by the utility ofthose labels for communication We use a particular word foran object because we know that others will know what we aretalking about when we use that label

I am not suggesting that we abandon theoretical devices likeprototypes exemplars and theories as constructs for helping usto understand categorization behavior Instead the theoreticalbasis of research on categorization needs to be modified

We must recognize that the tasks that we use to study psychol-ogy in the lab do not cut nature at its joints Instead we must lookfor the cross-cutting psychological mechanisms that are involvedin a variety of different tasks That means that theoretical con-structs from memory communication attention and motivationare all relevant to the study of category acquisition

In addition the valuable constructs from research on categor-ization must be exported to the study of other cognitive processeswhere they may be valuable Undoubtedly prototypes exem-plars and theories have value in understanding a variety ofaspects of cognitive processing beyond categorization Forexample exemplars play an increasingly important role in the-ories of automaticity (Logan 1988 2002)

Ultimately we must transition from a cognitive science inwhich psychological tasks organize our understanding of psycho-logical mechanisms to one based on an understanding of the waya core set of psychological mechanisms permit us to performthose tasks

Concepts versus conceptions (again)1

doi101017S0140525X10000440

Georges ReyDepartment of Philosophy University of Maryland College Park MD 20742

georeyearthlinknet

httpsitesgooglecomsitegeorgesrey

Abstract Machery neglects the crucial role of concepts in psychologicalexplanation as well as the efforts of numerous ldquoexternalistsrdquo of the last40 years to provide an account of that role He rightly calls attention tothe wide variation in peoplersquos epistemic relations to concepts ndash peoplersquosconceptions of things ndash but fails to appreciate how externalist andkindred proposals offer the needed stability in concepts themselves thatunderlies that variation

In proposing to dispense with concepts in Doing without ConceptsMachery (2009) neglects to notice how psychological explanationpresupposes them How could we begin even to describe forexample the Muller-Lyer illusion unless people share a conceptof longer than or the gamblerrsquos fallacy without them sharingmore likely Concepts seem to be natural kinds at least to theextent that they are the entities over which psychology generalizes

But what is a concept that can play this role Many recent phi-losophers (other than the atypical Peacocke [1992]) have offeredproposals that Machery amazingly ignores such as the variousforms of ldquoexternalistrdquo views that have been proposed by forexample Kripke (19721980) Putnam (1975) Dretske (1981)Devitt (1981 1996) Millikan (1984) Burge (1986) and Fodor(1990 1998) according to which the identity conditions for aconcept are provided in part by historical and counterfactualrelations the thinker bears to phenomena in the external worldAt most Machery considers some of the ldquointuitionsrdquo that motiv-ate externalism only to dismiss them as too cross-culturallyvariable and unreliable2

Machery correctly notes that a problem with many traditionalphilosophical accounts like Peacockersquos is that they fall afoul ofQuinersquos attack on the analyticsynthetic distinction (Machery2009 p 39) But he misses how that attack threatens anypurely epistemic proposal including his own appeal to ldquobodiesof knowledgerdquo (p 12) This is puzzling given Fodorrsquos extensivediscussion of many of these issues in at least five of his recentbooks a discussion quite often directed explicitly at the psycho-logical theories Machery considers The problem is that ldquobodiesof knowledgerdquo vary between people and stages in a personrsquos lifeUnless one restricts the relevant knowledge in some principledway no two people (or stages) will share a concept since shortof coincidence no two people (or stages) will bring exactly thesame knowledge or procedures to bear in making many of theirjudgments3 Call this the problem of epistemic variabilityRather than inviting us to abandon the notion of conceptperhaps this variability is simply a reason to abandon an episte-mic conception of it

Machery would probably reply that Irsquom merely pressing herethe philosophical notion of concept which is concerned withthe individuation of propositional attitudes (2009 pp 32ndash33) atopic that surprisingly he claims is not the concern of psycho-logical theory (p 34) Indeed he claims that when the twotasks are properly distinguished ldquomost philosophical attacksagainst the psychological theories of concepts are decisivelyunderminedrdquo (p 51)

ldquoDecisivelyrdquo There are a number of important reasons tothink not In the first place as Machery himself notes (2009pp 35ndash37) many psychologists themselves are hardly clearabout the difference between the two concerns often presentingtheir work as refuting the Classical View of traditional philosophyand presupposing some philosophical usually verificationistalternative (see Rey 1983)

Secondly concept identification would seem to be an issue notabout how people do think under pressure but how they couldthink if they were to reflect ndash what they could understand ndashand pace Machery (p 34) this seems as apt a topic as any forpsychological research It may well be that peoplesrsquo prototypeof a doctor is of a man in a white coat but if they found thethought of young woman doctor in a dark one as unintelligibleas a round square that would be a reason to think they didnrsquothave the concept doctor Itrsquos because people have a conceptdoctor that transcends their prototypes that itrsquos worth reasoningwith them that is modifying their epistemic position by citing



evidence or argument but keeping the concept the same As Iemphasized in Rey (1985) we need to distinguish the conceptof something from merely the (epistemic) conceptions of it thathave been too much the focus of the psychological researchMachery reviews Concepts are what remain stable across varia-bility in conceptions and so give argument a point framing thequestions of what people could learn and what might be thelimits of reason and thought

Lastly as Machery notes there is a need of a ldquocoherent frame-workrdquo (2009 p 247) for bringing together the psychologistsrsquo differ-ent proposals Something like the externalist strategies may be justthe sort of thing for the purpose They at least address the problemof epistemic variability in a promising way proposing that conceptsare constituted by what is ldquoexplanatorily basicrdquo4 A crucial feature ofthis strategy is that it makes no commitment to the character of therepresentations people use in ordinary circumstances requiringrapid reasoning or even in ldquoacquiringrdquo the concept (thereby alsoallowing for many concepts to be innate)

Nothing I have said here is meant to suggest that any external-ist (or other) proposals are yet satisfactory (see Segal [2000] andRey [2009a] for serious qualms) Itrsquos not that Machery shouldhave endorsed externalist or related strategies my point issimply that he should have discussed them particularly beforegiving up on the concept of concept entirely

NOTES1 This commentary summarizes my longer review of Mach-

eryrsquos book available at httpndprndedu httpndprndedureviewcfmidfrac1416608 which readers should consult for moredetail It and other relevant material of mine are also availableat httpsitesgooglecomsitegeorgesrey

2 See Devitt (forthcoming) for reasons this dismissal is rashNote that Machery fails to notice that for example Burgersquos(1986) arguments for externalism donrsquot rest on ordinary intuitionsalone but on explanatory features of a Marrian theory of vision

3 Of course there may be similarities and overlap in manypeoplersquos knowledge and procedures (at least relative to the stab-ility of their other concepts) but mere similarity and overlaparenrsquot identity and it is identity in concepts that is needed tosustain serious explanations such as ones about cognitive devel-opment vision or language cf Fodor (1998 pp 30ff)

4 Devitt (1996) and Horwich (1998a p 41) propose thattreating the meaning constitutive conditions as the ones onwhich all other uses of a symbol explanatorily depend can beseen as a mixed internalistexternalist variant of Fodorrsquos (1990)ldquoasymmetric dependencyrdquo that achieves some of its sameeffects but without its strong externalist commitments see Rey(2009a) for discussion

Why donrsquot concepts constitute a natural kind

doi101017S0140525X10000452

Richard Samuels and Michael FerreiraDepartment of Philosophy The Ohio State University Columbus OH 43210-1365

samuels58osuedu

ferreira23osuedu

Abstract Machery argues that concepts do not constitute a natural kindWe argue that this is a mistake When appropriately construed hisdiscussion in fact bolsters the claim that concepts are a natural kind

Introduction A central claim of Macheryrsquos Precis (targetarticle) ndash and of his book Doing without Concepts (Machery2009) ndash is that concepts do not constitute a natural kind Untilreading his work on the topic we would have been inclined toagree But he has changed our minds Macheryrsquos discussionwhen appropriately construed provides grounds to supposethat concepts do constitute a natural kind

What is a natural kind Though the notion of a natural kindhas been characterized many times over philosophers of sciencehave in recent years reached a consensus ndash or as close to consen-sus as philosophers ever get ndash according to which natural kinds areHomeostatic Property Clusters (HPC Boyd 1991 Machery 2005)According to the HPC account a kind K is natural if

H1 It is associated with a contingent property cluster a rangeof characteristics or symptoms which tend to be co-instan-tiated by instances of the kind but need not be genuinenecessary conditions for membershipH2 There is some set of empirically discoverable causal mech-anisms or structures ndash a causal essence ndash that explains the co-variation of these symptomsH3 To the extent that there is any real definition of what it isfor something to be a member of the kind it is not the symp-toms but the presence of the causal essence producing thesymptoms that are definitive of kind membership (Boyd 1989)

This conception of natural kinds is intended to capture the centralfeatures of those kinds paradigmatically studied by science Mostimportantly the fact that the existence of property-clustersdepends on the operation of an underlying causal essence helpsexplain why such kinds are typically subsumed by many non-accidental empirical generalizations (Griffiths 1997 Machery 2005)

Why suppose that concepts constitute a natural kind

Machery endorses the HPC account yet he denies that conceptsconstitute a natural kind We maintain however that if his claimsabout the cognitive scientistrsquos notion of a concept are correctthen concepts are plausibly a natural kind in the HPC senseFirst consider some of the characteristics that according toMachery concepts possess

1 Concepts consist in bodies of information2 Concepts are stored in long-term memory and persist even

when not actively deployed3 Concepts are non-proprietary4 Concepts are default representations automatically

retrieved from long-term memory5 Concepts are subject to temporal and inter-subjective

variation6 Concepts are internally connected7 Concepts are internally coherent

No doubt other generalizations hold of concepts as such Buteven the characteristics 1ndash7 just listed provide grounds tosuppose that concepts are a natural kind If true they suggestthat concepts satisfy the main conditions imposed by the HPCaccount Specifically

C1 Concepts exhibit a reliably covarying property cluster includ-ing internal coherence connectedness persistence and so on

C2 The existence of the property cluster depends largely onthe operation of an abstractly functionally characterizable kindof process Roughly Most of the properties depend on factsabout long-term memory the operations that it performs andits relations to other higher cognitive processes

C3 To the extent that inclusion in the kind concept can bedefined it should be the condition of being subject (in theright kind of way) to the relevant causal mechanism ndash ielong-term memory ndash that defines kind membership The con-dition that Machery labels ldquoDefaultrdquo partially captures this idea

In short Macheryrsquos comments appropriately construedprovide reason to suppose that concepts are a natural kind inthe HPC sense

Why would one think otherwise Clearly Machery is notinclined to accept this conclusion Why not Here are somebrief comments on the most plausible arguments that we couldextract from his discussion

Argument 1 Different kinds ndash for example prototypes andtheories ndash satisfy the conditions on concepthood So while




prototypes and theories might be natural kinds concepts as suchare not

Response There is nothing wrong with superordinate naturalkinds For example metal is plausibly a natural kind eventhough it contains subordinate kinds such as gold and lead Forall the present argument shows the same is true of concepts

Argument 2 Characteristics 1ndash7 are intended to explicate thecognitive scientistrsquos notion of a concept not to characterize thekind as such If so treating them as parts of the kind-syndromeis tantamount to confusing issues about the semantics of theword ldquoconceptrdquo with issues about the nature of the kind

Response Even if 1ndash7 are parts of the cognitive scientistrsquosnotion of a concept there is no inconsistency between thisclaim and the claim that they are also parts of the propertycluster exhibited by members of the kind

Argument 3 Members of a natural kind are supposed to sharemany covarying properties and hence be subsumed by manyempirical generalizations But while concepts share some proper-ties and are subsumed by some empirical generalizations thereare too few to merit natural kind status

Response How many properties must members of a kind sharein order for the kind to be natural This is a silly question becausethere is no reason to suppose any precise cut-off point Presum-ably some kinds manifest more common properties than conceptsdo and in that regard are better examples of natural kindhoodthan concepts But this is a very unsurprising conclusion (Whowould have supposed otherwise) Moreover it surely does notfollow from this that concepts are not a natural kind simpliciterOn the contrary there are two obvious considerations thatexplain the relative lack of properties and generalizations associ-ated with concepts without impugning their natural kind status

1 Concept is a psychological kind and psychological kindsquite generally appear to underwrite fewer empirically rich gen-eralizations than the kinds cited by many other disciplines (Exer-cise Compare chemistry or molecular biology with psychology)

2 Concept is a superordinate kind So in point of logic thekinds it subordinates ndashfor example prototypes and exemplars ndashwill have the characteristics of concepts as such plus their ownspecific characteristics Still it would be wrong to suppose thatonly subordinate kinds are natural while kinds like concept (ormetal or alkali etc) are not

Conclusion Concepts plausibly constitute a natural kind inthe HPC sense So pace Machery we do not advocate their elim-ination from cognitive science Instead we think that Macheryrsquosdiscussion supports a rather more banal conclusion namelyConcepts are in some regards a less good example of naturalkindhood than many other kinds studied by science But this isvery unsurprising and is largely explained by two facts Psychol-ogy appears to generate fewer robust empirical generalizationsthan many other sciences and in point of logic superordinatekinds manifest fewer regularities than the kinds they subordinate

Evidence of coordination as a cure forconcept eliminativism

doi101017S0140525X10000464

Andrea ScarantinoDepartment of Philosophy Georgia State University Atlanta GA 30302

ascarantinogsuedu

Abstract I argue that Machery stacks the deck against hybrid theoriesof concepts by relying on an unduly restrictive understanding ofcoordination between concept parts Once a less restrictive notionof coordination is introduced the empirical case for hybrid theories ofconcepts becomes stronger and the appeal of concept eliminativismweaker

In Doing without Concepts Machery (2009) makes a persuasivecase that there is no unique body of information that playsthe concept-role An important methodological consequencefollows Psychologists of concepts should stop asking whetherthe realizer of the concept-role is a prototype an exemplar ora theory None of these bodies of information alone can explainall phenomena pertaining to higher cognitive competencesWhat I reject is the additional thesis that the term ldquoconceptrdquoshould be eliminated from the vocabulary of psychologybecause it does not designate a natural kind roughly a maximalset of entities about which many scientifically interesting gener-alizations can be formulated

Machery is inspired by Griffithsrsquo (1997) proposal that we splitemotion theory into the study of several heterogeneous kinds ofemotions such as affect programs (eg fear of a suddenlylooming object) higher cognitive emotions (eg guilt abouthaving missed a friendrsquos birthday) and socially sustained pre-tenses (eg going postal after having been fired) Mere evidenceof heterogeneity however isnrsquot a sufficient reason for eliminatinga kind (Piccinini amp Scott 2006) If a set of heterogeneous entitiesis part of a higher-level structure about which scientifically inter-esting generalizations can be issued the heterogeneity of theparts is compatible with the existence of a higher-level naturalkind to which they jointly belong We do not think that ldquoele-phantrdquo does not designate a natural kind just because elephantshave heterogeneous parts these parts are integrated in a waythat allows biologists to formulate many scientifically interestinggeneralizations about elephants

Now in the case of emotions it is quite clear not only thataffect programs higher cognitive emotions and socially sus-tained pretenses are heterogeneous but also that there is nooverarching higher-level entity of which they are parts Severaldistinct causal mechanisms are responsible for the occurrenceand unfolding of emotion episodes of different kinds and thisprevents the emergence of a unified scientific psychology ofemotions

In the case of concepts the situation is considerably murkierbecause we do have at least preliminary evidence that co-referen-tial prototypes exemplars and theories are integrated parts of alarger whole (the Hybrid Hypothesis) To get clear on this topicwe must determine when bodies of information are parts of alarger whole Machery presents two individually necessary andjointly sufficient conditions Connection and Coordination(Machery 2009 p 64)

Connection between bodies of information requires that theretrieval and use of one body of information in a given cognitiveprocess facilitates the retrieval and use of the remaining bodies ofinformation in other cognitive processes Coordination betweenbodies of information requires that they never produce inconsistentoutcomes for example inconsistent categorization judgments

Armed with this understanding of the partndashwhole relationMachery proceeds to present evidence against CoordinationLanguage users he argues can judge that some liquid is waterbecause it fits the water prototype (water is transparent drink-able found in lakes etc) but at the same time that it is notwater because it does not fit the water definition (water isH2O) Since neither judgment is allegedly taken by the folk tobe authoritative over the other Machery concludes that theCoordination condition is violated and that this counts as astrike against the Hybrid Hypothesis

My main problem with this line of reasoning is that the Coordi-nation condition is inadequate In general terms the parts of agiven concept are coordinated just in case they work togetherin ways to be empirically discovered in at least some of theprocesses underlying higher cognitive competences

Macheryrsquos coordination condition offers a very narrowinterpretation of how such bodies of information are supposedto work together namely by avoiding conflict in all circumstancesin which they are jointly activated But the fact that this veryspecific principle of organization is not empirically supported



does not constitute evidence that bodies of information are notcoordinated in some other theoretically interesting way

First there are forms of coordination that have nothing to dowith joint activation For instance bodies of information can beacquired rather than deployed in a coordinated fashionFinding out how they are acquired demands unveiling whatwe may call generalizations of coordinated acquisition Forexample some have argued that the statistical information con-tained in prototypes and the causal functional and nomologicalinformation contained in theories are acquired in part through aprocess of abstraction from exemplars (Heit 1994b) This wouldrepresent an important aspect of integration between bodies ofinformation

A further aspect of integration is that the specific features thatare abstracted in a prototype appear to be heavily influenced bythe subjectrsquos background theory (Wisniewski amp Medin 1994)Some consider the integration between statistical and theoreticalbodies of information to be so tight that they have proposedrepresenting prototypes not as simple feature lists but ratheras schemata which make explicit the theory-based relationsbetween statistically common features (Cohen amp Murphy 1984)

Second there are forms of integrated activation that do not aimto avoid conflict under all circumstances Finding out about suchalternative forms of coordination demands unveiling what we maycall generalizations of coordinated deployment Some haveargued for instance that whether theories are activated or proto-types are activated in a categorization task depends on the percep-tual richness of the input perceptually rich inputs activateprototypes and perceptually poor inputs activate theories(Murphy 2002 p 168) This would be a way in which differentbodies of information can work together towards a cognitiveend by being differentially as opposed to jointly activated

There is also evidence that in some circumstances in whichjointly activated bodies of information lead to conflict the conflictis resolved according to a general trumping principle Forinstance Keil (1989) has argued that conflicts between proto-type-based and theory-based categorizations of biologicalcategories tend to be systematically resolved in favor of thetheory-based categorization (eg a raccoon that is made to lookexactly like a skunk while preserving its internal properties isjudged to be a raccoon even it if fits perfectly the skunk prototype)This particular form of coordination is lost on the other handwhen biological categories are substituted with artifact categories

My central point is that if enough empirical generalizations ofcoordination can be unveiled both of the acquisition and of thedeployment variety a case can be made that ldquoconceptrdquo paceMachery designates a higher-level natural kind for the purposesof scientific psychology Notice that this strategy for preservingthe scientific integrity of the notion of concept differs from thestrategy of unveiling empirical generalizations that are insensitiveto the differences between prototypes exemplars and theories(eg Weiskopf 2009b) Generalizations of coordination areeminently sensitive to the differences between prototypesexemplars and theories but they can provide evidence that theyare components of an integrated higher-level entity

ACKNOWLEDGMENTThanks to Gualtiero Piccinini for helpful comments on a previous draft

Conceptual atomism rethought

doi101017S0140525X10000579

Susan SchneiderDepartment of Philosophy Center for Cognitive Neuroscience Institute for

Research in Cognitive Science University of Pennsylvania Philadelphia

PA 19104-6304

susansdrgmailcom

Abstract Focusing on Macheryrsquos claim that concepts play entirelydifferent roles in philosophy and psychology I explain how one well-known philosophical theory of concepts Conceptual Atomism (CA)when properly understood takes into account both kinds of roles

In Doing without Concepts Machery (2009) argues that conceptsshould be eliminated from psychology He further claims thatpsychological and philosophical theories of concepts talk pasteach other ldquowhen philosophers and psychologists developtheories of concepts they are really theorizing about differentthingsrdquo (p 4) I am not convinced that concepts should beeliminated from psychological theorizing but today I focus onMacheryrsquos claim that concepts play entirely different roles inphilosophy and psychology I shall aim for a partial reconciliationin the context of one well-known philosophical theory of con-cepts Conceptual Atomism (CA) a view pioneered by JerryFodor (see Fodor 1998 Laurence amp Margolis 2002)

According to Machery (2009) philosophical theories aremainly interested in concepts as they figure as constituents inpropositional attitudes and hence find the matter of referencedetermination to be of key import Psychological theories in con-trast focus on topics such as categorization analogical reasoningand induction (p 34) As an example of how divorced philosophi-cal and psychological interests are Machery raises CA

