How Capuchin Monkeys (Cebus apella) Quantify Objects and ...How Capuchin Monkeys (Cebus apella)...

Journal of Comparative Psychology (2006) Vol. 120, Issue 4, pp. 416-426.©2006 American Psychological Association

How Capuchin Monkeys (Cebus apella) QuantifyObjects and Substances

Kristy vanMarle, Justine Aw, Koleen McCrink, & Laurie R. SantosYale University

Humans and nonhuman animals appear to share a capacity for nonverbal quantity representations.But what are the limits of these abilities? Previous research with human infants suggests that theontological status of an entity as an “object” or a “substance” affects infants’ ability to quantify it.We ask whether the same is true for another primate species--the new world monkey Cebusapella. We tested capuchin monkeys’ ability to select the greater of two quantities of eitherdiscrete objects or a nonsolid substance. Participants performed above chance with both objects(Experiment 1) and substances (Experiment 2); in both cases, the observed performance was ratio-dependent. This suggests that capuchins quantify objects and substances similarly and do so viaanalog magnitude representations.

A wealth of research on numericalcognition suggests that both human andnonhuman animals represent quantitynonverbally and use this information to guidetheir behavior. Preverbal human infants, forexample, reliably reason about numericalinformation in a variety of different tasks (seeFeigenson, Dehaene, & Spelke, 2004 andGallistel & Gelman, 2005 for review).

.This research was supported by Yale University.This work was approved by the Yale University IACUCcommittee and conforms to federal guidelines for the use ofanimals in research.

The authors would like to thank Miriam Bowring,Laura Edwards, Bryan Galipeau, Heidi Hansberry, NathanHerring, Iris Ma, Samantha Santos, Adena Schachner, andBrandon Schneider for their help in running these studies. Weare grateful to Rochel Gelman, Daniel Gottlieb, and threeanonymous reviewers for helpful comments on earlierversions of this manuscript.

Correspondence concerning this article should beaddressed to Dr. Kristy vanMarle, Rutgers Center forCognitive Science (RuCCS), Psych Bldg Addition, BuschCampus, 152 Frelinghuysen Road, Rutgers University-NewB r u n s w i c k , P i s c a t a w a y , N J , 0 8 8 5 4 o [email protected].

This article may not exactly replicate the finalversion published in the APA journal. It is not the copy ofrecord.

Similarly, numerical abilities havebeen documented in a wide range of non-linguistic species such as rats, pigeons,parrots, raccoons, ferrets, lemurs, monkeys,and apes (see Brannon & Roitman, 2003;Davis & Perusse, 1988; Dehaene, 1997;Gallistel & Gelman, 2000; and Nieder, 2005for review).

These nonverbal numerical abilitiesare quite general. Both human and nonhumananimals represent numerical informationregardless of whether the stimuli involveauditory or visual events (e.g., Hauser, Tsao,Garcia & Spelke, 2003; Jordan, Brannon,Logothetis, & Ghazanfar, 2005; McCrink &Wynn, 2004; Meck & Church, 1983; Whalen,Gallistel, & Gelman, 1999), objects in theworld (e.g., Hauser, MacNeilage, & Ware,1996; Wynn, 1992), or actions produced bythe animal (e.g., lever presses, Fernandes &Church, 1982; Mechner, 1958). In addition,rhesus macaques (Macaca mulatta) andcapuchin monkeys (Cebus apella) can orderpairs of stimuli on the basis of numerosity(Brannon & Terrace, 1998; 2000; Judge,

http://www.apa.org/journals/com/homepage.html

Kristy vanMarle

Line


Evans, & Nyas, 2005). Infants and someanimal species have also been observed toperform mental computations over theirnumber representations (e.g., Boysen &Berntson, 1989; Call, 2000; Wynn, 1992).Preverbal infants, for example, have beenshown to discriminate between correct andincorrect numerical results of large numberaddition and subtraction events shown on acomputer screen (i.e., 5 + 5 or 10 - 5 = 5 or10; McCrink & Wynn, 2004). Similarly,Brannon, Wusthoff, Gallistel, & Gibbon(2001) found that pigeons can be trained tomake a behavioral response on the basis of acomparison made between a standard numberand the number resulting from a numericalsubtraction. Additionally, human infants,chimpanzees and rhesus monkeys whoobserve an experimenter sequentially hidedifferent numbers of food items in twodifferent boxes will preferentially approachthe box containing the larger total number(Beran, 2001, 2004; Feigenson, Carey, &Hauser, 2002; Hauser, Carey, & Hauser,2000).