There is little point in blaming some philosophical theories of conceptssuch as Fodorrsquos theory for being unable to explain how we reason howwe categorize how we draw analogies or how we induce (as does egPrinz 2002) For simply a philosophical theory of concepts is not in thebusiness of providing such explanations (p 37)

CA claims that the nature of a concept is determined (or asphilosophers say ldquoindividuatedrdquo) at least in part by the infor-mation that the symbol carries about the world It further holdsthat lexical concepts are primitive being semantically unstruc-tured that is they are not comprised of further concepts(Fodor 1998 Margolis amp Laurence 1999 p 62 Prinz 2002p 89)

Many would agree with Macheryrsquos claim that CA ignores therole that concepts play in thought (ie categorization inductionetc) Still Macheryrsquos conclusion is premature Surprisingly CAhas a neglected resource for capturing the role the conceptplays in thought this is because as wersquoll see according to CAa concept is defined by its symbol type For consider that CAdefines primitive concepts in the following manner

Existence condition A primitive concept exists if and onlyif a primitive symbol in the language of thought (LOT) hasa broad contentIdentity condition Primitive concepts are identical if andonly if they are of the same symbol type and have the samebroad content (Fodor 1998 p 37)

(Where the broad content of a symbol is roughly what thesymbol refers to)

The reader may be surprised that I construe CA as saying thatsymbols individuate concepts But observe that Fodor himselfacknowledges that broad content alone is inadequate for thepurpose of individuating primitive concepts because it fails to dis-tinguish co-referring concepts (eg groundhogwoodchuckCiceroTully) (Fodor 1998 2008) He explains that it is for thisreason that he distinguishes concepts in terms of their mode ofpresentation (MOP) types as well as their broad contents(Fodor 1998 Chs 1 and 2 2008 Ch 3 especially p 70) Andas philosophers know Fodor regards MOPs as being symbolsHence even working within Fodorrsquos original framework CAhas the resources to individuate concepts along two dimensionsa symbolic dimension and a referential one

Now let us see how CArsquos symbolic element captures the rolethe concept plays in thought I have argued that symbols aredefined by the role they play in computation (Schneider 2009)Although Fodor ironically challenges aspects of my viewnotice that even Fodor himself writes that MOPs (ie symbols)




are individuated by their role in mental processing ldquoIf MOPs areboth in the head and functionally individuated then a MOPrsquosidentity can be constituted by what happens when you entertainitrdquo (Fodor 1998 p 20 see also Fodor 2008 p 92)

Piecing these observations together we are now ready for thepayoff When CA is properly understood both psychological andphilosophical interests are brought together into a singularpackage A lexical concept is a semantically unstructuredldquoatomrdquo that is individuated by both its meaning (in particularits broad content) and its symbol type where the symbol inturn is individuated by the computational role that it plays inonersquos cognitive economy (including importantly its role inmental processes such as categorization induction and analogi-cal reasoning) (Schneider 2009 forthcoming) So CA can say thatthe features of concepts that psychologists are traditionally inter-ested in are built into conceptsrsquo very natures For example con-sider the prototype view In the eyes of CA the experimentalresults in the literature on prototypes are indications of featuresof certain symbolsrsquo underlying computational roles and theseroles determine the relevant conceptrsquos natures

CA is ecumenical For now consider the competing theory-theory which suggests that concepts are mini-theories of the cat-egories that they represent encompassing our beliefs abouthidden features underlying causal mechanisms and ontologicaldivisions Advocates of the theory-theory suggest that it capturesexplanatory relations between features while the prototypetheory does not For instance in a well-known criticism of theprototype view children appear to use beliefs about a creatureor thingrsquos underlying essence to override categorization judg-ments based on superficial sensory features (Keil 1989) In theeyes of the conceptual atomist this debate provides insightsregarding conceptsrsquo underlying computational roles But nomatter how the debate plays out conceptsrsquo natures are neverthe-less determined by their broad contents and symbol typesIndeed perhaps some concepts have computational roles thatare explained by the prototype view while others have rolesthat are illuminated by the theories view

So the conceptual atomist who pays attention to the neglectedsymbolic element of concepts can offer a more comprehensivetheory of concepts than is normally supposed CA is ecumenicalenough to incorporate different sorts of concepts (prototypestheories etc) and it also speaks to philosophersrsquo traditionalinterest in reference determination Finally although I havefocused on the LOT approach my remarks can apply to philoso-phical approaches to concepts more generally insofar as theyindividuate concepts by both meaning and computational (orfunctional) role

Banishing the thought

doi101017S0140525X10000476

Nina Strohminger and Bradley W MooreDepartment of Psychology University of Michigan Ann Arbor

MI 48109-1043

humeanumichedu

bradmoorumichedu

Abstract The first seven chapters of Doing without Concepts offer aperfectly reasonable view of current research on concepts The lastchapter on which the central thesis of the book rests provides littleactual evidence that using the term ldquoconceptrdquo impedes scientificprogress It thus fails to demonstrate that this term should beeliminated from the scientific vernacular

The newly minted cognitive psychology student is likely to havebeen taught that there are three major models of categorization ndashprototype exemplar and theory ndash and that they are vying withone another for the title of one true theory The better part of

Doing without Concepts (Machery 2009) is spent revising thisdoctrine rather than being mutually exclusive the overwhelmingevidence suggests that each of these strategies is used for deter-mining category membership depending on context

From these modest conclusions Machery derives a fairlyradical idea What we have been referring to as ldquoconceptual pro-cessingrdquo is actually several distinct processes so distinct that noscientifically meaningful relationship exists between themBecause these processes do not form a natural kind ndash they donot overlap at the cognitive or neural level ndash any further discus-sion of ldquoconceptsrdquo as a unit is both incoherent and detrimental tothe study of these processes Thus Machery advances in the finalchapter a proposal for concept eliminativism where the notionof a concept must be banished from science entirely

One would think that such a proposal requires some kind of evi-dence that these heuristic groupings generally prevent scientificadvancement Curiously Machery provides virtually no supportto this effect He does mention in passing a few instanceswhere scientific taxonomies have undergone restructuring basedon deference to natural kinds (as when chemical classificationsshifted to the periodic table) but this is quite separate fromdemonstrating that theories reflecting natural kinds are requisitefor scientific progress In any case examples of such sea changesin psychology are rare Machery cites ldquomemoryrdquo and ldquoemotionrdquoas two terms which psychologists have successfully done awaywith a claim we find puzzling The idea that emotion containsldquofew scientifically relevant properties that are common to allemotionsrdquo (2009 p 238) is far from the standard view ndash if any-thing it is the division of emotions into ever-smaller categoriesthat meets with resistance (Nesse amp Ellsworth 2009) Memory isan even worse example Machery asserts that the umbrella termldquomemoryrdquo has been tidily replaced with a suite of finer-grainedlabels like ldquoworking memoryrdquo and ldquoexplicit memoryrdquo In fact psy-chologists debate today not only where to carve memory into itsconstituent parts but also whether to disambiguate memory atall In one recent review Jonides and colleagues concluded thatthe longstanding division between long-term and short-termmemory is artificial owing to the fact that the same mechanismsare involved in encoding maintenance and retrieval in bothsystems (Jonides et al 2008) Here is a case where overzealouscompartmentalizing has led scientists to overlook alternate possi-bilities that were evident in the available data

Of course the memory case cannot directly inform whetherldquoconceptrdquo should be discarded in favor of exemplar prototypeand theory theories But if a decades-long distinction betweenshort-term memory and long-term memory has been overstatedthen even the extensive evidence that Machery gathers insupport of dividing concept into subcategories may erode afterfurther study and review

What of those psychological terms which were never divviedup into natural kinds ndash how have they fared It may be instruc-tive to look at research on concepts which by Macheryrsquosaccount has been hobbled with an imperfect taxonomy foryears Fortunately a survey of this research is convenientlylocated in the earlier chapters of his book which is rich withexperiments probing the nature of conceptual cognition Thearms race between camps of what were once considered compet-ing theories appears to have led to a flourishing of work on thistopic If the unnatural kind-ness of ldquoconceptrdquo has hinderedresearch on conceptual processing it certainly doesnrsquot show inMacheryrsquos book

There is another hurdle facing would-be converts to the elim-inativist project which is the delicate matter of what to do oncethe word ldquoconceptrdquo has been cast from the psychological verna-cular Machery is happy to supply us with a more technicallyaccurate replacement ldquobodies of knowledge used by default inthe processes underlying most higher cognitive competenciesrdquo(p 239) It is not clear why Machery endorses this ungainly locu-tion which would seem to retain all the scientific baggage ofldquoconceptrdquo And it is probably not a good sign that Machery



himself can hardly go more than a page without resorting to theforsaken term

Although Machery presents his project as a fundamentally prag-matic one concept eliminativism does not seem particularly prac-tical It requires that we not only convolute our vocabulary butalso renounce our natural inclination to think of these processesas related on a functional level A truly pragmatic revolution canhardly fail to take into account the minds that must work withinthe confines of the new paradigm In this sense concept eliminati-vism and old-fashioned eliminativism (which Machery takes painsto distance himself from) bear quite a bit in common The proposalsto eradicate inaccurate terminology even if technically moreprecise are so unwieldy that they would be unrealistic to adopt

If science were performed by robots eliminating conceptsmight well be a reasonable prescription Of course the degree towhich these rough intuitions hamper scientific progress wouldstill need to be demonstrated In considering proposals such asMacheryrsquos we must weigh the inconvenience of the newer lessfluent way of thinking against the advantages of doing so In thiscase the inelegance of the project is evident and the benefits ofadopting it are as yet untested and unknown We therefore seeno reason to jettison ldquoconceptrdquo from the scientific discourse

Are prototypes and exemplars used in distinctcognitive processes

doi101017S0140525X10000488

James Virtel and Gualtiero PiccininiDepartment of Philosophy University of MissourindashSt Louis St Louis MO

63121-4400

jlv4z3umsledu

piccininigumsledu

wwwumsledupiccininig

Abstract We argue that Machery provides no convincing evidence thatprototypes and exemplars are typically used in distinct cognitiveprocesses This partially undermines the fourth tenet of theHeterogeneity Hypothesis and thus casts doubts on Macheryrsquos way ofsplitting concepts into different kinds Although Machery may be rightthat concepts split into different kinds such kinds may be differentfrom those countenanced by the Heterogeneity Hypothesis

In Doing without Concepts (2009) Macheryrsquos argument that con-cepts split into different kinds is bold and inspiring but not fullypersuasive We will focus on the lack of evidence for the fourthtenet of Macheryrsquos Heterogeneity Hypothesis (HH) accordingto which ldquoprototypes exemplars and theories are typically usedin distinct cognitive processesrdquo (Machery 2009 p 4)

Machery proposes three types of evidence that may supporthis fourth tenet (p 124) If any of the following is shown fortwo kinds of concept then it is likely that the two kinds ofconcept are used in distinct cognitive processes

1 The neural systems implementing the cognitive processesthat use the two kinds of concept are doubly dissociable

2 The cognitive processes that use the two kinds of conceptexhibit a difference in their input-output functions

3 The cognitive processes that use the two kinds of conceptdo so by means of different algorithms

We accept these three criteria with one exception pertaining toCriterion 2 While we agree that a difference in outputs is evi-dence of distinct processes we deny that a difference in theinputs alone is good evidence of distinct processes

Machery maintains that an input difference in categorizationfor example categorizing some items by means of prototypesand other items by means of exemplars is evidence enough forhis fourth tenet (Machery 2009 p 124) Assuming for the sake

of the argument that some items are indeed categorized bymeans of prototypes and others by means of exemplars thisshows only that we possess both prototypes and exemplars anduse both in categorizing Given this evidence it may be that pro-totypes and exemplars are used in distinct processes but it mayalso be that both prototypes and exemplars are used in thesame process Neither possibility is favored by an input differ-ence With this caveat in place we argue that Machery doesnot fulfill any of his three criteria with respect to prototypesand exemplars Some of our considerations go even further andraise doubts about Macheryrsquos splitting of concepts into proto-types exemplars and theories

As to Criterion 1 Machery presents no evidence of doubly dis-sociable neural systems involving prototypes and exemplars Theonly evidence he describes is of a single dissociation in amnesicpatient EP (Machery 2009 p 214) Since EP could not recog-nize previously seen items EP was unable to add new exemplarsto his long-term memory But EP could still correctly categorizesimple dot patterns in a way that suggests he used prototypes(Machery 2009 p 215)

After citing EPrsquos case however Machery cites evidence thatprevious exposure to category members is not necessary toperform well in the dot pattern task used to test EP A similarperformance may be obtained by relying solely on short-termmemory (Palmeri amp Flanery 1999) Thus as Machery pointsout (2009 p 217) EPrsquos performance does not show that EPcategorized by means of prototypes in the absence of exemplars(a single dissociation) And even if that were shown a double dis-sociation would also require an additional case in which exem-plars are used without prototypes

As to Criterion 2 Macheryrsquos best evidence comes fromexperiments in which subjects learn some new categories Aand B and then categorize some new stimuli as either A or B(Malt 1989) Under some conditions subjects appeared to cat-egorize a new stimulus by comparing it to the old stimulus mostclosely resembling it Malt interpreted this as exemplar-basedcategorization Under other conditions subjects appeared tocategorize a new stimulus by determining which features it pos-sessed among those that were typical of a category Malt inter-preted this as prototype-based categorization Based on Maltrsquosexperiments Machery concludes that people categorize someitems using prototypes and others using exemplars and heimplies that the processes involved are distinct (Machery2009 pp 180ndash82) We reject Macheryrsquos conclusion for tworeasons

First as we pointed out earlier input differences alone are notgood evidence of distinct processes Hence experiments such asMaltrsquos do not support the fourth tenet of HH unless there are alsooutput differences Malt (1989) reports no output differencesShe does report a priming effect that occurs only under the alleg-edly exemplar-based strategy But the priming effect changes asubjectrsquos response time not the output

Second Maltrsquos experiments do not even show that at least inthe wild subjects store both prototypes and exemplars properlyso called (ie representations of particular objects) and useboth in categorization The stimuli in Experiments 1ndash3 weredrawings of real animals from different species (see Malt 1989Fig 1) They depict what a typical member of a species lookslike without any features that would distinguish one particularanimal from another Such stimuli are too generic to provide con-vincing evidence that subjects store exemplars properly so calledBy contrast the stimuli in Experiments 4ndash6 were artificial cat-egories (Malt 1989 Figs 2 and 3) whose structure is too differentfrom natural categories to warrant any firm conclusion about theordinary process of categorization Machery himself worries thatldquothese experiments tap into ad hoc strategies only used by sub-jects to deal with abnormal learning and categorization con-ditionsrdquo (Machery 2009 p 183)

As to Criterion 3 Machery points out that while both proto-type-based models and exemplar-based models postulate that




categorization involves a computation of similarity the twoclasses of models are different in one respect Prototype-basedmodels usually employ a linear similarity measure while exem-plar-based models usually employ a nonlinear similaritymeasure This may suggest that prototypes are processed follow-ing an algorithm that uses a linear similarity measure whileexemplars are processed following an algorithm that uses a non-linear similarity measure

But as Machery also points out linear measures of similarityare not required for prototype-based models (2009 p 90) andnonlinear measures of similarity are not required for exemplar-based models (2009 p 98) Thus the use of linear versus non-linear measures does not determine whether an algorithm is pro-totype-based or exemplar-based Therefore there is no clearevidence that prototypes and exemplars are used in processesthat follow different algorithms

In conclusion Machery has provided no convincing evidencethat prototypes and exemplars are typically used in distinct cog-nitive processes This lack of evidence is enough to at least partlyundermine the fourth tenet of HH The considerations we havepresented are part of a larger set of doubts on Macheryrsquos way ofsplitting concepts into prototypes exemplars and theoriesMachery may yet be right that there are different kinds ofconcept but there might be a more fruitful way to split conceptsinto kinds than that postulated by HH

Specifically one of us (Piccinini) has argued that the twomain kinds of concept are implicit concepts and explicit con-cepts Implicit concepts encode information about a categoryin an implicit form that cannot be accessed directly by thelanguage faculty whereas explicit concepts encode informationin an explicit form that can be manipulated by the languagefaculty (Piccinini forthcoming Piccinini amp Scott 2006) Thisproposal fits with and may contribute to explain a wide rangeof evidence about implicit versus explicit cognition (Evans ampFrankish 2009)

ACKNOWLEDGMENTSThanks to Edouard Machery and Andrea Scarantino for helpfulcomments

Doing with development Moving toward acomplete theory of concepts

doi101017S0140525X1000049X

Haley A Vlach Lauren Krogh Emily E Thom and CatherineM SandhoferDepartment of Psychology University of California Los Angeles Los Angeles

CA 90095-1563

haleyvlachuclaedu

laurenkroghgmailcom

emilyt0623uclaedu

sandhofpsychuclaedu

Abstract Machery proposes that the construct of ldquoconceptrdquo detractsfrom research progress However ignoring development also detractsfrom research progress Developmental research has advanced ourunderstanding of how concepts are acquired and thus is essential to acomplete theory We propose a framework that both accounts fordevelopment and holds great promise as a new direction for thinkingabout concepts

In Doing without Concepts Machery (2009) provides a solidargument for how the current construct of ldquoconceptrdquo has led touseless controversies While agreeing that research on conceptsneeds to be refocused we contend that Macheryrsquos proposal isonly a small step towards a new framework for thinking aboutconcepts We suggest that a promising direction for concepts

exists in research that is too often ignored ndash the acquisition for-mation and development of concepts

Unfortunately like many theories in cognitive science andphilosophy Macheryrsquos proposal largely ignores developmentDoing without Concepts avoids issues that are central to atheory of concepts how concepts are acquired how conceptualstructures change across development and how concepts thatare coordinated early in development become uncoordinatedover time In fact Macheryrsquos proposal acknowledges the senti-ment that development is less important than other areas ofresearch on concepts (2009 p 18) This blanket rejection ofdevelopment is erroneous and dangerous ndash identifying themechanisms by which concepts are acquired how knowledgechanges over time and how cognitive processes give rise tosuch changes is essential to our understanding of how conceptsoperate and are organized Because of the focus on such issuesmany developmental researchers have taken a step back fromthe assumption that concepts originate from existing mentalstructures or representations

One way that developmental research has advanced our under-standing of concepts is by introducing the idea that concepts areformed in the moment (eg Gibson 1969 Samuelson et al 2009Smith et al 1999 Spencer amp Schoner 2003 Thelen amp Smith1994) By this account performance on tasks is a reflection ofthe dynamic interaction between the learner and the learningenvironment ndash prior experiences recalled from long-termmemory act together with task dynamics perception andaction to generate behavior For example one study (Samuelsonet al 2009) had children generalize novel nouns for rigid anddeformable objects in two different tasks a forced-choice taskin which children had to choose one of three objects thatshared the same name with an exemplar and a yesno task inwhich children had to respond whether each of the threeobjects shared the same name with an exemplar Childrenrsquos per-formance in the two tasks suggested conflicting conclusions abouthow children organized categories Performance on the forced-choice task suggested that children treat rigid and deformablethings differently when assigning labels Rigid things are namedby similarity in shape whereas deformable things are named bysimilarity in material substance However performance on theyesno task suggested that children did not distinguish betweenrigid and deformable things in naming and generalized namesfor both kinds by shape similarity Given that children were allat the same developmental level and were presented with theexact same stimuli differences in performance between the twotasks were likely not attributable to possessing different objectconcepts Instead childrenrsquos object concepts were formed inthe moment given the demands of the different tasks

Machery notes that the feasibility of concepts being formed inthe moment fails because there is not a significant body ofresearch suggesting a ldquovariability across contexts of the knowl-edge brought to bear on tasksrdquo (2009 p 22) HoweverMachery ignores countless examples of developmental researchdemonstrating great variability in the knowledge that is broughtto bear in a particular task (eg Plumert 2008 Samuelsonet al 2009 Sandhofer amp Doumas 2008 Siegler 1994 Smithet al 1999 Thelen amp Smith 1994 van Geert amp van Dijk 2002Vlach et al 2008) This literature has demonstrated that perform-ance on tasks is flexible (eg Sloutsky amp Fisher 2008) contextdependent (eg Samuelson amp Smith 1998) and altered by see-mingly minor changes in the conditions of the task (eg Sandho-fer amp Doumas 2008 Sandhofer amp Thom 2006 Vlach et al 2008)For example altering the timing of exemplar presentation by amatter of seconds can lead to marked differences in childrenrsquosperformance on a generalization task (eg Vlach et al 2008)