Much debate has surrounded thenature of the representations underlying thisnonverbal number capacity. Over the past fewdecades, researchers have proposed severaltypes of mechanisms for representing number.One prominent class of models includes thosein which quantity is represented via analogmagnitudes. Although these models comedifferent flavors (e.g., “accumulator model”,Meck & Church, 1983; “neural filteringmodel”, Dehaene & Changeaux, 19931), the

1 Recent research by Nieder, Freedman, and Miller(2002) and Nieder and Miller (2003) found single cellsin the primate prefrontal cortex that are tuned torespond to specific numerosities. The activation ofthese cells can explain scalar variability in behavioralnumerical discrimination data since the tuning curvesbecome wider as the preferred numerosity becomeslarger. According to Nieder and Miller (2004), aspects

signature property of all analog magnitudemodels is that the discriminability of twovalues depends on their proportionatedifference (i.e., ratio), rather than theirabsolute difference. Thus, it should be easierto discriminate 4 from 8, than 8 from 12, eventhough the values differ by the same numberof units in both cases. To date, a great deal ofevidence has been gathered in support ofanalog magnitude models. Researchers haveobserved ratio-dependent performance in avariety of nonhuman animal species (e.g.,Beran, 2001, 2004; Beran & Beran, 2004;Beran & Rumbaugh, 2001; Lewis, Jaffe, &Brannon, 2005; Mechner, 1958; Meck &Church, 1983; Nieder & Miller, 2004), aswell as in infant and adult humans (e.g.,Cordes, Gallistel, & Gelman, & Whalen,2001; McCrink & Wynn, 2004; Xu & Spelke,2000; Xu, Spelke, & Goddard, 2005).

Other researchers have proposed adifferent kind of model to explain numericalperformace: the object tracking mechanism(e.g., Kahneman, Treisman, & Gibbs, 1992;Pylyshyn, 1989). Object tracking mechanismsare not mechanisms for number processingper se, but instead operate as a series of visualattention processes consisting of a limitednumber of indexes that “point” to individualobjects in the world. Such pointers allow anorganism to keep track of objects as theymove through space and undergo occlusion.The signature property of this type of

of this research (finding parallel, rather than serialprocessing of numerosity) accord particularly well withDehaene and Changeaux’s (1993) “neural filteringmodel” in comparison to Meck and Church’s (1983)“accumulator model”. It should be noted that althoughthese cells fire for arrays of simultaneously presenteditems, it is unclear whether or not they respond to thenumerosity of sequentially presented stimuli. Meck &Church’s (1983) accumulator model, however, canexplain how numerosity might be represented for bothsimultaneous and sequentially presented stimuli.


mechanism is its limited capacity--it can onlytrack as many objects as it has indexes. Inadult humans, this limit appears to be aboutfour indexes (Pylyshyn, 1989; Pylyshyn &Storm, 1988).

Interestingly, human infants andvarious animals seem abstract the numerosityof sets smaller than four objects withconsiderable precision. These findings haveled some investigators to adopt an objecttracking account of these numerical abilities(Feigenson et al., 2002; Hauser et al., 2000;Simon, 1997). For instance, Simon (1997)proposed that infants in Wynn’s (1992)“addition/subtraction” task were not doingarithmetic (as Wynn had originallysuggested), but rather were detectingviolations of object physics (i.e., that objectscannot spontaneously appear or disappear).The task is as follows: in the additioncondition, infants first see one object placedon a stage. A moment later, a screen is raisedto hide the object and then a second object isbrought onto the stage and placed behind thescreen. Finally, the screen is removed toreveal either two objects (the correct number)or one object (an incorrect number). Infants inthis situation look longer when the incorrectnumber of objects is revealed.

On Simon’s (1997) and similaraccounts, the object tracking system in thissituation deploys an index for each object,such that it represents both objects as being‘behind the screen’. When the screen isremoved, the active indexes are placed in one-to-one correspondence with the revealed set.A mismatch between the number of indexescurrently deployed and the number of itemsvisible is thought to account for the fact thatinfants, monkeys, and lemurs look longer atnumerically incorrect compared tonumerically correct outcomes in situationslike this (Hauser et al., 1996; Santos, Barnes,

& Mahajan, 2005; Simon, 1997; Wynn,1992).

Previous research with both infantsand nonhuman primates has examined whichof these two types of mechanisms--analogmagnitude representations or object trackingmechanisms--underlies nonverbal numberprocessing. Much of this work has sought torule out one mechanism or the other as anexplanation for animals’ ability to respond onthe basis of number. Recently however, agrowing body of evidence has begun tosupport the idea that human and nonhumananimals may employ both of the proposedmechanisms for enumeration, albeit indifferent circumstances (see Flombaum,Junge, & Hauser, 2005; Hauser & Carey,2003). The new objective for work innumerical cognition, then, is to examine theconditions under which one or both of themechanisms are engaged in a given task andwhy, which in turn provides informationabout the nature and limits of each of the twomechanisms.

One recent proposal is that analogmagnitudes and object indexes operate overdifferent parts of the number range (e.g.,Feigenson et al., 2004; Hauser & Spelke,2004; Xu, 2003). Clearly, this could be truefor the object tracking mechanism, which bydefinition is limited to representing small sets.A more striking proposal is that analogmagnitudes represent only large values (i.e.,more than four). Limited evidence for thiscomes from studies of infant numericalprocessing2. For example, human infants failto discriminate small values at a 1:2 ratio 2Note however, that evidence from human adults(Cordes et al., 2001) shows that the accumulatorrepresents both large and small values. When asked topress a key the same number of times as a target value,there was a constant coefficient of variation within thedistribution of responses across both the small (four orfewer) and large number ranges (five or more).


(e.g., 1 vs. 2 dots: Xu, 2003), even thoughthey readily discriminate larger values at a 1:2ratio (e.g., 8 vs. 16 dots: Xu, 2003; Xu &Spelke, 2000; Xu et al., 2005). Convergingevidence comes from infants’ performance inan ordinal choice task. When asked to choosebetween two hidden quantities of grahamcrackers, 10-month-old infants reliably chosethe container with the larger amount (e.g., 2crackers over 13, 3 crackers over 2, seeFeigenson et al., 2002). Performance fell tochance, however, whenever there were morethan 3 crackers hidden in a given container(e.g., 3 vs. 4, 1 vs. 4). This was true eventhough the ratio between the quantities shouldhave been large enough to allow infants tojudge which container had more if analogmagnitudes were being used (i.e., 1 vs. 4crackers). To account for this ‘set sizesignature’, researchers concluded that infantsuse the object tracking system alone to tracksmall sets of objects, and that in infants thismechanism is limited to tracking only 3objects (Feigenson & Carey, 2003; Feigensonet al., 2002).

A second proposal concerningdifferences in the use of analog magnitudesand object indexes involves the nature of theun i t s over which these two types ofmechanisms operate. Previous researchinvestigating object tracking processes in

3 Differential performance with small sets (failing todiscriminate one vs. two in habituation studies, butsuccessfully choosing two crackers over one in theordinal choice task) can be explained by the fact thatthe relevant habituation studies (Xu, 2003; Xu &Spelke, 2000) controlled for continuous properties ofthe displays so that number was the only dimensionavailable as a basis for discrimination. Studies usingthe ordinal choice task, in contrast, generally do notcontrol for continuous properties such that they areconfounded with number. Under these conditions,infants successfully discriminate between small valuesas long as the number of items in a set does not exceedthree.

adults suggests that this system might belimited to tracking rigid, cohesive objects(vanMarle & Scholl, 2003). Using a standardMultiple Object Tracking task (Pylyshyn &Storm, 1988), adults were asked to track 4 outof 8 identical items as they moved around acomputer screen. When the items stoppedmoving, participants had to report which ofthe eight items were the original targets.Adults performed at ceiling when the itemswere rigid objects, but were near chance whenthe items were non-cohesive substances that“poured” from location to location. Takentogether with studies showing that adults havedifficulty tracking other non-object entities(e.g., parts, Scholl, Pylyshyn, & Feldman,2001), this suggests that the natural units forthis mechanism are bounded, cohesiveobjects.