Machery also rejects the idea that concepts are formed in themoment too quickly dismissing the proposal that there are notenduring mental structures for prototypes exemplars and the-ories The book refers to this perspective as the ldquoanti-representa-tionalist argumentrdquo (2009 p 222) However developmental



research has provided substantial evidence suggesting that varia-bility and behavior can be explained without the presence of amental structure or representation (eg Samuelson et al 2009Smith et al 1999 Spencer amp Schoner 2003 Thelen amp Smith1994) For example Piaget (1954) originally attributed errors inthe A-not-B task to infantsrsquo lack of an ldquoobject conceptrdquo (ie anenduring mental structure about the spatial and temporal con-sistencies of objects) Over the last few decades research hasdemonstrated that errors on this task result from factors otherthan a lack of an object concept (eg Smith et al 1999) suchas visual input (eg Butterworth et al 1982) direction of gaze(eg Horobin amp Acredolo 1986) and memory (eg Diamond1985) Moreover performance on this task can be explained bythe interaction of these processes alone (eg Smith et al1999) Altogether this work has provided a compelling argumentfor how concepts and performance are shaped in the momentfrom a confluence of factors rather than being accessed froman enduring mental structure

To conclude thinking about concepts needs a better directionHowever any theory of concepts would be remiss if it did notaccount for development We propose that thinking about con-cepts should be situated within a dynamic framework thatincludes the learner and learning environment Future researchshould explore the mechanisms by which concepts emerge in themoment ndash perception action and memory interacting withproperties of the environment will reveal how this processworks This framework accounts for how concepts arise andchange over time and thus holds great promise as the new direc-tion for thinking about concepts

The theoretical indispensability of concepts

doi101017S0140525X10000506

Daniel A WeiskopfDepartment of Philosophy Georgia State University Atlanta GA 30302

dweiskopfgsuedu

httpwww2gsueduphldaw

Abstract Machery denies the traditional view that concepts areconstituents of thoughts and he more provocatively argues thatconcepts should be eliminated from our best psychological taxonomy Iargue that the constituency view has much to recommend it (and ispresupposed by much of his own theory) and that the evidence givesus grounds for pluralism rather than eliminativism about concepts

What are concepts A long tradition in philosophy and psychol-ogy takes them to be the constituents of thoughts In Doingwithout Concepts Machery rejects this defining them insteadas bodies of knowledge stored in long-term memory and usedby default in a range of higher cognitive processes (Machery2009 p 12)

Macheryrsquos arguments against the constituency view howeverare not compelling He suggests that the notion of a constituent isill-understood (2009 p 26) But he also notes that the languageof thought (LOT) hypothesis (Fodor 1975 2008) gives us onefairly clear sense of what this might mean (Machery 2009p 27) So why shouldnrsquot we adopt precisely this sense Moreoverit is hard to understand many of his own claims about conceptualprocessing without appealing to constituency Prototypes exem-plars and theories are all complex representations that bearstructural relations to their parts over which inferences simi-larity computations and the like might operate He may wishto remain neutral on issues of what the vehicles of thought arebut connectionist and dynamical systems models of thesephenomena are notably unpromising The best candidates forbodies of knowledge that can fill the role he posits are ones orga-nized inter alia by relations of constituency

Constituency also plays a role in psychological explanationNot everything about a category in long-term memory is or canbe accessed in a single task only some packets are extractedand used at once Tokening a complex representation makes itsconstituents available to working memory for processing Andcomplex representations may make greater processing demandsthan simpler ones The notion of a representational constituentis needed in describing what packets are retrieved from thevast reserves of long-term memory and how these copies inworking memory affect task performance

So we can safely embrace the notion that concepts are con-stituents of thoughts But I agree with Machery that these con-stituents are heterogeneous (non-uniform in my terms) Myown provisional list of types of concepts includes prototypesexemplars causal models ideals and norms and some perceptualand linguistic representations This is central to the pluralist viewof concepts I defend (Weiskopf 2009a 2009b) While Macheryand I agree on much of the empirical data we disagree on itsimport Where I see evidence for pluralism he favors eliminati-vism I suggest that we should be optimistic about the study ofconcepts as such

Concepts are a functional kind like most in psychology andneuroscience (Weiskopf forthcoming) Consider how functionalexplanation proceeds We decompose cognitive systems into ahost of nested and interconnected subsystems and populatethem with representations processes and resources such asmemory stores This is obviously true in explaining competenciessuch as visual perception and numerical cognition and it is noless true for concepts Inductive and deductive reasoningdecision making long-term planning theory construction andtesting language use and a host of ldquohigherrdquo capacities requireexplanation and concepts are the representations whateverthey may be that are proprietary to the system that underliesthese capacities

Machery argues that concepts have nothing in commonbeyond this functional description and hence are not aldquonatural kindrdquo in his sense (2009 pp 243ndash44) But functionalkinds are empirically discovered and are posited in order toexplain a (possibly open-ended) range of capacities that creaturespossess And concept possessors are strikingly different fromcreatures lacking concepts They have a cognitive repertoirethat is flexible ndash that is sensitive to but substantially indepen-dent of ongoing perceptual input in terms of both content andprocessing ndash and that displays integration of information freelyacross domains (Weiskopf 2010) Concepts also explain the pro-ductive character of human thought in virtue of being able tocombine open-endedly Machery does not discuss productivitybut it is widely taken to be a central property of conceptualthought and one that separates concepts from other types ofrepresentation

The fact that a separate cognitive system is needed to explainthese capacities is a discovery not an a priori deliverance Other-wise we could have predicted from the armchair that Skinnerianbehaviorism and its modern descendents (eg Brooksianrobotics) were doomed to fail Instead the limits of thesemodels are demonstrated by their failure to capture the relevantphenomena Concepts constitute a kind because positing themgives us the needed explanatory leverage over a wide range ofcreatures and their capacities If we posit them we simul-taneously gain the ability to account for phenomena that wouldotherwise have been inexplicable and to capture similaritiesamong otherwise dissimilar creatures This is how the functionalkinds posited in models of cognitive systems typically earn theirdistinctive status If a model containing a functional category Fhas greater explanatory and unifying power than ones that lackit then F is prima facie a kind The failure of models of humancognition that lack anything corresponding to concepts showsthat they satisfy this condition

But suppose we followed Macheryrsquos lead and eliminated theterm ldquoconceptrdquo talking only of prototypes exemplars and so




on These representations may occur in many cognitive subsys-tems Visual perception may involve generating and storingsuch representations hence we often find talk of ldquoperceptual pro-totypesrdquo in the psychological literature Without distinguishingconcepts as such we would be unable to state the ways inwhich perceptual prototypes differ from their conceptual kinConceptual prototypes as opposed to perceptual ones arecapable of free recombination with other conceptual represen-tations are capable of being generated for non-perceptual cat-egories and so forth Talk of prototypes alone will not do thiscrucial taxonomic work for us For this we need the theoreticalnotion of a concept

Indeed this explanatory need can be seen even in Macheryrsquosown definition of concepts as being involved in ldquohigherrdquo cognitivecapacities For what makes one capacity ldquohigherrdquo than another Atempting answer is that the ldquohigherrdquo ones are just the concept-involving ones The fact that we need to appeal to conceptseven to isolate these various types of representations in the firstplace suggests that concepts will be an essential part of our taxon-omy of psychological kinds Happy news for pluralists but not foreliminativists

Developing without concepts

doi101017S0140525X10000518

Yevdokiya Yermolayeva and David H RakisonDepartment of Psychology Carnegie Mellon University Pittsburgh PA 15213

yyermolaandrewcmuedu

rakisonandrewcmuedu

httpwwwpsycmuedupeopleyermolayevahtml

httpwwwpsycmuedupeoplerakisonhtml

Abstract We evaluate the heterogeneity hypothesis by considering thedevelopmental time course and the mechanism of acquisition ofexemplars prototypes and theories We argue that behavioral andmodeling data point to a sequential emergence of these three types ofconcepts within a single system This suggests that similar or identicalunderlying cognitive processes ndash rather than separate ones ndash underpinrepresentation acquisition

Doing without Concepts (Machery 2009) proposes an interestingsolution to the problem of applying the term ldquoconceptrdquo to proto-types exemplars and theories which according to the authorare unrelated Each type of concept engages a distinct cognitiveprocess ndash such as similarity comparison or causal inference ndash sothat a unified label is inappropriate The book synthesizes animpressive amount of literature in psychology and philosophyto provide evidence for this heterogeneity hypothesis Fromthe point of view of developmental psychology however twokey questions remain unanswered First what is the timecourse for the emergence of prototypes exemplars and theoriesSecond and more importantly what is the mechanism behindtheir formation Specifically does each require a dedicatedmechanism or is a single system sufficient In our view ananswer to the second question is particularly important for ourability to evaluate the proposal that distinct cognitive processesunderlie the use of prototypes exemplars and theories

Answering the first question is an important component toanswering the second question If exemplars prototypes andtheories emerge in succession and not simultaneously then it ispossible that they build upon each other This could suggestthe development of a single mechanism or at the very leastthe development of three related mechanisms While no singlestudy provides definitive evidence a pattern of successive emer-gence can be observed across studies As an example we canexamine infantsrsquo knowledge about individuals Three-month-oldinfants can discriminate an image of their motherrsquos face fromthat of a stranger (Barrera amp Maurer 1981) which suggests

that they have stored an exemplar of their motherrsquos appearanceBy 6 months of age infants can extract a prototype from a seriesof faces and display a preference for a novel face when it is pre-sented with either a familiar face or the previously unseen proto-type (Rubenstein et al 1999) By 10 months infants display moretheory-like knowledge about individuals in that they do not gen-eralize goal-directed actions such as reaching for an object fromone individual to another (Buresh amp Woodward 2007) This taskrequires not only theoretical knowledge about the properties ofgoals but also the ability to store exemplars of the individualsso that goals may be matched correctly Taken together thesestudies provide some support for the sequential emergence ofexemplars prototypes and theories

In addition to determining the time course for these processesthe most important developmental question with respect to theheterogeneity hypothesis relates to the mechanism of acquisitionDo dedicated mechanisms exist for of prototypes exemplars andtheories or are all three acquired within the same system In ourview the literature points to the latter We suggest that if theunderlying mechanism of acquisition is shared then entirely dis-tinct cognitive processes do not underlie different types ofconcepts

According to Machery prototypes and theories have little incommon the former involves extraction of the statistics of a cat-egory while the latter involves causal inference However as hasbeen suggested by Sobel and Kirkham (2007) statistical learningis involved in the emergence of causal reasoning In the back-wards blocking paradigm childrenrsquos and adultsrsquo judgmentsabout an objectrsquos causal effectiveness are influenced by priorknowledge about frequencies with which causal and noncausalobjects are present in the environment Sobel and Kirkhamargue that causal knowledge involves probability distributionsand likelihoods of particular hypotheses ndash to reason causallychildren must have the ability to extract statistical regularitiesfrom the environment The formation of prototypes and theoriesthen relies on processing of statistical information

Work in computational modeling provides additional evidencefor a shared mechanism by demonstrating that reasoning basedon similarities (using exemplars or prototypes) and reasoningbased on theories does not require separate architecturesChaput and Cohen (2001) used hierarchical self-organizingmaps to model changes in infantsrsquo perception of simple collisionevents in which one ball causes another to move Studies haveshown that younger infants respond to such events based on tem-poral or spatial similarity Older infants respond based on causalfeatures of events Noncausal events with a temporal gap in thesequence are perceived to be the same as those with a spatialgap and different from continuous causal events Chaput andCohen (2001) produced a model in which the intermediatelayers responded based on temporal and spatial componentsearly in training much like younger infants As training pro-gressed the top layer integrated these components and beganto respond based on causal information

Similarly Verguts and Fias (2009) used modeling to demon-strate that similarity and rule-based responding can be thoughtof as lying on the same continuum Similarity judgments aremade based on many readily perceivable features rule judg-ments are made based on fewer internally generated featuresThe model replicated human performance on a prediction taskin which participants who received little training used similarityto observed cases to make predictions and those who receivedmore training used rules With an increased number of trainingtrials the model progressed from making similarity judgmentsusing the components of the input to making rule judgmentsby extracting regularities among components Taken togetherwith Chaput and Cohenrsquos (2001) work these findings suggestthat separate mechanisms are not necessary for the emergenceof prototypes exemplars and theories and that theories canemerge through the reorganization of similarity informationwithin the same system



The proposal that exemplars prototypes and theories areunderwritten by distinct processes is a convenient way toaccount for the conflicting psychological data on conceptsHowever an examination of the developmental literature isnecessary for the evaluation of this proposal Behavioral andmodeling studies suggest that exemplars prototypes and the-ories develop sequentially and can do so within the samesystem without the need for three dedicated mechanisms Inour view if the mechanism of acquisition is shared then the cog-nitive processes underlying prototypes exemplars and theoriesmust be partially if not completely overlapping casting doubton the heterogeneity hypothesis From the developmental per-spective the three are not so distinct and ldquodoing without con-ceptsrdquo may be unnecessary

Parsimony and the triple-system model ofconcepts

doi101017S0140525X10000531

Safa Zakia and Joe Cruzb

aDepartment of Psychology and Program in Cognitive Science Williams

College Williamstown MA 01267 bDepartment of Philosophy and Program in

Cognitive Science Williamstown MA 01267

szakiwilliamsedu

httpwwwwilliamseduPsychologyFacultyZakizakihtml

jcruzwilliamsedu

httpwwwwilliamseduphilosophyfourth_layerfaculty_pages

jcruzjcruzhtml

Abstract Macheryrsquos dismissive position on parsimony requires that weexamine especially carefully the data he provides as evidence for hiscomplex triple-system account We use the prototype-exemplar debateas an example of empirical findings which may not in fact support amultiple-systems account We discuss the importance of consideringcomplexity in scientific theory

In Doing without Concepts Machery (2009) postulates a novelmultiple-system theory of concepts a triple-system theoryThat is he proposes a model that has a prototype process atheory process and an exemplar process He cites experimentalevidence in favor of each of these component systems andargues that none of them alone is sufficient to account forextant experimental data In defense of the manifest complexityof the model Machery argues that parsimonious theories areonly to be preferred if they have historically ldquowon outrdquo in a par-ticular domain

[O]ne can take parsimonious theories as more likely to be empiricallysupported than less parsimonious theories if and only if in the past par-simonious theories have been better supported than less parsimonioustheories (Machery 2009 p 126)

In his view the unified parsimonious view of cognition does notpossess this historical advantage so a lack of parsimony is no chal-lenge to his multi-system theory

We find Macheryrsquos treatment of parsimony odd If a single-process theory does a credible job in matching a multiple-systems theory it ought to be favored on the basis of parsimonyMachery however has closed off this appeal His historical pre-sumption makes it impossible for scientifically meritorious simpletheories to triumph if they are late-comers because by then pre-sumably there would exist data sets that at least equally favorcomplex models After all by their nature complex theorieswill be able to accommodate more of these data points thanless complex models Therefore all that the complex modelshave to do is get there first

Of course the serious downside to the ease with which acomplex system can accommodate data is the loss of predictiveability In our view this is the point of privileging parsimony

Evidently Machery maintains that a multiple-systems theory isto be assessed solely in terms of data in its favor regardless ofcomplexity To his credit he canvases an impressive number ofstudies in the field But because there is no presumption infavor of parsimony and therefore in favor of single-systemaccounts we are inclined to look closely at the data that heuses as evidence of the several component processes in categorylearning In at least one area Machery erroneously concludesthat the data favor his triple-system theory over a single-systemaccount

Machery cites a series of articles by Smith and Minda tosupport the idea that exemplars are insufficient (Minda amp Smith2001 Smith amp Minda 1998 Smith et al 1997) In these articlesSmith and Minda presented evidence that they claimed chal-lenged the predictions of the exemplar-based models of classifi-cation and that supported predictions of prototype models Thebasis for these researchersrsquo claims was that the prototypemodels provided better quantitative fits to certain sets of individ-ual subject categorization data than did the context modelHowever Nosofsky and Zaki (2002) subsequently noted that inall of these quantitative-fit comparisons Smith and colleaguesconsidered the predictions from only a constrained version ofthe exemplar model which did not allow the model to capturethe levels of deterministic responding that are often evidencedby individual subjects (eg Maddox amp Ashby 1993) When aresponse-rule parameter was allowed to vary in the model anexemplar model consistently outperformed the prototypemodel Machery does not cite the debate that followed the orig-inal Smith and Minda articles and therefore leaves readers withthe incorrect impression that the exemplar-prototype debatewas settled in favor of a model that required both representations

Similarly in Chapter 6 Machery cites a series of dot-patternstudies (Smith 2002 Smith amp Minda 2001 2002) that heclaims provide support for the existence of a prototype processIn these studies Smith and Minda fit models to data from a par-ticular version of the classic dot-pattern paradigm (Knowlton ampSquire 1993) In the Knowlton-Squire version of the dot-pattern paradigm during a study phase participants are shownhigh-level distortions of a single pattern of nine dots arrangedin a fixed shape In a test that follows participants judge the cat-egory membership of various new dot patterns that were derivedfrom this category prototype but distorted at different levels (iethe prototype itself low-level distortion high-level distortionsand random patterns) The classic result is that the prototypeis classified as a member of the category with the highest prob-ability followed by the low distortions high distortions andrandom patterns Although both exemplar and prototypemodels predict the ordering of this typicality gradient Smithand Minda claimed that only the prototype model couldcapture that steepness of the gradient However Zaki andNosofsky (2004 2007) demonstrated that this steepness was atleast in part attributable to confounded properties of the stimulusset This paradigm was simply not diagnostic in terms of tellingapart the models Macheryrsquos claim that these dot-patternstudies in some way provided evidence for the existence of a pro-totype system and an exemplar system is simply not warranted

We do not claim that a single-system exemplar account of thedata is the correct account of concept learning Indeed the ideathat observers might use different systems to represent categoriesis highly plausible And we note that we have only addressed asmall portion of the data in Macheryrsquos book In evaluating evi-dence we would however prefer to see a more careful treat-ment of the cost of complexity Is a more complex modelwarranted by the data We have no doubt that a triple-processor even a quadruple theory of categorization (see Machery2009 p 118 for a candidate fourth system) would predict avast number of phenomena in the field After a certain pointhowever the important question is what would a complexmodel not predict Many researchers are moving in a directionwhere models are penalized for being more complex (eg




Myung et al 2000 Myung amp Pitt 2009 Navarro et al 2004) bydeveloping sophisticated metrics of complexity that go beyondtraditional approaches of simply penalizing models for additionaldegrees of freedom Machery seems to be pulling us in the otherdirection and we should resist

Authorrsquos Response

The heterogeneity of knowledgerepresentation and the elimination of concept

doi101017S0140525X10000932

Edouard MacheryDepartment of History and Philosophy of Science University of Pittsburgh

Pittsburgh PA 15260

Macherypittedu

wwwpittedumachery

Abstract In this response I begin by defending and clarifying thenotion of concept proposed in Doing without Concepts (Machery2009) against the alternatives proposed by several commentatorsI then discuss whether psychologists and philosophers whotheorize about concepts are talking about distinct phenomena orabout different aspects of the same phenomenon as argued insome commentaries Next I criticize the idea that the cognitive-scientific findings about induction categorization conceptcombination and so on could be explained by positing a singlekind of concept and I insist that many categories (substancestypes of events etc) are represented by distinct coreferentialconcepts that belong to very different kinds of concept This isfollowed by an assessment of the hybrid theories of conceptsoffered by commentators according to which categoriessubstances and types of events are represented by hybridconcepts made of several parts Finally I defend the proposalthat it may be useful to eliminate concept from the theoreticalvocabulary of psychology

R1 Introduction

While writing Doing without Concepts (Machery 2009henceforth DwC) I had several goals One of them wasto clarify the notion of concept used in cognitive scienceand to regiment its use I also wanted to put an end touseless controversies between philosophers and psycholo-gists about what concepts are Even more important Iwanted to make a case for a picture of knowledge rep-resentation that has emerged from 30 years of cognitive-scientific research on the topic According to that picture(which I called the Heterogeneity Hypothesis) cognitivecompetencies are often subserved by several distinct pro-cesses There are many ways to categorize draw induc-tions and so forth These processes are defined overdistinct kinds of concepts which have very little incommon Thus the class of concepts divides into severalheterogeneous kinds Finally I proposed a radicalreform It may be useful for cognitive scientists to elimin-ate the theoretical term concept from their theoreticalvocabulary

Judging by the commentaries elicited by the book andthe Precis it seems that the scientific community in cogni-tive science shares some of these views although few seem

willing to accept the whole picture I proposed In thisresponse I defend this picture I am grateful for thethoughtful and challenging commentaries which havehighlighted some significant limits of the argument madein DwC attracted my attention to some literature I unfor-tunately ignored while writing this book and cast somedoubts on the strength of particular arguments putforward in support of my views

Here is how I proceed in my response After providingsome clarifications in section R2 I turn to the characteriz-ation of the notion of concept used in cognitive science insection R3 and then examine the relation between thephilosophy and the psychology of concepts in section R4In section R5 I respond to the commentaries thatdefend the received view ndash that is the view that there isonly a single kind of concept (eg exemplars or proto-types) In section R6 I examine whether concepts thatbelong to different kinds (specifically prototypes sets ofexemplars and theories) can be coreferential SectionR7 focuses on the view that the class of concepts is evenmore heterogeneous than pictured in DwC In sectionR8 I defend my criticism of hybrid theories of conceptsbefore examining in section R9 three responses to my pro-posal to eliminate concept from the theoretical vocabularyof cognitive science