The same limits appear to affectinfants’ tracking capacities. Infants canreadily track small numbers of discreteobjects, but not small numbers of nonsolidsubstances4 (i.e., piles of sand, Huntley-Fenner, 1995; Huntley-Fenner, Carey, &Solimando, 2002).Specifically, using a modified paradigm basedon Wynn’s (1992) addition/subtraction task,Huntley-Fenner et al. (2002) comparedinfants’ ability to track either rigid objects ornonsolid substances (i.e., piles of sand).Consistent with previous results, infants inthis study looked reliably longer when ‘oneobject’ + ‘one object’ was shown to equalonly ‘one object’ (a physically impossible and

4 Throughout this paper, we shall use the term“substance” to refer exclusively to nonsolid substancessuch as sand or water, rather than solid substances suchas wood or metal. This further distinction betweensolid and nonsolid substances is orthogonal to thecontrast of interest here since solid substances, likewood, though they have no inherent form, maintaintheir form under movement, making them more likediscrete objects than nonsolid substances.


numerically incorrect outcome). In contrast,when piles of sand were used instead ofbounded, cohesive objects, infants lookedequally long regardless of how many pileswere revealed following the 1 + 1 operation.Thus, not only did they fail to notice thatthere were only “half as many” piles as thereshould have been, they also failed to noticethat there was only “half as much” sand asthere should have been.

Interestingly, however, infants canquant ify substances under somecircumstances. vanMarle (2004) tested 10- to12-month-old infants’ ability to select thelarger of two hidden quantities of eitherdiscrete objects (graham crackers) or portionsof substance (Cheerios). The object conditioninvolved the sequential lowering of individualcrackers into two opaque cups. The substancecondition differed in two ways thatemphasized the substance-like nature of theCheerios. First, each amount of Cheerios waspresented on a plate as a single, boundedportion of stuff (i.e., the individual Cheerioswere bunched together so that they formed agroup with a single bounding contour).Second, each portion was poured into anopaque cup such that its non-cohesivenesswas made salient. Results indicated that in theobject condition, infants readily selected 2crackers over 1 (replicating previous findings,Feigenson et al., 2002). In the substancecondition, however, an interesting patternemerged. Infants who were given a choicebetween two quantities of Cheerios thatdiffered by a 1:2 ratio (10 v 20 Cheerios)performed at chance. In contrast, those givena choice between quantities differing by a 1:4ratio (5 v 20 Cheerios) reliably chose thelarger amount. Since the food in this caseviolated cohesion (a property known todisrupt visual tracking in human adults), andsince the number of individual grains ofcereal in each portion clearly exceeded the

proposed capacity limits of the object trackingmechanism, these data suggest that infantsmay be able to use analog magnitudes torepresent and compare substance quantities(vanMarle, 2004)5. A different patternemerged when infants were given a choicebetween small portions in which the numberof individual Cheerios in each portion waswithin or just outside the set size limit.Specifically, infants reliably chose 2 Cheeriosover 1 (as they had done with crackers), butperformed near chance when given a choicebetween 1 and 4 Cheerios. This “set sizesignature” indicates that infants may havebeen using object indexes when faced withsmall portions of Cheerios and thus weretreating the small portions very differentlyfrom the large portions where they succeededat a 1:4 ratio, but not a 1:2 ratio (vanMarle,2004).

So far, we have seen evidencesuggesting that infants may use object indexesto track small numbers of objects, and analogmagnitude representations to quantify largenumbers and substances. But what aboutnonhuman animals? Interestingly, while thereis a great deal of evidence supporting the useof analog magnitudes in both animals andhumans, there is less evidence for the objecttracking mechanism in nonhuman animals.This is partly because the idea of an objecttracking mechanism was only recentlyborrowed to explain human infants’ and

5 The reader may be skeptical that infants were treatingthe Cheerios portions as substances rather than a largecollection of objects. However, note that if they wereconstruing them as objects, then they should havesucceeded at a 1:2 ratio, which even much youngerinfants can discriminate (e.g., McCrink & Wynn, 2004;Xu & Spelke, 2000; Xu et al., 2005). The fact that theyrequired a 1:4 ratio to succeed with large portions ofCheerios suggests that they were in fact construingthem as portions of substance rather than as a largecollection of discrete objects.


nonhuman animals’ performance on tasksinvolving small numbers of objects. However,it is also the case that many studies withnonhuman animals reveal a ratio signature,rather than a set size signature, across both thesmall and large number ranges (e.g., Beran,2001, 2004; Beran & Beran, 2004; Beran &Rumbaugh, 2001; Lewis et al., 2005; Nieder& Miller, 2004). For example, using acomputer joystick, chimpanzees (P a ntroglodytes) were asked to “collect dots” oneat a time until they had reached the numberindicated by an Arabic numeral displayed atthe beginning of the trial (target numerals 1through 7). The chimps were above chancewith all values, and performance wasconsistent with analog representations ofnumber; the larger the quantity to be matched,the worse the chimps performed (Beran &Rumbaugh, 2001). Similar results wereobtained using a different task in whichchimpanzees saw different numbers ofM&M’s sequentially placed into two cups.Here, chimpanzees chose the greater numberof candies significantly more often thanchance, even when it contained up to ninecandies. Again, performance was ratio-dependent. As the proportional differencebetween the quantities increased, so didperformance. The same was true whenmultiple sets of candies were hidden in thecups, such that the chimps had to performaddition in order to choose the quantity withthe greatest total number of candies (Beran,2001, 2004).

However suggestive, the chimpanzeesin these studies received hundreds of trialsand, consequently, may have beenrepresenting quantity differently than theywould in the wild. One way to avoid this issueis to try and observe nonhuman animals’spontaneous number abilities. Hauser et al.(2000) did just this by giving free-rangingrhesus monkeys a choice between two hidden

quantities of discrete food items. Monkeysreliably chose the box containing the largernumber of food items for a variety ofcomparisons including one vs. two, two vs.three, and three vs. four, but chose randomlywhen offered four vs. six, three vs. eight andfour vs. eight. Hauser and colleagues arguedthat this set-size limit demonstrated thatrhesus monkeys were using object indexes,rather than accumulator representations, todetermine which container had the most food.Increasing the ratio between the quantities didnot always increase performance. Althoughmonkeys reliably chose five over three fooditems, they were at chance on discriminationsof four vs. eight and three vs. eight,suggesting that monkeys, like infants,generally failed to use accumulatormagnitudes representations in this trackingtask, even for highly discriminable values inthe large number range (Hauser et al., 2000).

At present, the Hauser et al. (2000)tracking task provides the only strongevidence to date that a nonhuman animalrepresents and discriminates between smallnumbers of objects using object indexes.Unfortunately, however, this study does notfully rule out the possibility that the monkeyswere using analog magnitudes in this task.First, the design of these studies preventsthem from providing information about whatdiscrimination function obtains in this task;because each monkey was tested only once,there is no way to see if the rhesus monkeyparticipants showed subtle differences inperformance that depended on the ratiocomparison. Moreover, since differentanimals participated in different conditions,meaningful comparisons cannot be madeacross conditions. The present study aims toaddress this issue by examining how anotherprimate species, the capuchin monkey,performs in a similar ordinal choice task.


The present set of experiments wasdesigned with three goals in mind. First, wewere interested in documenting numericalabilities in another primate species which todate has largely been neglected in studies ofnumerical cognition. Second, we wanted toexamine whether evidence for analogmagnitudes could be revealed in a task similarto that used to test nonverbal numericalabilities in other nonhuman primates andhuman infants. Third, we wanted to testwhether monkeys, like infants (vanMarle,2004), are able quantify substances in thistask and if so, what type of mechanismunderlies this ability.