R2 Clarifications

Before addressing the substantive criticisms made by thecommentators I should briefly clarify a few misunder-standings Hayes amp Kearney misinterpret my concernswith the term concept I do not recommend that conceptbe eliminated from cognitive scientistsrsquo classificatoryscheme because of the context-sensitivity of knowledgeretrieval from long-term memory (By contrast VlachKrogh Thom amp Sandhofer [Vlach et al] recommendeliminating concept on precisely these grounds ndash moreon this in section R3) Indeed I concur with Hayes ampKearney that behavioral and neuroimaging evidenceshows that while knowledge retrieval is indeed context-sensitive some knowledge is also retrieved in a context-insensitive manner (Machery forthcoming see discussionin sect R3) as I explained in two distinct places of DwC(sects 141 and 811) In fact I propose to identify con-cepts with those bodies of knowledge that are retrievedin a context-insensitive manner I also agree with Hayesamp Kearney that models of cognitive processes thatassume the existence of such context-insensitive bodiesof knowledge are compatible with the contextual variationobserved in experimental tasks So what are for me thegrounds for eliminating concept Instead of being con-cerned with flexibility and context-sensitivity I suggesteliminating concept because failing to pick out a naturalkind this term does not fulfill its classificatory function(see sects R91 and R92) and because keeping thisterm would have some important drawbacks and veryfew benefits (see sects R94 and R95)

Vlach et al object to my criticism of the anti-represen-tationalist approaches in cognitive science but they seemto misunderstand the claim made by these approaches(illustrated eg by Rodney Brooksrsquos work and by someversions of the dynamical systems theory) namely thatcognitive science should explain behavior without

ResponseMachery Precis of Doing without Concepts


ascribing to organisms representations of their environ-ment Thus contrary to what Vlach et al seem tobelieve anti-representationalist approaches to cognitivescience do not merely reject the idea of enduring rep-resentations ndash namely representations stored in long-term memory they reject the idea that organisms haveany kind of representations including representationsconstructed on the fly

R3 What are concepts

One of the goals of DwC was to clarify and regiment theslippery notion of concept used in cognitive scienceAccording to the characterization C (introduced anddefended in Chapter 1) a concept of x is a body of knowl-edge about x that is used by default in the processes under-lying our higher cognitive competencies when these resultin judgments about x Blanchard Lalumera Margolisamp Laurence Markman Rey Vlach et al and Weis-kopf raise some questions about this characterization ofthe notion of concept

Lalumera objects that C is an incomplete character-ization of the notion of concept in cognitive science buther own discussion belies this claim Although the notionof a higher cognitive competence was left vague it isclear that the examples of competencies that accordingto her I ignored (viz generalization and induction)straightforwardly fall under this notion Indeed inductionis one of the examples I have repeatedly used to illustratewhat higher cognitive competencies are

Blanchard doubts that the notion of default is anappropriate way of cashing out the notion of conceptused in cognitive science Noting that many experimentsmeant to support the theory theories of concepts are notrun under time pressure he concludes that eithertheory theorists do not view their work as bearing onconcepts ndash which I agree with Blanchard is dubious ndashor the notion of default fails to capture the notion ofconcept they use

Blanchardrsquos objection is an excellent occasion to clarifythe notion of default (see also Machery forthcoming) Onmy view three properties are characteristic of the bodiesof knowledge retrieved by default speed automaticityand context-independence The bodies of knowledgeretrieved by default come to mind quickly their retrievaldoes not depend on onersquos intentional control (althoughthey may also be intentionally retrieved) and they areretrieved in all contexts Of these three propertiescontext-independence is the essential one while speedand automaticity are likely effects of context-indepen-dence Because a body of knowledge is retrieved in acontext-insensitive manner its retrieval from memorymight be simpler and thus faster and it can be automa-tized On my view being retrieved quickly ndash forexample being retrieved under time pressure in an exper-imental context ndash is evidence but not a necessary con-dition for being retrieved by default Thus experimentaltasks that do not involve time pressure can be used toexamine peoplersquos concepts

Rey criticizes my alleged identification of concepts withthe bodies of knowledge used under time pressure (see myresponse to Blanchard above) and he proposes to ident-ify concepts with those bodies of knowledge we would use

if we were to reflect However if we were to followthis proposal psychologists working on conceptswould be unable to explain how people categorizemake inductions understand words and so on inmost circumstances

Like Barsalou (1993) and Prinz (2002) Vlach et alhold that concepts are temporary constructs in workingmemory on the grounds that there is ldquoa great variabilityof the knowledge that is brought to bear in a particulartaskrdquo I concur with Vlach et al that the knowledgebrought to bear on a given task is influenced by contextindeed I explicitly acknowledge this variability in DwCand the useful references they give in their commentaryprovide further evidence of this variability However asshown in DwC and in the Precis several remarkable be-havioral and neuroimaging studies also support the claimthat some bodies of knowledge are retrieved by default(Barsalou 1982 Weisberg et al 2007 Whitney et al1985) Vlach et al do not address this body of evidenceand provide no reason to doubt the conclusion it seemsto support

Other findings provide further evidence for the claimthat some bodies of knowledge are retrieved by defaultAlthough their goal was to highlight the context-sensitivityof knowledge retrieval Hoenig et al (2008) haveshown that brain areas involved in representing actionsor physical manipulations are automatically activated in afeature verification task when the question concerns theappearance of a tool while conversely the brain areasinvolved in representing visual appearances are automati-cally activated when the question concerns the movementassociated with a natural object James and Gauthier(2003) have similarly shown that auditory areas and areasdedicated to the processing of movements areactivated by the mere visual presentation of novelobjects (greebles) when participants have respectivelybeen trained to recognize the sounds of these objectsand their movements In both experiments knowledgethat is not relevant to the tasks at hand is accessed in acontext-insensitive manner

Furthermore the variability of the knowledge used indifferent contexts is consistent with the fact that knowl-edge retrieval from long-term memory is partly context-insensitive Indeed Hayes amp Kearney discuss severalmodels that account for the variability while assumingthat some bodies of knowledge are retrieved by default(see also sect 141 of DwC)

Both Margolis amp Laurence and Weiskopf concur thatdefining concepts as the constituents of thoughts providesa better characterization of the notion of concept than CBefore discussing this proposal note that contrary to whatWeiskopf claims I do not deny that concepts are the con-stituents of thoughts Rather I insist that concepts shouldnot be defined as the constituents of thoughts (sect 143of DwC)

Margolis amp Laurence and Weiskopf propose toclarify the notion of a constituent of thought by appealingto the language-of-thought hypothesis However it wouldbe a mistake to build an empirical hypothesis as controver-sial as this in the very notion of concepts since the validityof the work of cognitive scientists working on concepts(eg their account of the bodies of knowledge retrievedwhen we categorize or draw an induction as well as theiraccounts of the categorization or induction processes)




does not hang on the truth of the language-of-thoughthypothesis

Finally there is little benefit to defining concepts asconstituents of thoughts since this definition does notcast light on cognitive scientistsrsquo research While Margolisamp Laurence mention numerous psychologists who docharacterize concepts as constituents of thoughts (Careyetc) they do not show that this characterization playsany role in these psychologistsrsquo actual empirical work onconcepts

Weiskopf mentions the need to explain the productivityof human thought (see also Hill 2010) This suggests thathe does not simply define concepts as constituents ofthoughts Instead for him a body of knowledge is aconcept only if it can combine freely with any otherbody of knowledge (ie only if it meets Evansrsquos [1982]generality constraint) First it is not clear that productivityis among the explananda of a psychological theory of con-cepts but I will not press this point here (Machery 2010)Rather I propose that Weiskopfrsquos definition is arbitrary Ifconcepts are defined this way organisms that are able tocombine all bodies of knowledge freely (as humans doaccording to Weiskopf) have concepts whereas organismsthat are able to combine many such bodies but not all donot even if the bodies of knowledge in the former andlatter kinds of organism are very similar

Here is another way of illustrating the arbitrariness ofWeiskopfrsquos proposal Suppose we humans have somebodies of knowledge that are used in categorization induc-tion and the like in almost the same way as our otherbodies of knowledge For example they could beprototypes and they could be used in similarity-basedprocesses But while the other bodies of knowledge canbe combined freely these prototypes can only be com-bined with a limited number of other prototypes(perhaps because they belong to a modular cognitivesystem) Weiskopf would conclude that these prototypesare not concepts since they cannot combine freely withall the other bodies of knowledge but this is the wrongconclusion to draw Instead the similarities betweenthese prototypes and the other bodies of knowledgeshould lead us to conclude that they are concepts and toreject Weiskopfrsquos proposal that a body of knowledgecounts as a concept only if it can combine with everyother body of knowledge

Rey argues that my characterization of concept entailsthat two individuals or even a single individual at twodifferent times cannot have the same concept Howeverit is incorrect that a single individual at two differenttimes cannot have the same concept since a givenconcept can remain the same although its parts (viz theelements of knowledge that are by default retrieved fromlong-term memory) change exactly as objects remain thesame despite their parts changing (Machery 2010) It istrue that given my characterization of concept differentindividuals are likely to have different concepts (egdifferent concepts of dogs) but I think that this is a clearvirtue since this explains why they categorize differentlyand make different inductions

Finally Markman argues that it is a mistake to charac-terize psychological constructs such as the notion ofconcept by means of particular cognitive competenciesand experimental tasks since adequate psychological con-structs should explain performance in a range of tasks

and he regrets that I did not pay attention to a broaderrange of competences and tasks I agree with Markmanrsquosargument Although DwC examined only three cognitivecompetencies and focused on a few psychological tasksthe proposed characterization of concept ndash namely C ndashis consistent with Markmanrsquos point

R4 Philosophers and psychologists on concepts

Another goal of DwC was to show that although philoso-phers and psychologists use the term concept and developtheories of concepts psychological and philosophical the-ories of concepts are really about different things As aresult many criticisms of psychological theories of con-cepts by philosophers (and vice versa) are emptyEdwards Margolis amp Laurence Rey and Schneiderdisagree with this claim Keil also seems to assume thatphilosophical and psychological theories of concepts areabout the same thing but does not press this point

Edwards and Margolis amp Laurence are right to claimthat the fact that psychologists and philosophers havedifferent explanatory interests does not entail that theyare theorizing about different things because they couldbe theorizing about different aspects of the samephenomenon

I deny the univocity of concept across philosophy andpsychology on the following grounds Generally in philos-ophy of science a candidate explication of a scientific term(eg ether force) is taken to be a failure if it entails thatwhat scientists say when they use this term is typically mis-taken This is in fact a commonsense idea If my interpret-ation of what someone means by a particular word entailsthat most what she says when she uses this word is falsemy interpretation is probably mistaken Applied to thenotion of concept in cognitive science a satisfactory expli-cation of this notion should not entail that psychologistsrsquoclaims about concepts are literally false C but not Mar-golis amp Laurencersquos characterization meets this con-straint (A similar objection applies to Reyrsquos andEdwardsrsquos characterizations) Since Margolis amp Laurencepropose that the bodies of knowledge psychologists havebeen focusing on (viz prototypes exemplars etc)belong to the processing structure of concepts (whateverthat is) they are bound to claim that psychologists are lit-erally mistaken when they say that concepts are prototypesor exemplars (etc)

Margolis amp Laurence take the mutual influence ofphilosophers and psychologists to be evidence that psy-chologists and philosophers are talking about the samething when they theorize about concepts but I am notswayed by this argument First I am more impressed byhow often psychologists and philosophers talk past eachother when they exchange arguments about conceptsthan by how useful these exchanges have been Secondthe extent to which philosophers and psychologists havefruitfully influenced each other is perfectly consistentwith the idea that they do not theorize about the samething when theorizing about concepts Cell biologistsand physicists working on quantum mechanics do not the-orize about the same thing but the former are usefullyappealing to the theories and findings of the latter (egCollini et al 2010)



Because I hold that philosophers and psychologists whotheorize about concepts are not talking about the samething I find Edwardsrsquos and Reyrsquos critiques puzzlingClearly it is important to understand better how conceptsrefer but this is not one of the explananda of the theoriesdeveloped by psychologists theorizing about conceptsFurthermore as argued in section 24 of DwC the kindof arguments exchanged by philosophers which oftenrely on intuitions about what concept an individual in aparticular circumstance possesses or fails to possess orabout whether two individuals have the same belief (egBurge 1979) seems unlikely to be appropriate for deter-mining how concepts refer

Schneider notes that psychologistsrsquo findings about howwe categorize draw inductions and so forth can be co-opted by some philosophical theories of concepts ndashthose that individuate concepts by means of the functionalrole of mental symbols While correct this observationdoes not undermine the claim that psychologists and phi-losophers tend to theorize about different things whenthey theorize about concepts Similarly while cell biol-ogists use biochemistsrsquo findings biochemists and cell biol-ogists are not developing theories about the sameprocesses

R5 Rejecting the heterogeneity of the class ofconcepts

To my surprise few commentaries objected to the claimsthat there are several kinds of concepts and that proto-types exemplars and theories are among these kindsStrohminger amp Moore and Rey seem even to havefound this claim entirely unsurprising However I worrythat this lack of resistance is due to a confusion betweentwo different claims (1) that prototypes exemplars andtheories are three distinct kinds of concept and (2) thatour long-term memory includes prototypes exemplarsand theories These two claims are not identical sinceone could grant that we have say exemplars and theoriesbut insist that concepts are prototypes (eg Hampton2001) or that we have prototypes and exemplars butinsist that concepts are really theories While Claim 2 isindeed not very controversial Claim 1 is less obviouslycorrect

Hayes amp Kearney Harnad and Hampton are theonly ones to raise doubts about the heterogeneity of con-cepts Harnad rejects the idea that the bodies of knowl-edge we use in higher cognitive tasks are prototypes orexemplars Rather sensorimotor processes play a centralrole in cognition They enable living creatures to navigatethe world by making object identification and appropriateaction possible There is no doubt that such sensorimotorprocesses exist since recognition must involve them butthese processes cannot underwrite the bulk of cognitionLike behaviorists and like some roboticists (see Machery2006b on Harvey) Harnad underestimates the complexityof higher cognition when he proposes that the processesthat can explain perceptual and motor processing canscale up to higher cognition Appealing to words asHarnad does does not help much since words have tobe understood and mapped onto bodies of knowledge inlong-term memory

Hampton contends that the distinction between exem-plars and prototypes breaks down for at least some cat-egories since the exemplars of superordinate categoriessuch as the category of vehicles are prototypes of subordi-nate categories such as the category of cars (see also Maltforthcoming) However superordinate categories could berepresented by representations of particular cars planesbikes and so on rather than by prototypes of cars plansbikes et cetera This is naturally an empirical questionand Hampton might well be right Supposing he is rightit would nonetheless be a mistake to say as he does thatthe exemplars of superordinate categories are prototypesof the subordinate categories Rather using the termexemplar to refer only to representations of individualsone should say that superordinate categories are rep-resented by sets of prototypes of subordinate categories

In any case Hamptonrsquos interesting comment brings tothe fore a shortcoming of DwC ndash the neglect of superordi-nate concepts and the focus on basic-level categories (egdogs) It might be that the heterogeneity of conceptualrepresentations is even larger than proposed in DwCwith superordinate categories being represented differ-ently from basic-level and subordinate categories Impor-tantly far from undermining the crucial message of thebook this outcome would reinforce it It is hopeless tolook for a theory of conceptual representations thatapplies to all default bodies of knowledge

Hayes amp Kearney defend the received view Theycontend that exemplar models of a range of phenomenaoutperform prototypes models (see also Zaki amp Cruz)and that the role of theories can naturally be included inexemplar models (on this latter point see also Yermo-layeva amp Rakison) But some of the studies they them-selves cite in fact undermine this claim Far fromproviding evidence that categorization can be explainedby means of a single kind of default bodies of knowledgeAllen and Brooks (1991) provided evidence that in at leastsome circumstances we have two distinct representationsof a single category (a rule and a set of exemplars) and thatthese representations can lead to conflicting categorizationjudgments Smith et al (1998) have replicated these find-ings and provided evidence that two neural networks areinvolved in each categorization judgment Furthermoreexemplar models seem able to explain the empirical find-ings about concepts only if one takes exclusively into con-sideration the category-learning studies that involveartificial stimuli such as patterns of points Research onthe knowledge of expert physicians (reviewed in Normanet al 2006) shows for instance that during their trainingin medical school physicians acquire different types ofbodies of knowledge that are largely independent fromone another

Further for Hayes amp Kearney the role of theories incognition is indirect They influence which exemplars arelearned in category-learning tasks or which exemplars areretrieved from long-term memory (which explainsperhaps why Hayes amp Kearney prefer to speak of priorknowledge instead of theories) However a less partialreview of the literature suggests that the use of theories isnot so limited and that causal theories are directly used tocategorize and to make inductions (Chs 6 and 7 of DwC)

Finally in DwC I argued that simplicity can be used tochoose between scientific hypotheses in a domain ofinquiry (eg in psychology) only if past evidence




inductively supports the belief that in this domain simplerhypotheses tend to be better supported than morecomplex hypotheses By contrast Zaki amp Cruz arguethat simplicity should always be preferred on the groundsthat models with more free parameters fit better a givenset of data points than models with less free parametersand they conclude that everything else being equal weshould prefer a theory that posits a single kind of conceptto a theory that posits several distinct kinds of conceptAlthough the question cannot be resolved in a few wordstheir argument should be resisted for two reasons Firstit is not necessarily the case that simpler models fit betterthan more complex ones when fit is evaluated by cross-vali-dation since models with more parameters can overfitSecond model fitting provides a poor albeit common(eg Forster amp Sober 1994) analogy for understandingthe use of simplicity as a criterion for theory choice Morecomplex theories are not necessarily better supportedthan simpler theories since they often have empirical impli-cations that simpler theories simply do not have Forinstance the heterogeneity hypothesis ndash but not (eg) pro-totypes theories ndash predicts that in at least some casespeoplersquos categorization (or induction) judgments aregoing to be slower or less reliable because for examplethe prototype-based and theory-based categorization (orinduction) processes conflict with one another (for consist-ent evidence see Allen amp Brooks 1991 Kulatanga-Moruziet al 2001 Regehr et al 1994 Smith et al 1998 and seesects 515 66 and 715 of DwC)

R6 Can prototypes exemplars and theories becoreferential

While I proposed that we often have several coreferentialconcepts ndash for example we might have a prototype ofdogs a set of exemplars about dogs and a theory ofdogs ndash Hampton Jacobson and Margolis amp Laurencedoubt that prototypes exemplars and theories can co-refer

In DwC I intentionally refrained from proposing atheory of how bodies of knowledge such as prototypessets of exemplars and theories denote (see Edwardsrsquosand Reyrsquos commentaries) and this is not the place topropose one However for present purposes it is suffi-cient to note that according to several influential theoriesof reference a prototype a set of exemplars and a theorycan be coreferential Consider for instance Fodorrsquos infor-mational semantics (eg Fodor 1990) According to thisview roughly a concept refers to the property that it isnomologically linked to As argued convincingly by Prinz(2002) informational semantics can naturally be com-bined with prototype theories A prototype of dogs refersto dogs because its occurrence (ie its retrieval fromlong-term memory) is nomologically linked to the pres-ence of dogs in the cognizerrsquos environment Informationalsemantics can similarly be combined with theory theoriesand with exemplar theories Thus it is perfectly possiblefor a prototype a set of exemplars and a theory to becoreferential

Jacobson and Margolis amp Laurence assert mista-kenly that a prototype a set of exemplars and a theorycannot be coreferential because they fail to distinguishreference and categorization (on this distinction see

sect 222 of DwC) The fact that objects can be mis-categorized (eg a wolf can be misclassified as a dog)shows that reference differs from categorizationIndeed categorization judgments could not be mistakenif every object that is categorized as an x (eg as a dog)really belonged to the extension of the concept of x (egthe concept of dog) Because reference differs from cat-egorization the fact that prototype-based and theory-based categorization processes can occasionally classifysome objects differently (see the examples in sect 33of DwC and in Machery amp Seppala forthcoming) doesnot entail that a prototype and a theory cannot becoreferential

R7 Should the heterogeneity hypothesis bebroadened

Couchman Boomer Coutinho amp Smith (Couchmanet al) Dove and Khemlani amp Goodwin argue thatthe heterogeneity hypothesis fails to capture the extentto which concepts form an heterogeneous kind and theypropose to extend the idea that concepts divide into verydifferent kinds in two distinct directions I am sympatheticto this kind of proposal One of the important questions tobe addressed by future research on knowledge represen-tation is whether prototypes exemplars and theoriesexhaust the fundamental kinds of default bodies of knowl-edge (see the conclusion of DwC)