To establish more sensitive measuresthan Hauser et al. (2000), we tested monkeyson multiple trials and in multiple conditions.Such a within-subject design allowed us toreveal potentially subtle differences inperformance across different ratiocomparisons and to make more meaningfulcomparisons across conditions. To minimizeany effect of training, the monkeys werenever trained on test conditions and were notgiven any feedback as to whether they hadmade an appropriate choice.

Our first experiment was designed toestablish simply whether or not this speciescould choose the greater of two hiddenquantities of food in an ordinal choice task.Experiment 2 then went on to explore whethercapuchins can quantify non-object entities andif so, what mechanism may underlie thisability.

Experiment 1

Method

Participants. We tested 6 brown capuchinmonkeys (Cebus apella). Capuchins are large NewWorld primates (see Fragaszy, Visalberghi, & Fedigan,2004 for an excellent survey of this species’ ecologyand behavior). Unlike other New World species,

capuchins are an extremely dexterous species and areknown to be adept tool-users both in the wild and incaptivity. Our participants were housed at theComparative Cognition Laboratory at Yale University(New Haven, CT); participants were born in captivityeither at the present facility or at the Living LinksCenter capuchin colony in Yerkes Regional PrimateResearch Center (Atlanta, GA). Our participant groupconsisted of three males (FL, AG, NN) and threefemales (HG, JM, MD) ranging in age from 3y4m to9y1m, with a mean age of 6y5m at the time of testing.All animals were housed together in a large socialenclosure filled with toys, swings, and naturalbranches. The ambient temperature in the mainenclosure (and thus during testing) remained relativelystable and was approximately 85° F with 85%humidity. Animals had ad libitum access to water andwere fed a diet of chow and fruits supplemented by thefood treats they receive during testing. All participantshad previously participated in experiments concerningphysics, social cognition, and tool use; theseexperiments sometimes involved reaching for differentnumbers of objects, but never involved representingdifferent numbers of occluded objects, as was requiredin the present study.

Materials. The experiments were conducted ina cubic enclosure (82.5 cm3) elevated 76 cm from thefloor and attached to the main enclosure. The walls ofthe experimental enclosure were made of wire mesh.After entering the experimental enclosure, a Plexiglasdoor was closed behind the monkey. The panel facingthe experimenter was made of wire mesh and includeda platform on the inside of the enclosure (25.4 cmabove the panel floor) on which the monkeys couldstand (see Figure 1). The panel had two openings (5 cmhigh x 9 cm long), spaced such that the participantscould reach through one, but not both, of the openingsat the same time, (approximately 25 cm apart).Attached to this panel (on the outside of theexperimental enclosure) was a square wooden frame(61 cm x 61 cm) with two sets of rails. These railssupported an acrylic tray (58.5 cm long x 30.5 cmwide) that served as a presentation platform. When thetray was placed on the lower rails, participants wereable to view, but not reach the tray. When it was liftedto the upper rails, participants could easily reach thequantities on it through the hand holes.

To ensure that the quantities were presented inthe same place on each trial and that they would beaccessible through the hand holes in the panel, twowhite plastic plates (15.25 cm in diameter) weresecured on the front half of the tray acting as place


markers. The plates were covered with grey duct tapeso the food items placed on them would be visible.This also minimized any potential auditory cues madeby the sound of the food items dropping onto theplates6. To facilitate comparison to subsequentexperiments, two red plastic cups (~4” tall and 3” indiameter at the opening) were used to cover food itemsby placing them upside-down over the quantities. Themonkeys could not see the food items once hidden.

Figure 1. An experimenter’s eye view of the testing chambersetup. The Plexiglas tray with two plastic plates serving asplace markers was presented on the wood frame. The two redplastic cups were movable and used to occlude food itemsduring presentations.

We used yogurt raisins7, a highly preferredfood with which the participants had prior experience.Since successful performance in our task requiredselecting the larger of two amounts of food, we wantedto use highly desirable food items to increase thechance that the monkeys would attempt to obtain thelarger amount. All test sessions were videotaped usinga Sony Digital Handycam (DCR-TRV140).