Dove proposes that in addition to distinguishingprototypes exemplars and theories it is necessary todistinguish at least two types of format amodal andmodal (the Representational Heterogeneity Hypothesis)Contra Barsalou Prinz and others Dove agrees with methat not all concepts have a perceptual format (Dove2009 Machery 2006c 2007 forthcoming) Howeverfollowing the dual-coding tradition (Paivio 1991) he alsoholds that some concepts have such a format While theevidence reviewed by Dove is surely suggestive I remainto be convinced It is again important to keep in mindthe distinction (which is apparently not challenged byDove) between concepts and the knowledge used inhigher cognition Concepts are just a subset of the knowl-edge used in higher cognition (see sect 11 of DwC andsect R3 here for a discussion of how to draw the distinc-tion) There is no doubt that we use perceptual represen-tations to solve some tasks and it is plausible indeed asDove argues that dedicated cognitive systems areused for this purpose However this does not entailthat these representations are concepts since they mightonly be used in particular circumstances in a context-sensitive manner

Khemlani amp Goodwin propose to add two other kindsto the fundamental kinds of concepts rules and genericrepresentations Goodwinrsquos work about conceptual illu-sions provides some striking evidence that people areable to learn rules that determine category membershipHowever as I explained in section 414 of DwC theconcern with the rule-based approach to concepts is notthat people are unable to learn and apply rules butrather that natural categories outside of contrived labora-tory conditions do not have the definitions that rule-basedaccounts assume In contrast to rules it is extremely plaus-ible that people store some knowledge about generics as



Prasadarsquos work (among others) suggests In DwC follow-ing many psychologists I took knowledge of genericsand knowledge of causal relations to be constitutive of the-ories but unfortunately I did not defend this proposalKhemlani amp Goodwin want to distinguish representationsof generic information from theories apparently on thegrounds that generic knowledge cannot be identifiedwith causal knowledge However because generic knowl-edge and causal knowledge about some xrsquos (eg dogs) areplausibly intertwined it would be a mistake to hold thatgeneric knowledge and theories form distinct kinds of con-cepts In the terminology of DwC (sect 331 see also sectR8 here) they seem to be linked (or connected) and coor-dinated As a consequence they should be thought of asparts of the same concept our theory of dogs ratherthan as two distinct concepts In this sense theories aretrue hybrids They are made of distinct types of knowl-edge perhaps used in distinct processes that are linkedand coordinated

R8 Hybrid theories

Proponents of hybrid theories of concepts agree withsome but not all tenets of the heterogeneity hypothesisThey propose that the heterogeneous bodies of knowledgeabout a given category (eg dogs) are not distinct con-cepts but rather the parts of a single concept Keil Gon-nerman amp Weinberg Margolis amp Laurence andScarantino defend hybrid theories against the attackmounted in section 33 of DwC As I understood thenotion of part if two bodies of knowledge A and B areparts of the same concept using A enables the use of B(A and B are linked or connected) and A and B mustnot lead to incompatible judgments that are taken to beequally authoritative (A and B are coordinated) Keil Gon-nerman amp Weinberg Margolis amp Laurence and Scaran-tino all reject Coordination as a necessary condition fortwo bodies of knowledge to be parts of the sameconcept Although Coordination may finally turn out tobe an inappropriate way to characterize the notion of con-ceptual parthood I will defend it here Note that ifCoordination is rejected an alternative characterizationof the notion of conceptual parthood should be provided(which Keil fails to do)

Before discussing their key arguments let me clarifyCoordination a bit Coordination does not state that theparts of a hybrid concept cannot underlie incompatiblejudgments This would be an inaccurate way of cashingout the notion of a part of a concept since some well-known hybrid theories of concepts (eg Osherson ampSmith 1981) assume that the judgments underwritten byprototypical information can be defeated by (and canthus be incompatible with) the judgments underwrittenby theoretical or definitional information What Coordi-nation excludes is that the parts of a concept give rise toincompatible judgments that are taken to be equallyauthoritative

Keil and Gonnerman amp Weinberg propose twodifferent arguments purporting to show that Coordinationcannot be a necessary condition for distinct bodies ofknowledge to count as parts of the same concept Keilrightly notes that predicates such as ldquotallrdquo can lead toapparent contradictions Someone can be tall with

respect to some standard and not tall with respect toanother standard He infers that if Coordination were anecessary condition for conceptual parthood we wouldhave to conclude absurdly that all these judgmentsinvolve distinct concepts However what is going on inthese cases is clearly quite different from what is goingon when people both agree and disagree with ldquoTomatoesare vegetablesrdquo In the former case when one bothagree and disagree with the claim that say John is tallit is not because ldquotallrdquo is ambiguous The meaning ofldquotallrdquo is the same in ldquoJohn is tallrdquo and ldquoJohn is not tallrdquoInstead one can both agree and disagree with the claimthat John is tall because two distinct standards areapplied (compared with Tom Cruise John is tall but com-pared with Shaquille OrsquoNeal John is not tall) By contrastone can both agree and disagree with the claim thattomatoes are vegetables because ldquovegetablerdquo is ambiguousThe meaning of ldquovegetablerdquo (ie the concept this wordexpresses) is not the same in ldquoTomatoes are vegetablesrdquoand in ldquoTomatoes are not vegetablesrdquo

Gonnerman amp Weinberg note insightfully that thedistinct exemplars of for example dogs (which accordingto me form a concept of dogs) seem to violate Coordi-nation which shows that Coordination cannot be a necess-ary condition for distinguishing concepts Exemplars ofsay dogs can indeed underlie incompatible judgments ndashsuch as inconsistent categorization judgments ndash sincedifferent exemplars might be retrieved from long-termmemory in different circumstances but it does notfollow that these judgments are taken to be equally author-itative (although to my knowledge exemplar theoristshave said little about this question) Perhaps the judgmentunderwritten by the larger number of exemplars is takento defeat the judgment underwritten by the smallernumber of exemplars Although evidence is lacking toevaluate this proposal it might thus be that exemplarsdo not violate Coordination

Scarantino proposes to replace Coordination withweaker conditions which are satisfied by prototypesexemplars and theories and concludes that these countas parts of concepts rather than as distinct concepts Themain problem with Scarantinorsquos conditions is that theyare neither sufficient nor necessary for two bodies ofknowledge to be parts of the same concept Considerfor instance his first proposal Two bodies of knowledgeare coordinated when the knowledge stored in one ofthem influences the acquisition of the knowledge storedin the other A first problem with this proposal is that gen-uinely distinct concepts meet this condition For instancewhen one forms a body of knowledge about a new animalspecies our theoretical body of knowledge about animalsin general is likely to influence this acquisition processA second problem is that the parts of a single conceptneed not meet this condition For instance the elementsof knowledge about the typical properties of dogs areparts of the same prototype of dogs but my knowledgeabout a typical property of dogs needs not influence theacquisition of my knowledge about another typicalproperty

Finally Keil argues that hybrid theories of concepts canaccount for the findings that I argued undermine them ndashnamely the fact that people are willing to endorse appar-ent contradictions such as ldquoTomatoes are vegetablesrdquo andldquoTomatoes are not vegetablesrdquo particularly when such




sentences are prefaced with hedges such as ldquoin a senserdquo(see Machery amp Seppala [forthcoming] for some relevantfindings and discussion) There are two issues that needto be distinguished here First would people be willingto endorse apparent contradictions such as ldquoTomatoesare vegetablesrdquo and ldquoTomatoes are not vegetablesrdquo ifthey were not prefaced by hedges such as ldquoin a senserdquo Isuspect that this is the case but there is no clear evidencefor this claim Second supposing that people would agreethat tomatoes are vegetables and that they are not veg-etables even without such edges hybrid models wouldthen be compatible with peoplersquos judgments only ifCoordination were rejected But if hybrid theoristsreject Coordination they then need to explain why distinctbodies of knowledge about say tomatoes are parts of thesame concept of tomato instead of being distinct conceptsof tomato

R9 Eliminativism

Most commentators reject the eliminativist conclusion putforward in DwC even when they agree with the existenceof three different types of concepts

R91 Are there generalizations about concepts

A key step in the eliminativist argument proposed in DwCconsists in denying that the class of default bodies ofknowledge forms a natural kind on the grounds that fewscientifically interesting generalizations are true of thisclass Blanchard Danks Lombrozo Virtel amp Picci-nini and Yermolayeva amp Rakison challenge this claim

Inspired by her fascinating work on explanation Lom-brozo proposes that prototypes exemplars and theories(together with perhaps a host of other kinds of knowl-edge) are used to explain While she takes this finding tounify concepts I disagree Ribosomes and transfer RNAare both involved in the production of proteins out ofamino acids but it does not follow that they form asingle kind In fact I propose that a different conclusionfollows from Lombrozorsquos work This work suggests that itmight not be possible to characterize the notion oftheory by means of the notion of explanation as I did inDwC (following many psychologists) since being used toexplain is not a distinctive property of theories Just likeinduction categorization or concept combination expla-nation might be one of the cognitive competencies thatare subserved by distinct processes defined over differentkinds of default body of knowledge

Blanchard Virtel amp Piccinini and Yermolayeva ampRakison challenge the claim that prototypes exemplarsand theories are used in distinct processes (eg distinctcategorization processes) If they are right then general-izations about cognitive processes are true of all con-cepts and the class of concepts is a genuine naturalkind

Blanchard notes that some evidence for the existenceof theories (Luhmann et al 2006) is compatible with the-ories being used in similarity-based processes just likeexemplars and prototypes (see also Hayes amp Kearneyand Yermolayeva amp Rakison) However research oninduction (reviewed in sect 71 of DwC) shows that theprocess underlying theory-based induction and the

processes using prototypes and exemplars differ Thelatter are similarity-based ndash representations are comparedwith one another and their match is evaluated by somesimilarity measure ndash while phenomena like the causalasymmetry effect show that the inductive processes usingcausal knowledge are not based on similarity Virtel ampPiccinini hold that there is no evidence that exemplarsand prototypes are used in different kinds of cognitiveprocess However even if exemplars and prototypeswere used in the same kind of process it would still bethe case that theories are used in a different kind ofprocess and thus that there are no generalizations abouthow concepts are used in cognitive processes

Yermolayeva amp Rakison rightly bemoan the fact thatDwC paid little attention to developmental psychologyincluding the acquisition pattern of prototypes exem-plars and theories (but see the brief discussion ofSmith amp Minda [1998] in sect 644 of the book) andfurther work on the developmental questions raised intheir commentary is called for However I find the pro-posed developmental sequence unconvincing First Iobject to the way exemplar is used in Yermolayeva ampRakisonrsquos commentary An exemplar is not just any rep-resentation of an individual it is a representation that isused by default in higher cognition Thus the fact thatbabies acquire representations of individuals (eg oftheir mother) early does not entail that they acquiregenuine exemplars Such representations are exemplarsonly if they can be used for example to categorize inaddition to identify the individuals they are aboutSecond in some category-learning experiments withadult participants prototypes seem to be acquiredbefore exemplars (Smith amp Minda 1998) Research onmedical expert knowledge also suggests that causal the-ories are acquired at the beginning of physiciansrsquo train-ing before physicians acquire any knowledge ofparticular cases (in the second half of their training)and form prototypes (for review see Norman et al2006) It would thus seem that there is no necessaryacquisition sequence which casts doubts on the ideathat a single process underlies the acquisition of proto-types exemplars and theories

Danks puts forward a distinct criticism based on hisfinding that prototype- exemplar- and theory-basedformal models of categorization can be seen as distinctgraphical models (Danks 2007) However the fact thatdifferent formal models are instances of a more abstractformalism does not entail that the processes describedby these models form a unified kind Lokta-Volterraequations in ecology Hodgkin and Huxleyrsquos model ofthe action potential in neuroscience and the Cagan mon-etary model in economics are all differential equationsWould Danks conclude that they form a unified kind ofprocess that is the object of a unified empirical theory

R92 Do concepts form a natural kind

Samuels amp Ferreira acknowledge perhaps for the sakeof the argument that the class of bodies of knowledgeused by default in higher cognition includes severalkinds that have little in common (viz prototypes exem-plars and theories) but they insist that the bodies ofknowledge used by default in higher cognition are anatural kind on the grounds that they form a genuine



homeostatic property cluster kind However being ahomeostatic property cluster kind is not sufficient forbeing a natural kind Natural kinds are those kinds thatsupport a large number of scientifically relevant induc-tions Because few generalizations are true of them thebodies of knowledge used by default in higher cognitiondo not form a natural kind

Samuels amp Ferreira rightly note that there is no precisecutting point for distinguishing those kinds that supportrespectively few and many generalizations Howeverthose kinds that support many generalizations are differentfrom those kinds that support only few generalizationsexactly as white differs from black even if there is noprecise cutting point when one moves from white to blackthrough gray (Machery 2005) The bodies of knowledgeused by default in higher cognition are an instance of thelatter type of kinds and for this reason are not a natural kind

R93 Concepts as a functional kind

While Couchman et al Khemlani amp Goodwin Lalu-mera and Weiskopf acknowledge that prototypes exem-plars and theories form distinct kinds of concept and thatthere are few generalizations true of all of them theyreject the conclusion that the notion of concept shouldbe eliminated from the theoretical vocabulary of psychol-ogy on the grounds that concepts form a functional kindWhile concepts are indeed a functional kind ndash my ownexplication of the notion of concept C is functional ndashthis alone does not settle the issue of whether conceptshould be eliminated

Being a functional kind is not sufficient for earning onersquoskeep in a scientific classification Some functional kindshave their place and others do not Functional kindshave their place in scientific classificatory schemes eitherbecause they are natural kinds or because denoting themfulfills some useful function and does not have any draw-back So what needs to be shown is either that many gen-eralizations are true of the bodies of knowledge used bydefault in higher cognition or that denoting this classfulfills some useful function in cognitive science

R94 Is the notion of concept useful for cognitivescientists

Hampton Lalumera Strohminger amp Moore andWeiskopf contend that the notion of concept has auseful role to play in cognitive science and that as aresult eliminating it would be detrimental

Hampton contends that the term concept is useful forbringing attention to the way prototypes exemplars andtheories are organized but I do not see exactly why thisterm would be needed It seems straightforward to askwhether prototypes exemplars and theories are coordi-nated whether they acquired in any particular develop-mental sequence and so on

Lalumera argues that we need the notion of concept toexplain why our representation of say dogs changes overtime We start with exemplars of particular dogs thendevelop a prototype and finally build a theory Thiswonrsquot do however for ndash as I have argued at length ndashthere is no such thing as our representation of dogsRather we simultaneously have several distinct represen-tations of dogs and Lalumera provides no reason to doubt

this claim Furthermore as noted earlier in my reply toYermolayeva amp Rakison (sect R91) it is not the casethat prototypes exemplars and theories are necessarilyacquired in any particular order

Weiskopf provides a different intriguing reason forkeeping the term concept in the classificatory scheme ofcognitive science There is an important distinctionbetween those organisms that have this type of body ofknowledge and those that do not However I doubt thatthe class of organisms that have concepts would be ofinterest to comparative psychologists It will probably bemore fruitful to examine which organisms have proto-types and which processes in which species use theseor to compare the causal theories in humans and the the-ories (or proto-theories) in other species That is my con-cerns about the usefulness of the notion of concept forpsychologists working on human higher cognition carryover to comparative psychology It might even well bethat the term concept misleads us in thinking that theclass of organisms that have concepts is an interestingclass for comparative psychology exactly as it misleadsus in thinking that in the human mind they form an inter-esting class for cognitive scientists working on highercognition

Strohminger amp Moore note that keeping conceptwithin the classificatory scheme of psychology has numer-ous virtues Further research might undermine the evi-dence suggesting that there are very different kinds ofbody of knowledge used by default in higher cognitionand not eliminating concept might keep psychologistsaware of this possibility However I doubt that cautionis needed here because the evidence in support of theexistence of distinct kinds of concept seems unlikely tobe undermined Strohminger amp Moore also note that thedescription ldquobodies of knowledge used by default inhigher cognitionrdquo is ungainly and as a result unlikely tobe adopted by cognitive scientists as a replacement forconcept (see similar concerns in Mercier 2010) I hopethat the benefits that I argue would fall out from eliminat-ing concept will convince cognitive scientists that this is acost worth paying

R95 Does the eliminativist argument against ldquoconceptrdquoovergeneralize

According to Gonnerman amp Weinberg Khemlani ampGoodwin and Margolis amp Laurence my eliminativistargument cannot be valid because if it were we wouldhave to eliminate numerous notions that have earnedtheir keep in science in general and in cognitive sciencein particular such as the notions of representationmodule algorithm and nutrient After all representationsand nutrients are probably no more natural kinds thanconcepts

In response first I do not hold that concept should beeliminated merely because it fails to pick out a naturalkind Rather in addition to failing to pick out a naturalkind keeping concept has numerous drawbacks and fewbenefits Second the case of concepts is very differentfrom the case of say representations Psychologists donot attempt to discover generalizations about represen-tations in general and to encompass these generalizationswithin a theory of representations while they do preciselythis for concepts There is thus no theoretical habit to curb




in the case of representations while if I am right abouthow knowledge is organized cognitive scientistsrsquo ten-dencies to develop theories of concepts should becurbed (for some recent attempts see eg Gallese ampLakoff 2005 Martin 2007 Prinz forthcoming)

One might wonder whether the use of concept reallyimpedes the progress of cognitive science Strohmingeramp Moore rightly note that I provide little actual evidencein support of this claim Couchman et al seem to agreewith this criticism I acknowledge that this claim is partlyspeculative but it strikes me as plausible While the useof concept in cognitive science and the attempts todevelop a unified theory of concepts have not preventedcognitive scientists for making numerous findings aboutconcepts some important questions have not attracted suf-ficient attention such as How are prototypes exemplarsand theories used concomitantly And what happens whenthey yield incompatible judgments I am less convincedthan Couchman et al and Markman that we havealready acquired an extensive knowledge about thesequestions although some noticeable work has alreadybeen conducted (including by Smith and Markman) Ifurther hypothesize that much more work would bedone on these questions if cognitive scientists stopped the-orizing about concepts and started theorizing about proto-types exemplars and theories

References

[Letters ldquoardquo and ldquorrdquo appearing before authorsrsquo initials refer to target article

and response references respectively]

Allen S W amp Brooks L R (1991) Specializing the operation of an explicit ruleJournal of Experimental Psychology General 1203ndash19 [BKH arEM]

Anderson J R (1978) Arguments concerning representations for mental imageryPsychological Review 85249ndash77 [aEM]

Anderson J R amp Betz J (2001) A hybrid model of categorization PsychonomicBulletin and Review 8629ndash47 [aEM]

Ashby F G Alfonso-Reese L A Turken A U amp Waldron E M (1998) Aneuropsychological theory of multiple systems in category learning Psycho-logical Review 105(3)442ndash81 [ABM]

Ashby F G amp Ell S W (2002) Single versus multiple systems of learning andmemory In Stevensrsquo handbook of experimental psychology vol4 Method-ology in experimental psychology 3rd ed ed J Wixted amp H Pashler pp 655ndash92 Wiley [aEM]

Ashby F G amp Maddox W T (2004) Human category learning Annual Review ofPsychology 56149ndash78 [aEM]

Ashby F G amp Maddox W T (2005) Human category learning Annual Review ofPsychology 56149ndash78 [JJC BKH]

Barrera M E amp Maurer D (1981) Recognition of motherrsquos photographed face bythe three-month-old infant Child Development 52714ndash16 [YY]

Barsalou L W (1982) Context-independent and context-dependent information inconcepts Memory and Cognition 1082ndash93 [arEM]

Barsalou L W (1985) Ideals central tendency and frequency of instantiationas determinants of graded structure in categories Journal of ExperimentalPsychology Learning Memory and Cognition 11629ndash54 [aEM]

Barsalou L W (1987) The instability of graded structures Implications for thenature of concepts In Concepts and conceptual development Ecological andintellectual factors in categorization ed U Neisser (pp 101ndash38) CambridgeUniversity Press [aEM]

Barsalou L W (1993) Flexibility structure and linguistic vagary in conceptsManifestations of a compositional system of perceptual symbols In Theories ofmemory ed A C Collins S E Gathercole amp M A Conway pp 29ndash101Erlbaum [arEM]

Barsalou L W (1999) Perceptual symbol systems Behavioral and Brain Sciences22(4)577ndash660 [GD SH aEM]

Barsalou L W (2008a) Cognitive and neural contributions to understanding theconceptual system Current Directions in Psychological Science 1791ndash95[aEM]

Barsalou L W (2008b) Grounded cognition Annual Review of Psychology59617ndash45 [aEM]

Barsalou L W (2009) Simulation situated conceptualization and predictionPhilosophical Transactions of the Royal Society of London Biological Sciences3641281ndash89 [aEM]

Barsalou L W amp Hale C R (1993) Components of conceptual representationFrom feature lists to recursive frames In Categories and conceptsTheoretical views and inductive data analysis ed I van Mechelen J AHampton R S Michalski amp P Theuns pp 97ndash144 Academic Press[JAH]

Barsalou L W Simmons W K Barbey A K amp Wilson C D (2003) Groundingconceptual knowledge in modality-specific systems Trends in CognitiveSciences 784ndash91 [aEM]

Berndt R S Haendiges A N Burton M W amp Mitchum C C (2002) Gram-matical class and imageability in aphasic word production Their effects areindependent Journal of Neurolinguistics 15353ndash71 [GD]

Binder J Westbury C McKiernan K Possing E amp Medler D (2005) Distinctbrain systems for processing concrete and abstract concepts Journal of Cog-nitive Neuroscience 17905ndash17 [GD]

Bird H Howard D amp Franklin S (2003) Verbs and nouns The importance ofbeing imageable Journal of Neurolinguistics 16113ndash49 [GD]

Blair M amp Homa D (2003) As easy to memorize as they are to classify The 5-4categories and the category advantage Memory and Cognition 311293ndash1301 [JJC]

Blondin-Masse A Chicoisne G Gargouri Y Harnad S Picard O amp MarcotteO (2008) How is meaning grounded in dictionary definitions Paper presentedat TextGraphs-3 Workshop ndash 22nd International Conference on Compu-tational Linguistics 18 August 2008 [SH]

Bloom P (1996) Intention history and artifact concepts Cognition 601ndash29[aEM]

Bloom P (2000) How children learn the meanings of words MIT Press [EL]Boulenger V Hauk O amp Pulvermuller F (2009) Grasping ideas with the motor

system Semantic somatotopy in idiom comprehension Cerebral Cortex191905ndash14 [aEM]

Boyd R (1989) What realism implies and what it does not Dialectica 435ndash29[RS]

Boyd R (1991) Realism anti-foundationalism and the enthusiasm for naturalkinds Philosophical Studies 61127ndash48 [aEM RS]

Boyd R (1999) Kinds complexity and multiple realization Philosophical Studies9567ndash98 [aEM]

Braisby N (2005) Similarity and categorisation Getting dissociations inperspective In Proceedings of the Twenty-Sixth Annual CognitiveScience Society ed K Forbus D Getner amp T Regier pp 150ndash55Erlbaum [CG]

Buresh J S amp Woodward A L (2007) Infants track action goals within and acrossagents Cognition 104287ndash314 [YY]

Burge T (1979) Individualism and the mental Midwest Studies in Philosophy473ndash121 [rEM]

Burge T (1986) Individualism and psychology Philosophical Review 95(1)3ndash46 [GR]

Butterworth G Jarrett N amp Hicks L (1982) Spatiotemporal identity in infancyPerceptual competence or conceptual deficit Developmental Psychology18435ndash49 [HAV]

Cangelosi A amp Harnad S (2001) The adaptive advantage of symbolic theft oversensorimotor toil Grounding language in perceptual categories Evolution ofCommunication 4(1)117ndash14 [SH]

Caramazza A (1986) On drawing inferences about the structure of normalcognitive systems from the analysis of patterns of impairedperformance The case for single-patient studies Brain and Cognition541ndash66 [aEM]

Carey S (1985) Conceptual change in childhood MIT Press [TL]Carey S (2009) The origin of concepts Oxford University Press [EMar]Carey S amp F Xu (2001) Infantsrsquo knowledge of objects Beyond object files and

object tracking Cognition 80(1ndash2)179 [AJJ]Carlson G N amp Pelletier F J (1995) The generic book University of Chicago

Press [SSK]Carmichael C amp Hayes B K (2001) Prior knowledge and exemplar encoding in

childrenrsquos concept acquisition Child Development 721071ndash90 [BKH]Chao L L amp Martin A (1999) Cortical representation of perception

naming and knowledge of color Journal of Cognitive Neuroscience1125ndash35 [aEM]

Chaput H H amp Cohen L B (2001) A model of infant causal perception and itsdevelopment In Proceedings of the Twenty-Third Annual Conference of theCognitive Science Society ed JD Moore amp K Stenning pp 182ndash87Erlbaum [YY]

Chin-Parker S amp Ross B H (2002) The effect of category learning on sensitivityto within-category correlations Memory and Cognition 30(3)353ndash62[ABM]

ReferencesMachery Precis of Doing without Concepts


Churchland P M (1981) Eliminative materialism and the propositional attitudesJournal of Philosophy 7867ndash90 [aEM]

Cohen B amp Murphy G L (1984) Models of concepts Cognitive Science 827ndash58 [AS]

Collini E Wong C Y Wilk K E Curmi P M G Brumer P amp Scholes G D(2010) Coherently wired light-harvesting in photosynthetic marine algae atambient temperature Nature 463644 [rEM]

Connolly A C Fodor J A Gleitman L R amp Gleitman H (2007) Why stereo-types donrsquot even make good defaults Cognition 1031ndash22 [SSK]

Couchman J J Coutinho M V C amp Smith J D (in press) Rules and resem-blance Their changing balance in the category learning of humans (Homosapiens) and rhesus monkeys (Macaca mulatta) Journal of Experimental Psy-chology Animal Behavior Processes [JJC]

Crepaldi D Aggujaro S Arduino L S Zonca G Ghirardi G Inzaghi M GColombo M Chierchia G amp Luzzatti C (2006) Noun-verb dissociation inaphasia The role of imageability and functional location of the lesion Neu-ropsychologia 44(1)73ndash89 [GD]

Danks D (2004) Psychological theories of categorization as probabilistic modelsTechnical report CMU-PHIL-157 July 15 2004 [DD]

Danks D (2007) Theory unification and graphical models in human categorizationIn Causal learning Psychology philosophy and computation ed A Gopnikamp L Schulz pp 173ndash89 Oxford University Press [DD rEM]

Devitt M (1981) Designation Columbia University Press [GR]Devitt M (1996) Coming to our senses Cambridge University Press [GR]Devitt M (forthcoming) Experimental semantics Philosophy and Phenomenolo-

gical Research [GR]Diamond A (1985) Development of the ability to use recall to guide action as

indicated by infantsrsquo performance on A-not-B Child Development 56868ndash83 [HAV]

Dove G (2009) Beyond perceptual symbols A call for representational pluralismCognition 110412ndash31 [GD arEM]

Dretske F (1981) Knowledge and the flow of information BlackwellMIT Press[KE GR]

Dunn J C (2008) The dimensionality of the remember-know task A state-traceanalysis Psychological Review 115(2)426ndash46 [BKH]

Dunn J C amp Kirsner K (1988) Discovering functionally independent mentalprocesses The principle of reversed association Psychological Review 9591ndash101 [aEM]

Dunn J C amp Kirsner K (2003) What can we infer from double dissociationsCortex 391ndash7 [aEM]

Edwards K (2009) What concepts do Synthese 170289ndash310 [KE aEM]Edwards K (2010) Concept referentialism and the role of empty concepts Mind

and Language 25(1)89ndash118 [KE]Elder C L (1994) Higher and lower essential natures American Philosophical

Quarterly 31255ndash65 [CG]Evans G (1982) The varieties of reference Oxford University Press [rEM]Evans J S B T (2007) Hypothetical thinking Dual processes in reasoning and

judgement Routledge [aEM]Evans J S B T amp Frankish K eds (2009) In two minds Dual processes and

beyond Oxford University Press [aEM JV]Feeney A amp Heit E eds (2007) Inductive reasoning Experimental Develop-

mental and computational approaches Cambridge University Press [aEM]Feldman J (2000) Minimization of Boolean complexity in human concept learning

Nature 407630ndash33 [SSK]Feldman J (2006) An algebra of human concept learning Journal of Mathematical

Psychology 50339ndash68 [SSK]Field H (2001) Truth and the absence of fact Oxford University Press [KE]Flynn J J Nedbal M A Dragoo J W amp Honeycutt R L (2000) Whence the

red panda Molecular Phylogenetics and Evolution 17(2)190ndash99 [AJJ]Fodor J A (1974) Special sciences (Or The disunity of science as a working

hypothesis) Synthese 28(2)97ndash115 [KE]Fodor J A (1975) The language of thought Crowell [aEM DAW]Fodor J A (1987) Psychosemantics The problem of meaning in the philosophy of

mind MIT Press [KE]Fodor J A (1990) A theory of content and other essays MIT Press [rEM GR]Fodor J A (1992) A theory of content and other essays MIT Press [KE]Fodor J A (1997) Special sciences Still autonomous after all these years

Philosophical Perspectives Mind Causation and World 249ndash63 (Nous 31Suppl) [KE]

Fodor J A (1998) Concepts Where cognitive science went wrong Oxford Uni-versity Press [FK aEM GR SS]

Fodor J A (2003) Is it a bird Problems with old and new approaches to the theory ofconcepts Times Literary Supplement January 17 2003 pp 3ndash4 [aEM]

Fodor J A (2008) LOT 2 The language of thought revisited Oxford UniversityPress [aEM SS DAW]

Forster M amp Sober E (1994) How to tell when simpler more unified or less adhoc theories will provide more accurate predictions British Journal for thePhilosophy of Science 451ndash36 [rEM]

Gallese V amp Lakoff G (2005) The brainrsquos concepts The role of the sensory-motorsystem in conceptual knowledge Cognitive Neuropsychology 21455ndash79[arEM]

Garrod S amp Anderson A (1987) Saying what you mean in dialogue A study inconceptual and semantic co-ordination Cognition 27181ndash218 [ABM]

Garrod S amp Doherty G (1994) Conversation co-ordination and convention Anempirical investigation of how groups establish linguistic conventions Cogni-tion 53181ndash215 [ABM]

Gelman R (2004) Cognitive development In Stevensrsquo handbook of experimentalpsychology vol 3 Memory and cognitive processes ed H Pashler amp D LMedin pp 533ndash60 Wiley [aEM]

Gelman S A (2003) The essential child Origins of essentialism in everydaythought Oxford University Press [TB SSK]

Gelman S A (1988) The development of induction within natural kinds and arti-facts categories Cognitive Psychology 2065ndash95 [aEM]

Gelman S A (2003) The essential child Origins of essentialism in everydaythought Oxford University Press [aEM]

Gelman S A amp Markman E (1986) Categories and induction in young childrenCognition 23183ndash209 [aEM]

Gibson E J (1969) Principles of perceptual learning and development PrenticeHall [HAV]

Giesbrecht B Gamblin C amp Swaab T (2004) Separable effects of priming andimageability on word processing in human cortex Cerebral Cortex 14521ndash29 [GD]

Gigerenzer G amp Regier T P (1996) How do we tell an association from a rulePsychological Bulletin 11923ndash26 [aEM]

Gigerenzer G Todd P M amp the ABC Research Group (1999) Simple heuristicsthat make us smart Oxford University Press [aEM]

Glenberg A M (1997) What memory is for Behavioral and Brain Sciences 201ndash55 [GD aEM]

Glenberg A M amp Robertson D A (2000) Symbol grounding and meaning Acomparison of high-dimensional and embodied theories of meaning Journal ofMemory and Language 43379ndash401 [SH]

Glymour C (1994) On the methods of cognitive neuropsychology British Journalfor the Philosophy of Science 45815ndash35 [aEM]

Goldberg R F Perfetti C A amp Schneider W (2006) Distinct and commoncortical activations for multimodal semantic categories Cognitive Affectiveand Behavioral Neuroscience 6214ndash22 [GD]

Goodwin G P amp Johnson-Laird P N (2010) Conceptual illusions Cognition114253ndash65 [SSK]

Goodwin G P amp Johnson-Laird P N (submitted) Models as the representation ofBoolean concepts [SSK]

Gopnik A (2003) The theory theory as an alternative to the innateness hypothesisIn Chomsky and his critics ed L Antony amp N Hornstein pp 238ndash54Blackwell [aEM]

Gopnik A Glymour C Sobel D Schulz L Kushnir T amp Danks D (2004) Atheory of causal learning in children Causal maps and Bayes nets Psycho-logical Review 1111ndash31 [aEM]

Gopnik A amp Meltzoff A N (1997) Words thoughts and theories MIT Press[TL aEM]

Greene J D Sommerville R B Nystrom L E Darley J M amp Cohen J D(2001) An fMRI investigation of emotional engagement in moral JudgmentScience 2932105ndash108 [aEM]

Griffiths P E (1997) What emotions really are Chicago University Press [aEMAS RS]

Griffiths T L Steyvers M amp Tenenbaum J B (2007) Topics in semantic rep-resentation Psychological Review 114211ndash44 [aEM]

Hampton J A (1979) Polymorphous concepts in semantic memory Journal ofVerbal Learning and Verbal Behavior 18441ndash61 [TB aEM]

Hampton J A (1981) An investigation of the nature of abstract concepts Memoryand Cognition 9149ndash56 [aEM]

Hampton J A (1993) Prototype models of concept representation In Categoriesand concepts Theoretical views and inductive data analysis ed I VanMechelen J A Hampton R S Michalski amp P Theuns pp 67ndash95 AcademicPress [aEM]

Hampton J A (1998) Similarity-based categorization and fuzziness of naturalcategories Cognition 65137ndash65 [JAH]

Hampton J A (2001) The role of similarity in natural categorization In Similarityand categorization ed U Hahn amp M Ramscar pp 13ndash28 Oxford UniversityPress [rEM]

Hampton J A (2006) Concepts as prototypes In The psychology of learning andmotivation Advances in research and theory vol 46 ed B H Ross pp 79ndash113 Academic Press [aEM]

Hampton J A (2007) Typicality graded membership and vagueness CognitiveScience 31355ndash84 [aEM]

Hampton J A Storms G Simmons C L amp Heussen D (2009) Feature integrationin natural language concepts Memory and Cognition 371721ndash30 [FK]

Harnad S (1990) The symbol grounding problem Physica D 42335ndash46 [SH]




Harnad S (2005) To cognize is to categorize cognition is categorization In Handbookof Categorization ed C Lefebvre amp H Cohen pp 20ndash42 Elsevier [SH]

Harnad S (2006) Cohabitation Computation at 70 cognition at 20 In Essays inHonour of Zenon Pylyshyn ed D Dedrick pp 245ndash57 MIT Press [SH]

Harnad S (2007) From knowing how to knowing that Acquiring categories byword of mouth Paper presented at the Kaziemierz Naturalized EpistemologyWorkshop (KNEW) Kaziemierz Poland September 2 2007 [SH]

Harnad S (2008) The annotation game On Turing (1950) on computingmachinery and intelligence In Parsing the Turing test Philosophical andmethodological issues in the quest for the thinking computer ed R Epstein ampG Peters pp 23ndash66 Springer [SH]

Hauk O Johnsrude I amp Pulvermuller F (2004) Somatotopic representation ofaction words in human motor and premotor cortex Neuron 41301ndash07[aEM]

Heit E (1994a) Models of the effects of prior knowledge on category learningJournal of Experimental Psychology Learning Memory and Cognition201264ndash82 [BKH]

Heit E (1994b) Similarity and property effects in inductive reasoning Journal ofExperimental Psychology Learning Memory and Cognition 20411ndash22[AS]

Heit E (2000) Properties of inductive reasoning Psychonomic Bulletin and Review7569ndash92 [aEM]

Heit E amp Hayes B (2008) Predicting reasoning from visual memory In Pro-ceedings of the 29th Meeting of the Cognitive Science Society ed V SloutskyB Love amp K McCrae pp 83ndash88 Erlbaum [BKH]

Hill C (2010) I love Macheryrsquos book but love concepts more Philosophical Studies149411ndash21 [rEM]

Hilton D J amp Slugoski B R (1986) Knowledge-based causal attributionThe abnormal conditions focus model Psychological Review 9375ndash88[TL]

Hoenig K Sim E-J Bochev V Herrnberger B amp Kiefer M (2008) Conceptualflexibility in the human brain Dynamic recruitment of semantic maps fromvisual motor and motion-related areas Journal of Cognitive Neuroscience20(10)1799ndash814 [arEM]

Holcomb P J Kounios J Anderson J E amp West W C (1999) Dual-codingcontext-availability and concreteness effects in sentence comprehension Anelectrophysiological investigation Journal of Experimental PsychologyLearning Memory and Cognition 25721ndash42 [GD]

Homa D Sterling S amp Trepel L (1981) Limitations of exemplar-based gener-alization and the abstraction of categorical information Journal of Exper-imental Psychology Human Learning and Memory 7418ndash39 [JJC]

Horobin K amp Acredolo L (1986) The role of attentiveness mobility history andseparation of hiding sites on stage IV search behavior Journal of ExperimentalChild Psychology 41114ndash27 [HAV]

Horwich P (1998a) Meaning Clarendon Press [KE GR]Horwich P (1998b) Truth 2nd edition Clarendon Press [KE]Jackendoff R (2002) Foundations of language Brain meaning grammar evol-

ution Oxford University Press [EMar]James T W amp Gauthier I (2003) Auditory and action semantic features activate

sensory-specific perceptual brain regions Current Biology 131792ndash96[arEM]

Jonides J Lewis R Nee D E Lustig C A Berman M G amp Moore K S(2008) The mind and brain of short-term memory Annual Review of Psy-chology 59151ndash1532 [NS]

Kable J W Kan I P Wilson A Thompson-Schill S L amp Chatterjee A (2005)Conceptual representations of action in the lateral temporal cortex Journal ofCognitive Neuroscience 171855ndash870 [aEM]

Keil F C (1989) Concepts kinds and cognitive development MIT Press [TBFK aEM AS SS]

Keil F C (2006) Explanation and understanding Annual Review of Psychology57227ndash54 [TL]

Keil F C Carter Smith W Simons D J amp Levin D T (1998) Two dogmas ofconceptual empiricism Implications for hybrid models of the structure ofknowledge Cognition 65103ndash35 [aEM]

Keil F C amp Newman G E (2010) Darwin and development Why ontogeny doesnot recapitulate phylogeny for human concepts In The making of humanconcepts ed D Mareschal P Quin amp S Lea pp 317ndash34 Oxford UniversityPress [FK]

Kelemen D (1999) Function goals and intention Childrenrsquos teleologicalreasoning about objects Trends in Cognitive Sciences 3461ndash68 [TL]

Kellenbach M L Wijers A A Hovis M Mulder J amp Mulder G (2002) Neuraldifferentiation of lexico-syntactic categories or semantic features Eventrelated potential evidence for both Journal of Cognitive Neuroscience 14561ndash77 [GD]

Kemp C amp Tenenbaum J (2009) Structured statistical models of inductivereasoning Psychological Review 11620ndash58 [BKH]

Khemlani S Leslie S J amp Glucksberg S (submitted) Generics modulate defaultinferences [SSK]

Kiefer M Sim E-J Liebich S Hauk O amp Tanaka J (2007) Experience-dependent plasticity of conceptual representations in human sensoryndashmotorareas Journal of Cognitive Neuroscience 19525ndash42 [aEM]

Kim J (1992) Multiple realization and the metaphysics of reduction Philosophyand Phenomenological Research 52(1)1ndash26 [KE EL]

Kim J (1998) Mind in a physical world An essay on the mind-body problem andmental causation MIT Press [KE]

Knowlton B J amp Squire L R (1993) The learning of categories Parallel brainsystems for item memory and category knowledge Science 2621747ndash49[SZ]

Knowlton B J Squire L R amp Gluck M A (1994) Probabilistic classificationlearning in amnesia Learning and Memory 1106ndash20 [ABM]

Kripke S (19721980) Naming and necessity Harvard University Press [GR]Kruschke J K (2005) Category learning In The handbook of cognition ed K

Lamberts amp R L Goldstone pp 183ndash201 Sage [BKH]Kulatanga-Moruzi C Brooks L R amp Norman G R (2001) Coordination of

analytical and similarity based processing strategies and expertise in derma-tological diagnosis Journal of Experimental Psychology Learning Memoryand Cognition 30563ndash72 [rEM]

Lakoff G (1972) Hedges A study in meaning criteria and the logic of fuzzy con-cepts In Papers from the Eighth Regional Meeting Chicago LinguisticSociety Chicago IL pp 183ndash228 Chicago Linguistic Society [FK]

Laurence S amp Margolis E (1999) Concepts and cognitive science In Conceptscore readings ed E Margolis amp S Laurence pp 3ndash82 MIT Press [TLEMar aEM]

Laurence S amp Margolis E (2002) Radical concept nativism Cognition 8625ndash55 [SS]

Lauritzen S L (1996) Graphical models Oxford University Press [DD]Lawler J (1973) Studies in English generics University of Michigan Papers in

Linguistics vol 1 University of Michigan Press [SSK]Logan G D (1988) Toward an instance theory of automaticity Psychological

Review 95492ndash527 [ABM]Logan G D (2002) An instance theory of attention and memory Psychological

Review 109(2)376ndash400 [ABM]Lombrozo T (2006) The structure and function of explanations Trends in Cog-

nitive Sciences 10(10)464ndash70 [TL]Lombrozo T (2009) Explanation and categorization How ldquowhyrdquo informs ldquowhatrdquo