6 In fact, it is highly unlikely that monkeys could haveused the sound of food items dropping onto the platesfor two additional reasons. First, our food items had ayogurt coating that softened within 2-3 minutes ofbeing brought into the testing area due to the high heatand humidity, so that they made little sound uponlanding on the plates. Additionally, since the testingtook place inside the same room in which the monkeyswere housed, there was a substantial amount ofambient noise (e.g., almost constant vocalizations ofthe monkeys, noise generated by monkeys leapingfrom swing to swing and climbing around theenclosure, etc.) that would have masked any potentialnoise made by the dropping of the raisins.7 Yogurt raisins are raisins that have been coated in acreamy yogurt confection. They are a common snackfood in the USA.

Design and Procedure. Participants weretested individually inside the experimental enclosure.At the beginning of each session, the participantentered the experimental enclosure for a food treat (apeanut). The experimenter then closed the Plexiglaspanel behind the participant, securing them in theenclosure. This was done to ensure that the othermonkeys could not interrupt the test participant.

Participants stood or seated themselves on theplatform in the presentation area. Each sessionconsisted of 10 test trials. At the start of each trial, theexperimenter faced the monkey and placed the tray onthe lower rails. The experimenter then tapped the cupson the tray to draw the monkey’s attention and thenplaced them upside-down, one on each plate. Each cuphad a small hole poked through the bottom so that theraisins could be dropped into them, but the monkeycould not view the resulting quantity. Raisins were thendropped into a cup while the monkey watched. Foreach placement, the experimenter retrieved a singleyogurt raisin from a small container, held it up to theparticipant, and then dropped it onto the plate throughthe hole, saying “Look” followed by the participants’name. To ensure the monkeys saw each placement, theexperimenter did not place a yogurt raisin into the cupunless the monkey was attending to the event. Since itwas not uncommon for monkeys to look away duringthe presentation, the amount of time it took to hide aparticular number of raisins and the rate at which theraisins were hidden often varied. Thus, although‘duration of presentation’ and ‘rate of presentation’were generally confounded with ‘number of items’,they were at best only partially reliable cues.Nonetheless, it is possible that monkeys could haveused these cues on some of the trials.

One quantity was presented first and then theother with the order (larger-first or smaller-first)counterbalanced across trials. Once all the raisins weredropped into the cups, the experimenter tapped thecenter of the tray frame until the monkey centered itselfbetween the two hand-holes. Once the participantcentered himself or herself, the experimenterimmediately lifted the tray onto the upper rack, withinreach of the participant. To avoid cuing the monkeys inany way, the experimenter always lifted the tray in thesame manner and looked directly ahead (rather than atthe tray or either of the two cups). Monkeys indicatedtheir choice by reaching through one of the hand-holes,knocking over the cup and obtaining the raisins on thatplate. After the participant obtained the selectedquantity, the tray was immediately removed (out of theparticipants’ reach) and replaced on the lower rack.


Finally, the unselected pile was uncovered, revealingthe unselected quantity underneath, which was placedback into the original container. Participants wereallowed only one choice per trial.

Because participants had never participated ina choice experiment of this nature before, we beganwith an initial training condition in which participantswere given a choice between one raisin and zeroraisins. This initial condition ensured that participantsunderstood how to make choices and could easilydisplace the cups to obtain the yogurt raisins. Becausethe comparison was between one and zero raisins,participants were reinforced only when they made acorrect choice; participants who mistakenly chose theplate with zero raisins received no food. Participantswere required to achieve a criterion of 80% correct fortwo consecutive sessions in order to move on to thefirst test condition.

Once participants had completed this initialtraining condition, they received an initial one vs. twotest session consisting of ten trials. In this test session,participants were presented with a choice between oneraisin and two raisins. Because test sessions wereaimed at exploring how participants responded to aparticular comparison in the absence of training,participants were allowed to eat whichever quantitythey chose. Thus, correct choices were differentiallyrewarded only to the extent that they provided doublethe amount of food as was obtained on the incorrectchoices; wrong choices were penalized only in thesense that participants did not obtain the largestpossible food reward.