Cognition 110248ndash53 [TL]Lombrozo T (under review) Causal-explanatory pluralism How intentions func-

tions and mechanisms influence causal ascriptions [TL]Lombrozo T amp Carey S (2006) Functional explanation and the function of

explanation Cognition 99(2)167ndash204 [TL]Lombrozo T Kelemen D amp Zaitchik D (2007) Inferring design Evidence of a

preference for teleological explanations in patients with Alzheimerrsquos diseasePsychological Science 18(11)999ndash1006 [TL]

Lopez A Atran S Coley J D Medin D L amp Smith E E (1997) The tree of lifeUniversal and cultural features of folkbiological taxonomies and inductionsCognitive Psychology 32251ndash95 [aEM]

Love B C Medin D L amp Gureckis T M (2004) SUSTAIN A network model ofcategory learning Psychological Review 111309ndash32 [BKH]

Luhmann C C Ahn W amp Palmeri T (2006) Theory-based categorizationunder speeded conditions Memory and Cognition 341102ndash11 [TBrEM]

Machery E (2005) Concepts are not a natural kind Philosophy of Science 72444ndash67 [arEM RS]

Machery E (2006a) How to split concepts Reply to Piccinini and Scott Philosophyof Science 73410ndash18 [aEM]

Machery E (2006b) Review of A Zilhao ed Evolution rationality andcognition A cognitive science for the twenty-first century Notre DamePhilosophical Reviews Retrieved from httpndprndedureviewcfmidfrac146342 [rEM]

Machery E (2006c) Two dogmas of neo-empiricism Philosophy Compass 1398ndash412 [arEM]

Machery E (2007) Concept empiricism A methodological critique Cognition10419ndash46 [GD arEM]

Machery E (2009) Doing without concepts Oxford University Press [TB JJCDD GD KE CG JAH SH BKH AJJ FK SSK EL TL EMar arEMABM GR RS AS SS NS JV HAV DAW YY SZ]

Machery E (forthcoming) Reply to Barbara Malt and Jesse Prinz Mind andLanguage [rEM]

Machery E (2010) Reply to my critics Philosophical Studies 149429ndash36 [rEM]Machery E amp Seppala S (forthcoming) Against hybrid theories of concepts

Anthropology amp Philosophy [arEM]Maddox W T amp Ashby F G (1993) Comparing decision-bound and exemplar

models of classification Perception and Psychophysics 5349ndash70 [SZ]Maddox W T amp Ashby F G (2004) Dissociating explicit and procedure-learning

based systems of perceptual category learning Behavioral Processes66(3)309ndash32 [ABM]



Mahon B Z amp Caramazza A (2008) A critical look at the embodied cognitionhypothesis and a new proposal for grounding conceptual content Journal ofPhysiology Paris 10259ndash70 [aEM]

Mahon B Z amp Caramazza A (2009) Concepts and categories A cognitive neu-ropsychological perspective Annual Review of Psychology 6027ndash51 [aEM]

Mallon R (2006) ldquoRacerdquo Normative not metaphysical or semantic Ethics116525ndash51 [aEM]

Mallon R Machery E Nichols S amp Stich S P (2009) Against argumentsfrom reference Philosophy and Phenomenological Research 79332ndash56[aEM]

Malt B C (1989) An on-line investigation of prototype and exemplar strategies inclassification Journal of Experimental Psychology Learning Memory andCognition 15(4)539ndash55 [JV]

Malt B C (1994) Water is not H2O Cognitive Psychology 2741ndash70 [aEM]Malt B C (forthcoming) Why we should do without concepts Mind and

Language [rEM]Malt B C amp Sloman S A (2007) Artifact categorization The good the bad and

the ugly In Creations of the mind Theories of artifacts and their represen-tation ed E Margolis amp S Laurence pp 85ndash123 Oxford University Press[aEM]

Malt B C Sloman S A amp Gennari S P (2003) Universality and languagespecificity in object naming Journal of Memory and Language 4920ndash42[ABM]

Margolis E (1994) A reassessment of the shift from the classical theory of conceptsto prototype theory Cognition 5173ndash89 [aEM]

Margolis E (1995) The significance of the theory analogy in the psychological studyof concepts Mind and Language 1045ndash71 [aEM]

Margolis E amp Laurence S (2006) Concepts Stanford Encyclopedia of PhilosophyAvailable at httpplatostanfordeduentriesconcepts [aEM]

Margolis E amp Laurence S (2007) The ontology of concepts ndash Are conceptsabstract objects or mental representations Nous 41(4)561ndash93 [EMar]

Margolis E amp Laurence S eds (1999) Concepts Core readings MIT Press[SS]

Markman A B amp Dietrich E (2000) In defense of representation CognitivePsychology 40(2)138ndash71 [ABM]

Markman A B amp Makin V S (1998) Referential communication and categoryacquisition Journal of Experimental Psychology General 127(4)331ndash54[ABM]

Markman A B amp Ross B (2003) Category use and category learning Psycho-logical Bulletin 129592ndash613 [BKH ABM]

Marr D (1982) Vision A computational investigation in the human representationof visual information Freeman [SSK]

Marshall J Pring T Chiat S amp Robson J (1996) Calling a salad a federation Aninvestigation of semantic jargon Part 1 ndash nouns Journal of Neurolinguistics9237ndash50 [GD]

Martin A (2007) The representation of object concepts in the brain AnnualReview of Psychology 5825ndash45 [arEM]

Martin A amp Chao L L (2001) Semantic memory and the brain Structure andprocesses Current Opinion in Neurobiology 11194ndash201 [aEM]

Martin A Haxby J V Lalonde F M Wiggs C L amp Ungerleider L G (1995)Discrete cortical regions associated with knowledge of color and knowledge ofaction Science 270102ndash05 [aEM]

Mayr E (1982) The growth of biological thought Diversity evolution andinheritance Harvard University Press [JAH]

Medin D L Lynch E B amp Solomon K O (2000) Are there kinds of conceptsAnnual Review of Psychology 51121ndash47 [aEM]

Medin D L amp Schaffer M M (1978) Context theory of classification learningPsychological Review 85207ndash38 [aEM]

Mellet E Tzourio N Denis M amp Mazoyer B (1998) Cortical anatomy of mentalimagery of concrete nouns based on their dictionary definition NeuroReport9803ndash08 [GD]

Mercier H (2010) How to cut a concept Review of Doing without concepts byEdouard Machery Biology and Philosophy 25269ndash77 [rEM]

Millikan R G (1984) Language thought and other biological categories Newfoundations for realism MIT Press [KE GR]

Millikan R G (1993) White Queen psychology and other essays for Alice MITPress [KE]

Millikan R G (1998) A common structure for concepts of individuals stuffs andbasic kinds More mama more milk and more mouse Behavioral and BrainSciences 2255ndash65 [EL]

Millikan R G (2000) On clear and confused ideas An essay about substanceconcepts Cambridge University Press [aEM]

Minda J P amp Smith J D (2001) Prototypes in category learning The effects ofcategory size category structure and stimulus complexity Journal of Exper-imental Psychology Learning Memory and Cognition 27775ndash99 [aEM SZ]

Morris C D Bransford J D amp Franks J J (1977) Levels of processing versustransfer appropriate processing Journal of Verbal Learning and VerbalBehavior 16519ndash33 [ABM]

Murphy D amp Stich S P (1999) Griffiths elimination and psychopathologyMetascience 813ndash25 [aEM]

Murphy G L (2002) The big book of concepts MIT Press [JJC JAH BKH AJJEL aEM AS]

Murphy G L amp Medin D L (1985) The role of theories in conceptual coherencePsychological Review 92289ndash316 [TB BKH TL aEM]

Myung I J Forster M R amp Browne M W eds (2000) Model selection [Specialissue] Journal of Mathematical Psychology 44190ndash204 [SZ]

Myung I J amp Pitt M A (2009) Optimal experimental design for model dis-crimination Psychological Review 116499ndash518 [SZ]

Navarro D J Pitt M A amp Myung I J (2004) Assessing the distinguishability ofmodels and the informativeness of data Cognitive Psychology 4947ndash84[SZ]

Nesse R M amp Ellsworth P C (2009) Evolution emotions and emotionaldisorders American Psychologist 64129ndash39 [NS]

Newell B amp Dunn J C (2008) Dimensions in data Testing psychological modelsusing state-trace analysis Trends in Cognitive Sciences 12(8)285ndash90[BKH]

Noppeney U amp Price C J (2004) Retrieval of abstract semantics NeuroImage22164ndash70 [GD]

Norman G Eva K Brooks L amp Hamstra S (2006) Expertise in medicine andsurgery In The Cambridge handbook of expertise and expert performance edK A Ericsson N Charness P J Feltovich amp R R Hoffman pp 339ndash53Cambridge University Press [rEM]

Nosofsky R M (1986) Attention similarity and the identification-categorizationrelationship Journal of Experimental Psychology Learning Memory andCognition 10104ndash14 [aEM]

Nosofsky R M (1992) Exemplar-based approach to relating categorizationidentification and recognition In Multidimensional models of perception andcognition ed F G Ashby pp 363ndash93 Erlbaum [aEM]

Nosofsky R M Palmeri T J amp McKinley S C (1994) Rule-plus-exception modelof classification learning Psychological Review 101266ndash300 [aEM]

Nosofsky R M amp Stanton R D (2005) Speeded classification in a probabilisticcategory structure Contrasting exemplar-retrieval decision-boundary andprototype models Journal of Experimental Psychology Human Perception andPerformance 31608ndash29 [aEM]

Nosofsky R M amp Zaki S R (2002) Exemplar and prototype models revisitedResponse strategies selective attention and stimulus generalization Journal ofExperimental Psychology Learning Memory and Cognition 285924ndash40[SZ]

Ogmen H (2007) A theory of moving form perception Synergy between maskingperceptual grouping and motion computation in retinotopic and non-retino-topic representations Advances in Cognitive Psychology 3(1ndash2)67ndash84 [AJJ]

Ogmen H Otto T U amp Herzog M H (2006) Perceptual grouping induces non-retinotopic feature attribution in human vision Vision Research 46(19)3234ndash242 [AJJ]

Osherson D N amp Smith E E (1981) On the adequacy of prototype theory as atheory of concepts Cognition 935ndash58 [arEM]

Osherson D N Smith E E Wilkie O Lopez A amp Shafir E (1990) Category-based induction Psychological review 97185ndash200 [aEM]

Paivio A (1987) Mental representations A dual-coding approach Oxford Univer-sity Press [GD]

Paivio A (1991) Dual coding theory Retrospect and current status CanadianJournal of Psychology 45255ndash87 [rEM]

Palmeri T J amp Flanery M A (1999) Learning about categories in the absenceof training Profound amnesia and the relationship between perceptualcategorization and recognition memory Psychological Science 10(6)526ndash30 [JV]

Peacocke C (1992) A study of concepts MIT Press [aEM GR]Peacocke C (2008) Truly understood Oxford University Press [aEM]Pecher D Zeelenberg R amp Barsalou L W (2004) Sensorimotor simulations

underlie conceptual representations Modality-specific effects of prior acti-vation Psychonomic Bulletin and Review 11164ndash67 [aEM]

Piaget J (1954) The construction of reality in the child Basic Books [HAV]Piccinini G (forthcoming) Two kinds of concept Implicit and explicit Dialogue

[JV]Piccinini G amp Scott S (2006) Splitting concepts Philosophy of Science 73390ndash

409 [GD EL aEM AS JV]Pinker S (1997) How the mind works Norton [EMar]Plaut D C (1995) Double dissociation without modularity Evidence from con-

nectionist neuropsychology Journal of Clinical and Experimental Neuropsy-chology 17291ndash21 [aEM]

Plumert J M (2008) Childrenrsquos thinking is not just about whatrsquos in the headUnderstanding the organism and environment as a unified system InAdvances in child development and behavior ed R V Kail pp 373ndash417Academic Press [HAV]

Prasada S amp Dillingham E M (2006) Principled and statistical connections incommon sense conception Cognition 9973ndash112 [TL]




Prasada S amp Dillingham E M (2009) Representation of principled connections Awindow onto the formal aspect of common sense conception CognitiveScience 33401ndash48 [SSK]

Prasada S Khemlani S Leslie S J amp Glucksberg S (submitted) Conceptualdistinctions amongst generics [SSK]

Prinz J J (2002) Furnishing the mind Concepts and their perceptual basis MITPress [GD arEM SS]

Prinz J J (2005) The return of concept empiricism In Handbook of categorizationin cognitive science ed H Cohen amp C Lefebvre pp 679ndash95 Elsevier[aEM]

Prinz J J (forthcoming) Can concept empiricism forestall eliminativism Acommentary on Machery Mind amp Language [rEM]

Proffitt J B Coley J D amp Medin D L (2000) Expertise and category-basedinduction Journal of Experimental Psychology Learning Memory andCognition 26811ndash28 [aEM]

Pulvermuller F (2005) Brain mechanisms linking language and action NatureReviews Neuroscience 6576ndash82 [aEM]

Pulvermuller F amp Hauk O (2006) Category-specific conceptual processing ofcolor and form in left fronto-temporal cortex Cerebral Cortex 161193ndash1201 [aEM]

Putnam H (1967) Psychological predicates In Art mind and religion ed W HCapitan amp D D Merrill pp 37ndash48 University of Pittsburgh Press [EL]

Putnam H (1975) The meaning of ldquomeaningrdquo In Collected papers vol 2Cambridge University Press [GR]

Pylyshyn Z W (1973) What the mindrsquos eye tells the mindrsquos brain A critique ofmental imagery Psychological Bulletin 801ndash24 [SH]

Pylyshyn Z W (2007) Things and places How the mind connects with the worldMIT Press [EMar]

Quine W V O (1969) Natural kinds In Ontological relativity and other essaysed W V O Quine pp 114ndash38 Columbia University Press [aEM]

Regehr G Cline J Norman G R amp Brooks L R (1994) Effect of processing ondiagnostic skill in dermatology Academic Medicine IS34ndashS36 [rEM]

Rehder B (2003a) A causal-model theory of conceptual representation andcategorization Journal of Experimental Psychology Learning Memoryand Cognition 291141ndash59 [aEM]

Rehder B (2003b) Categorization as causal reasoning Cognitive Science 27709ndash48 [TL]

Rehder B (2006) When causality and similarity compete in category-based prop-erty induction Memory and Cognition 343ndash16 [TL aEM]

Rehder B amp Kim S (2006) How causal knowledge affects classification A gen-erative theory of categorization Journal of Experimental Psychology LearningMemory and Cognition 32659ndash83 [aEM]

Rehder B amp Murphy G L (2003) A knowledge-resonance (KRES) model ofcategory learning Psychological Bulletin amp Review 10789ndash94 [BKH]

Rey G (1983) Concepts and stereotypes Cognition 15237ndash62 [aEM GR]Rey G (1985) Concepts and conceptions A reply to Smith Medin and Rips

Cognition 19297ndash303 [aEM GR]Rey G (2009a) Concepts defaults and internal asymmetric dependencies

Distillations of Fodor and Horwich In The a priori and its role in philosophyed N Kompa C Nimtz amp C Suhm Mentis pp 185ndash203 Mentis [GR]

Rey G (2009b) Review of E Machery Doing without Concepts Notre DamePhilosophical Reviews (Online journal Epub 20090715) Available at httpndprndedureviewcfmidfrac1416608 [KE aEM]

Rips L J (1989) Similarity typicality and categorization In Similarity andanalogical reasoning ed S Vosniadou amp A Ortony pp 21ndash59 CambridgeUniversity Press [TB TL aEM]

Roediger H L (2008) Relativity of remembering Why the laws of memory van-ished Annual Review of Psychology 59225ndash54 [BKH]

Rosch E amp Mervis C B (1975) Family resemblance Studies in the internalstructure of categories Cognitive Psychology 7573ndash605 [TB aEM]

Roth E M amp Shoben E J (1983) The effect of context on the structure ofcategories Cognitive Psychology 15346ndash78 [aEM]

Rubenstein A J Kalakanis L amp Langlois J H (1999) Infant preferences forattractive faces A cognitive explanation Developmental Psychology 35848ndash55 [YY]

Russell B (1948) Human knowledge Its scope and its limits Routledge amp KeganPaul [aEM]

Sabsevitz D Medler D Seidenberg M amp Binder J (2005) Modulation of thesemantic system by word imageability NeuroImage 27188ndash200 [GD]

Samuelson L K Schutte A R amp Horst J S (2009) The dynamic nature ofknowledge Insights from a dynamic field model of childrenrsquos novel noungeneralizations Cognition 110322ndash45 [HAV]

Samuelson L K amp Smith L B (1998) Memory and attention make smart wordlearning An alternative account of Akhtar Carpenter and Tomasello ChildDevelopment 6994ndash104 [HAV]

Sandhofer C M amp Doumas L A A (2008) Order of presentation effects in thelearning of color categories Journal of Cognition and Development 9194ndash221 [HAV]

Sandhofer C M amp Thom E E (2006) Taking the task seriously Reflections onmeasures of color acquisition Journal of Experimental Child Psychology94344ndash48 [HAV]

Sato J J Wolsan M Minami S Hosoda T Sinaga M H Hiyama KYamaguchi Y amp Suzuki H (2009) Deciphering and dating the red pandarsquosancestry and early adaptive radiation of Musteloidea Molecular Phylogeneticsand Evolution 53(3)907ndash22 [AJJ]

Schneider S (2009) The nature of symbols in the language of thought Mind andLanguage 24(4)523ndash53 [SS]

Schneider S (forthcoming) The language of thought New philosophical directionsMIT Press [SS]

Schwanenflugel P J amp Shoben E (1983) Differential context effects in thecomprehension of abstract and concrete verbal materials Journal of Exper-imental Psychology Learning Memory and Cognition 982ndash102 [GD]

Segal G (2000) A thin book about narrow content MIT Press [GR]Shafto P Coley JD amp Baldwin D (2007) Effects of time pressure on context-

sensitive property induction Psychonomic Bulletin amp Review 14890ndash94[BKH]

Shallice T (1988) From neuropsychology to mental structure CambridgeUniversity Press [aEM]

Shepard R N amp Cooper L A (1982) Mental images and their transformationsMIT PressBradford Books [SH]

Shepard R N Hovland C I amp Jenkins H M (1961) Learning and memorizationof classifications Psychological Monographs General and Applied 75(13)1ndash42 (Whole No 517) [SSK]

Siegler R S (1994) Cognitive variability A key to understanding cognitive devel-opment Current Directions in Psychological Science 31ndash5 [HAV]

Simmons W K Ramjee V Beauchamp M S McRae K Martin A amp BarsalouL W (2007) A common neural substrate for perceiving and knowing aboutcolor Neuropsychologia 452802ndash10 [aEM]

Simon H A (1995) Machine as mind In Android epistemology ed K M Ford CGlymour amp P J Hayes pp 23ndash40 MIT Press [aEM]

Sloman S A (1993) Feature-based induction Cognitive Psychology 25231ndash80[aEM]

Sloman S A (1996) The empirical case for two systems of reasoning PsychologicalBulletin 1193ndash22 [aEM]

Sloman S A amp Lagnado D (2005) The problem of induction In The Cambridgehandbook of thinking and reasoning ed K Holyoak amp R Morrison pp 95ndash116 Cambridge University Press [aEM]

Sloutsky V M amp Fisher A V (2008) Attentional learning and flexible inductionHow mundane mechanisms give rise to smart behaviors Child Development79639ndash51 [HAV]

Smith E E Patalano A L amp Jonides J (1998) Alternative strategies ofcategorization Cognition 65167ndash96 [arEM]

Smith E E amp Sloman S A (1994) Similarity- vs rule-based categorizationMemory and Cognition 22377ndash86 [TB]

Smith J D (2002) Exemplar theoryrsquos predicted typicality gradient can be testedand disconfirmed Psychological Science 13437ndash42 [aEM SZ]

Smith J D amp Minda J P (1998) Prototypes in the mist The early epochs ofcategory learning Journal of Experimental Psychology Learning Memoryand Cognition 241411ndash36 [JJC JAH arEM SZ]

Smith J D amp Minda J P (2000) Thirty categorization results in search of a modelJournal of Experimental Psychology Learning Memory and Cognition 263ndash27 [JJC aEM]

Smith J D amp Minda J P (2001) Journey to the center of the category Thedissociation in amnesia between categorization and recognition Journal ofExperimental Psychology Learning Memory and Cognition 27984ndash1002[SZ]

Smith J D amp Minda J P (2002) Distinguishing prototype-based and exemplar-based processes in dot-pattern category learning Journal of ExperimentalPsychology Learning Memory and Cognition 28800ndash11 [SZ]

Smith J D Murray M J Jr amp Minda J P (1997) Straight talk about linearseparability Journal of Experimental Psychology Learning Memory andCategorization 23659ndash80 [SZ]

Smith L B amp Samuelson L K (1997) Perceiving and remembering Categorystability variability and development In Knowledge concepts and categoriesed K Lamberts amp D Shanks pp 161ndash96 MIT Press [aEM]

Smith L B Thelen E Titzer R amp McLin D (1999) Knowing in the context ofacting The task dynamics of the A-not-B error Psychological Review 106235ndash60 [HAV]

Sobel D M amp Kirkham N Z (2007) Interactions between causal andstatistical learning In Causal learning Psychology philosophy andcomputation ed A Gopnik amp L E Schulz pp 139ndash53 Oxford UniversityPress [YY]

Solomon K O amp Barsalou L W (2001) Representing properties locally CognitivePsychology 43129ndash69 [aEM]

Solomon K O amp Barsalou L W (2004) Perceptual simulation in property veri-fication Memory and Cognition 32244ndash59 [aEM]



Spencer J P amp Schoner G (2003) Bridging the representational gap in thedynamic systems approach to development Developmental Science 6392ndash412 [HAV]

Sperber D amp Wilson D (1995) Relevance Communication and cognition 2nd edBlackwell [EMar]

Sperber D amp Wilson D (1998) The mapping between the mental and the publiclexicon In Language and thought Interdisciplinary themes ed P Carruthersamp J Boucher pp 184ndash200 Cambridge University Press [aEM]

Spivey M (2007) The continuity of mind Oxford University Press [ABM]Strevens M (2000) The essentialist aspect of naıve theories Cognition 74149ndash

75 [DD]Stich S P (1983) From folk psychology to cognitive science MIT Press [aEM]Stich S P (1996) Deconstructing the mind Oxford University Press [aEM]Storms G De Boeck P amp Ruts W (2000) Prototype and exemplar based infor-

mation in natural language categories Journal of Memory and Language4251ndash73 [JAH]

Swaab T Baynes K amp Knight R (2002) Separable effects of priming andimageability on word processing An ERP study Cognitive Brain Research1599ndash103 [GD]

Tenenbaum J B Griffiths T L amp Niyogi S (2007) Intuitive theories as grammarsfor causal inference In Causal learning Psychology philosophy andcomputation ed A Gopnik amp L Schulz pp 301ndash22 Oxford UniversityPress [aEM]

Thelen E amp Smith L B (1994) A dynamic systems approach to the development ofcognition and action MIT Press [HAV]

Thompson-Schill S L (2003) Neuroimaging studies of semantic memoryInferring ldquohowrdquo from ldquowhererdquo Neuropsychologia 41280ndash92 [aEM]

Turing A M (1950) Computing machinery and intelligence Mind 49433ndash60 [SH]Uttal W R (2001) The new phrenology MIT Press [ABM]van Geert P amp van Dijk M (2002) Focus on variability New tools to study

intra-individual variability in developmental data Infant Behavior andDevelopment 25340ndash74 [HAV]

Van Orden G C Pennington B F amp Stone G O (2001) What do doubledissociations prove Cognitive Science 25111ndash72 [aEM]

Verguts T amp Fias W (2009) Similarity and rules united Similarity- and rule-basedprocessing in a single neural network Cognitive Science 33243ndash59 [YY]

Vigo R (2009) Categorical invariance and structural complexity in human conceptlearning Journal of Mathematical Psychology 53203ndash21 [SSK]

Vlach H A Sandhofer C M amp Kornell N (2008) The spacing effect in childrenrsquosmemory and category induction Cognition 109163ndash67 [HAV]

Wattenmaker W amp Shoben E (1987) Context and the recallability of concreteand abstract sentences Journal of Experimental Psychology 13140ndash50[GD]

Weisberg J van Turrennout M amp Martin A (2007) A neural system for learningabout object function Cerebral Cortex 17513ndash21 [rEM]

Weiskopf D A (2007) Concept empiricism and the vehicles of thought Journal ofConsciousness Studies 14156ndash83 [GD]

Weiskopf D A (2009a) Atomism pluralism and conceptual content Philosophyand Phenomenological Research 79130ndash62 [DAW]

Weiskopf D A (2009b) The plurality of concepts Synthese169(1)145ndash73 [ELAS DAW]

Weiskopf D A (2010) Concepts and the modularity of thought Dialectica64107ndash30 [DAW]

Weiskopf D A (forthcoming) The functional unity of special science kinds BritishJournal for the Philosophy of Science [DAW]

Whitney P McKay T Kellas G amp Emerson W A Jr (1985) Semanticactivation of noun concepts in context Journal of Experimental PsychologyLearning Memory and Cognition 20804ndash23 [arEM]

Williams J J amp Lombrozo T (in press) The role of explanation in discoveryand generalization Evidence from category learning Cognitive Science[TL]

Wisniewski E J amp Medin D L (1994) On the interaction of theory and data inconcept learning Cognitive Science 18221ndash81 [AS]

Yamauchi T Love B C amp Markman A B (2002) Learning nonlinearly separablecategories by inference and classification Journal of Experimental PsychologyLearning Memory and Cognition 28(3)585ndash93 [ABM]

Yamauchi T amp Markman A B (1998) Category learning by inference andclassification Journal of Memory and Language 39(1)124ndash48 [ABM]

Yaxley R H amp Zwaan R A (2007) Simulating visibility during language com-prehension Cognition 105229ndash36 [aEM]

Zaki S R amp Nosofsky R M (2004) False prototype enhancement effects in dotpattern categorization Memory amp Cognition 32390ndash98 [SZ]

Zaki S R amp Nosofsky R M (2007) A high-distortion enhancement effect in theprototype-learning paradigm Dramatic effects of category learning duringtest Memory amp Cognition 352088ndash96 [SZ]

Ziff E (1972) Understanding understanding Cornell University Press [aEM]




all higher cognitive competences when they result injudgments about x I call this characterization ldquoCrdquo

It is important to highlight some significant propertiesof concepts so understood First concepts can be aboutclasses of objects (eg DOG) events (eg GOING TO THE

DENTIST) substances (eg WATER) and individuals(eg BARACK OBAMA) Second concepts are non-proprie-tary DOG is used by the processes underlying categoriz-ation induction linguistic understanding metaphorbuilding planning and perhaps other competencesThird the elements of information that are constitutiveof a concept can vary over time and across individualsFourth it might be unclear whether a given element ofinformation about x belongs to a concept of x Finally con-cepts are used by default in the cognitive processes under-lying higher cognitive competences (I call ldquoDefaultrdquo thehypothesis that some bodies of knowledge are retrievedby default when one is categorizing reasoning drawinganalogies making inductions and so on) This entailsthat our concept(s) of say dogs is (are) only a subset ofour whole knowledge about dogs The knowledge that isconstitutive of DOG is the knowledge about dogs that isretrieved by default from long-term memory when wereason about dogs categorize things as dogs and soforth I call ldquobackground knowledgerdquo about dogs theknowledge about dogs that is not part of the concept(s)of dogs Our background knowledge about a category asubstance some kind of events and the like can becalled upon occasionally when the default body of knowl-edge is insufficient to solve a cognitive task

At this point it is useful to spell out the notion of beingused by default at greater length A default body of knowl-edge about x is the body of knowledge that is presump-tively taken to be relevant when one reasons about xwhen one categorizes things as x and so on The knowl-edge that is stored in a concept of x is preferentially avail-able when we think reason and so forth about x So tospeak it spontaneously comes to mind

The proposed characterization of the notion of conceptcaptures much of what is implicit in the use of the termconcept in cognitive science However it is also clear thatC is partly at odds with some characterizations of thenotion of concept found in the literature which are dis-cussed at length in Chapter 1 of Doing without ConceptsIn any case with the proposed characterization of thenotion of concept I do not merely aim at capturing theuse of concept in cognitive science I also want to regimentit I contend that some bodies of knowledge are retrievedby default from long-term memory when one reasonscategorizes and so forth (see section 3 for a defense ofthis claim) and I propose that the term concept shouldbe used to refer to these bodies of knowledge

2 Individuating concepts

It is certainly possible for a given individual to have severalconcepts of the same category (eg several concepts ofchairs) or of the same substance (eg several conceptsof gold) For instance one might think of chairs inseveral distinct ways each of which corresponds to adifferent concept of chairs This possibility raises thefollowing question which is examined in Chapter 3of Doing without Concepts What does it mean for twobodies of knowledge about x (eg the knowledge that

water is typically transparent and the knowledge thatwater is made of molecules of H2O) to be part of thesame concept of x (WATER) rather than of two distinctconcepts (WATER1 and WATER2)

I propose the following individuation criteria (respect-ively Connection and Coordination) When two elementsof information about x A and B fulfill either of thesecriteria they belong to two distinct concepts

1 If retrieving A (eg water is typically transparent)from long-term memory and using it in a cognitiveprocess (eg a categorization process) does not facilitatethe retrieval of B (water is made of molecules of H2O)from long-term memory and its use in some cognitiveprocess then A and B belong to two distinct concepts(WATER1 and WATER2)

2 If A and B yield conflicting judgments (eg the judg-ment that some liquid is water and the judgment that thisvery liquid is not water) and if I do not view either judgmentas defeasible in light of the other judgment (ie if I holdboth judgments to be equally authoritative) then A and Bbelong to two distinct concepts (WATER1 and WATER2)

Letrsquos clarify these two criteria a bit further Connectionunpacks the idea that the components of a given conceptmust be connected (or ldquolinkedrdquo as I say in Doing withoutConcepts) If the beliefs that water is typically transparentand that water is made of molecules of H2O are part of thesame concept WATER then using the first belief to classifysome sample as water enables me to conclude that it ismade of molecules of H2O Accessing one part of aconcept makes the other parts accessible and poised tobe used in cognitive processing (for related discussionsee Millikan 2000 Ch 10)

Coordination expresses the idea that the parts of a singleconcept should not result in conflicting judgments that areboth taken to be equally authoritative Coordination iscompatible with the parts of a given concept yielding con-flicting judgments provided that all but one are viewed asdefeasible To illustrate although the belief that grand-mothers have gray hair and the belief that grandmothersare mothers of parents would yield different judgmentsif they were used to categorize a young-looking grand-mother (eg Sarah Palin) they could still both be partof the concept GRANDMOTHER if one of the two judgments(Sarah Palin is not a grandmother) is defeated by the otherone (Sarah Palin is a grandmother) In effect the firstbelief would be treated as a heuristic that sometimesleads us astray

These two criteria merely explain what it is for a givenindividual to have one or several concepts about someclass of objects some substance and so on (within-person individuation) It does not explain what it is fortwo or more individuals to have the same concept of x(between-person individuation) Now one might expecta theoretician of concepts to provide individuation criteriafor this situation too However I doubt that these areneeded for the psychology of concepts since they seemto play no explanatory role in psychology

3 Defending the proposed notion of concept

Some cognitive scientists and philosophers of psychologyreject Default (viz the assumption that some bodies ofknowledge are retrieved by default when one is categoriz-ing reasoning drawing analogies and making inductions)









































































9 Neo-empiricism







































12 Open questions




















doi101017S0140525X10000324


08901-1107















doi101017S0140525X10000336




















doi101017S0140525X1000052X



ddankscmuedu



















doi101017S0140525X10000348

















doi101017S0140525X10000543


kedwar02syredu














doi101017S0140525X1000035X

















doi101017S0140525X10000361


United Kingdom

hamptoncityacuk















doi101017S0140525X10000567




harnadugamca




















doi101017S0140525X10000373


Australia

bhayesunsweduau
















doi101017S0140525X10000385


ajjacobsonuhedu















doi101017S0140525X10000555


frankkeilYaleedu













doi101017S0140525X10000397



19104



















doi101017S0140525X10000403



Bologna Italy
















doi101017S0140525X10000415


lombrozoberkeleyedu


















doi101017S0140525X10000427




















names is arbitrary


doi101017S0140525X10000439


markmanpsyutexasedu





















doi101017S0140525X10000440


georeyearthlinknet




















doi101017S0140525X10000452


samuels58osuedu

ferreira23osuedu
































doi101017S0140525X10000464


ascarantinogsuedu




















doi101017S0140525X10000579



PA 19104-6304

susansdrgmailcom



















doi101017S0140525X10000476


MI 48109-1043

humeanumichedu

bradmoorumichedu















doi101017S0140525X10000488


63121-4400

jlv4z3umsledu

piccininigumsledu


























doi101017S0140525X1000049X


CA 90095-1563

haleyvlachuclaedu

laurenkroghgmailcom

emilyt0623uclaedu

sandhofpsychuclaedu













doi101017S0140525X10000506


dweiskopfgsuedu

















doi101017S0140525X10000518



rakisonandrewcmuedu















doi101017S0140525X10000531





szakiwilliamsedu


jcruzwilliamsedu


jcruzjcruzhtml

















doi101017S0140525X10000932


Pittsburgh PA 15260

Macherypittedu

wwwpittedumachery


R1 Introduction





R2 Clarifications





























































R8 Hybrid theories












R9 Eliminativism


































References































































































































































































































































































































































































































9 Neo-empiricism







































12 Open questions




















doi101017S0140525X10000324


08901-1107















doi101017S0140525X10000336




















doi101017S0140525X1000052X



ddankscmuedu



















doi101017S0140525X10000348

















doi101017S0140525X10000543


kedwar02syredu














doi101017S0140525X1000035X

















doi101017S0140525X10000361


United Kingdom

hamptoncityacuk















doi101017S0140525X10000567




harnadugamca




















doi101017S0140525X10000373


Australia

bhayesunsweduau
















doi101017S0140525X10000385


ajjacobsonuhedu















doi101017S0140525X10000555


frankkeilYaleedu













doi101017S0140525X10000397



19104



















doi101017S0140525X10000403



Bologna Italy
















doi101017S0140525X10000415


lombrozoberkeleyedu


















doi101017S0140525X10000427




















names is arbitrary


doi101017S0140525X10000439


markmanpsyutexasedu





















doi101017S0140525X10000440


georeyearthlinknet




















doi101017S0140525X10000452


samuels58osuedu

ferreira23osuedu
































doi101017S0140525X10000464


ascarantinogsuedu




















doi101017S0140525X10000579



PA 19104-6304

susansdrgmailcom



















doi101017S0140525X10000476


MI 48109-1043

humeanumichedu

bradmoorumichedu















doi101017S0140525X10000488


63121-4400

jlv4z3umsledu

piccininigumsledu


























doi101017S0140525X1000049X


CA 90095-1563

haleyvlachuclaedu

laurenkroghgmailcom

emilyt0623uclaedu

sandhofpsychuclaedu













doi101017S0140525X10000506


dweiskopfgsuedu

















doi101017S0140525X10000518



rakisonandrewcmuedu















doi101017S0140525X10000531





szakiwilliamsedu


jcruzwilliamsedu


jcruzjcruzhtml

















doi101017S0140525X10000932


Pittsburgh PA 15260

Macherypittedu

wwwpittedumachery


R1 Introduction





R2 Clarifications





























































R8 Hybrid theories












R9 Eliminativism


































References



















































































































































































































































































































































































































9 Neo-empiricism







































12 Open questions




















doi101017S0140525X10000324


08901-1107















doi101017S0140525X10000336




















doi101017S0140525X1000052X



ddankscmuedu



















doi101017S0140525X10000348

















doi101017S0140525X10000543


kedwar02syredu














doi101017S0140525X1000035X

















doi101017S0140525X10000361


United Kingdom

hamptoncityacuk















doi101017S0140525X10000567




harnadugamca




















doi101017S0140525X10000373


Australia

bhayesunsweduau
















doi101017S0140525X10000385


ajjacobsonuhedu















doi101017S0140525X10000555


frankkeilYaleedu













doi101017S0140525X10000397



19104



















doi101017S0140525X10000403



Bologna Italy
















doi101017S0140525X10000415


lombrozoberkeleyedu


















doi101017S0140525X10000427




















names is arbitrary


doi101017S0140525X10000439


markmanpsyutexasedu





















doi101017S0140525X10000440


georeyearthlinknet




















doi101017S0140525X10000452


samuels58osuedu

ferreira23osuedu
































doi101017S0140525X10000464


ascarantinogsuedu




















doi101017S0140525X10000579



PA 19104-6304

susansdrgmailcom



















doi101017S0140525X10000476


MI 48109-1043

humeanumichedu

bradmoorumichedu















doi101017S0140525X10000488


63121-4400

jlv4z3umsledu

piccininigumsledu


























doi101017S0140525X1000049X


CA 90095-1563

haleyvlachuclaedu

laurenkroghgmailcom

emilyt0623uclaedu

sandhofpsychuclaedu













doi101017S0140525X10000506


dweiskopfgsuedu

















doi101017S0140525X10000518



rakisonandrewcmuedu















doi101017S0140525X10000531





szakiwilliamsedu


jcruzwilliamsedu


jcruzjcruzhtml

















doi101017S0140525X10000932


Pittsburgh PA 15260

Macherypittedu

wwwpittedumachery


R1 Introduction





R2 Clarifications





























































R8 Hybrid theories












R9 Eliminativism


































References







































































































































































































































































































































































































9 Neo-empiricism







































12 Open questions




















doi101017S0140525X10000324


08901-1107















doi101017S0140525X10000336




















doi101017S0140525X1000052X



ddankscmuedu



















doi101017S0140525X10000348

















doi101017S0140525X10000543


kedwar02syredu














doi101017S0140525X1000035X

















doi101017S0140525X10000361


United Kingdom

hamptoncityacuk















doi101017S0140525X10000567




harnadugamca




















doi101017S0140525X10000373


Australia

bhayesunsweduau
















doi101017S0140525X10000385


ajjacobsonuhedu















doi101017S0140525X10000555


frankkeilYaleedu













doi101017S0140525X10000397



19104



















doi101017S0140525X10000403



Bologna Italy
















doi101017S0140525X10000415


lombrozoberkeleyedu


















doi101017S0140525X10000427




















names is arbitrary


doi101017S0140525X10000439


markmanpsyutexasedu





















doi101017S0140525X10000440


georeyearthlinknet




















doi101017S0140525X10000452


samuels58osuedu

ferreira23osuedu
































doi101017S0140525X10000464


ascarantinogsuedu




















doi101017S0140525X10000579



PA 19104-6304

susansdrgmailcom



















doi101017S0140525X10000476


MI 48109-1043

humeanumichedu

bradmoorumichedu















doi101017S0140525X10000488


63121-4400

jlv4z3umsledu

piccininigumsledu


























doi101017S0140525X1000049X


CA 90095-1563

haleyvlachuclaedu

laurenkroghgmailcom

emilyt0623uclaedu

sandhofpsychuclaedu













doi101017S0140525X10000506


dweiskopfgsuedu

















doi101017S0140525X10000518



rakisonandrewcmuedu















doi101017S0140525X10000531





szakiwilliamsedu


jcruzwilliamsedu


jcruzjcruzhtml

















doi101017S0140525X10000932


Pittsburgh PA 15260

Macherypittedu

wwwpittedumachery


R1 Introduction





R2 Clarifications





























































R8 Hybrid theories












R9 Eliminativism


































References























































































































































































































































































































































































9 Neo-empiricism







































12 Open questions




















doi101017S0140525X10000324


08901-1107















doi101017S0140525X10000336




















doi101017S0140525X1000052X



ddankscmuedu



















doi101017S0140525X10000348

















doi101017S0140525X10000543


kedwar02syredu














doi101017S0140525X1000035X

















doi101017S0140525X10000361


United Kingdom

hamptoncityacuk















doi101017S0140525X10000567




harnadugamca




















doi101017S0140525X10000373


Australia

bhayesunsweduau
















doi101017S0140525X10000385


ajjacobsonuhedu















doi101017S0140525X10000555


frankkeilYaleedu













doi101017S0140525X10000397



19104



















doi101017S0140525X10000403



Bologna Italy
















doi101017S0140525X10000415


lombrozoberkeleyedu


















doi101017S0140525X10000427




















names is arbitrary


doi101017S0140525X10000439


markmanpsyutexasedu





















doi101017S0140525X10000440


georeyearthlinknet




















doi101017S0140525X10000452


samuels58osuedu

ferreira23osuedu
































doi101017S0140525X10000464


ascarantinogsuedu




















doi101017S0140525X10000579



PA 19104-6304

susansdrgmailcom



















doi101017S0140525X10000476


MI 48109-1043

humeanumichedu

bradmoorumichedu















doi101017S0140525X10000488


63121-4400

jlv4z3umsledu

piccininigumsledu


























doi101017S0140525X1000049X


CA 90095-1563

haleyvlachuclaedu

laurenkroghgmailcom

emilyt0623uclaedu

sandhofpsychuclaedu













doi101017S0140525X10000506


dweiskopfgsuedu

















doi101017S0140525X10000518



rakisonandrewcmuedu















doi101017S0140525X10000531





szakiwilliamsedu


jcruzwilliamsedu


jcruzjcruzhtml

















doi101017S0140525X10000932


Pittsburgh PA 15260

Macherypittedu

wwwpittedumachery


R1 Introduction





R2 Clarifications





























































R8 Hybrid theories












R9 Eliminativism


































References


























































































































































































































































































































































Documents

Pre´cis of Doing without Concepts