After participants had completed this initialone vs. two test session, they continued on toadditional testing sessions, each of which involved oneof the three other numerical comparisons ofinterest—either one vs. four, two vs. three, and threevs. four. Again, each test session consisted of 10 trials.The order in which these three test sessions werepresented was counterbalanced across monkeys. As inthe initial one vs. two test session, participants were notdifferentially reinforced during these additional testsessions—participants received whichever pile ofraisins they chose during test trials, even if this was theincorrect smaller amount.

Unfortunately, because both choices led tosuccessful retrieval of some amount of food in theseadditional test sessions (i.e., incorrect choices were stillrewarded), we worried that participants might begin toignore the actual numerical comparisons and eitherresort to choosing randomly or develop side biases. Toreduce the possibility of such alternative strategies, we

ran participants on interim training sessions in betweenall of these additional test sessions, ensuring they knewthat the task was to always choose the larger amount.These interim training sessions presented participantswith a choice of one vs. zero (the same comparisonused in the initial training session). Note thatparticipants only obtained the raisin when theycorrectly chose the larger amount of one raisin. As inthe initial training conditions, participants wererequired to achieve a criterion of 80% correct for twoconsecutive sessions before moving on to the next testcondition.

Results

All 6 participants performed at 100%across the first two sessions of their initialtraining phase, allowing them to continueonto the initial one vs. two test session.Participants performed above chance (Mean+/- SD= 75 +/- 14%) on this initial testcondition; on average, all 6 monkeys chosetwo raisins over one raisin (t(5) = 4.44, p <0.003).

Participants then moved onto theadditional test conditions. Mean +/- SDpercent correct was 85 +/- 10% (CI8 = 77-93%), 65 +/- 12% (CI = 55-75%), and 57 +/-10% (CI = 48-65%) for one vs. four, two vs.three, and three vs. four, respectively (seeFigure 2). T-tests compared performance oneach comparison against chance performance(50%). Performance was significantly betterthan chance in the one vs. four condition (t(5)= 8.17, p = 0.0002) and the two vs. threecondition (t(5) = 3.00. p = 0.02), but was onlymarginally above chance in the three vs. fourcondition (t(5) = 1.58, p = 0.09). We thenperformed a repeated measures ANOVA withcondition (one vs. four, two vs. three, andthree vs. four) as a within subject factor. Weobserved a significant effect of condition

8 All confidence intervals (CIs) reported in this paperwere computed with 95% confidence limits.


(F(2,10) = 8.01, p = 0.008, η2G = 0.439).

0

20

40

60

80

100

PER

CEN

T C

OR

REC

T

1 vs 4 2 vs 3 3 vs 4 1 vs 4 2 vs 3 3 vs 4

OBJECTS SUBSTANCES

Column 7

Column 6

Column 5

Column 4

Column 3

Column 2

Column 1

Figure 2. Mean percentage correct (including 95%confidence intervals) across different numerical comparisonsin object (Experiment 1) and substance (Experiment 2)conditions.

Participants performed significantlydifferently across the three test comparisons,with highest performance on thediscriminations involving the largest the ratiobetween the two quantities. Given the strongeffect of condition, we wished to furtherexplore the hypothesis that performance in thetest conditions was ratio-dependent. To do so,

9 As a measure of effect size, we used η2

G (generalizedeta squared). To give the reader a sense of how thismeasure should be interpreted, consider it just like anyother correlation coefficient--it signifies whatproportion of the variation can be accounted for by theindependent variable (Howell, 1997). For example, ifη2

G = .43, as it does for the effect of ratio condition inthe Experiment 1, then 43% of the variation inperformance was attributable to the ratio condition.This measure is recommended over eta squared andpartial eta squared for effects obtained in analyseswith repeated measures because eta squared and partialeta squared can give biased estimates of the effect sizethat are larger than would be obtained in the samestudy using a between subjects design (Olejnik &Algina, 2003). This would defeat the purpose ofproviding effect size measures in the first place since itwould prevent one from comparing effect sizes acrosssimilar studies using different designs.

we performed a correlation between the ratiopresented (expressed as the proportionalsimilarity between the two numberspresented: one vs. four = 25%, two vs. three =67%, and three vs. four = 75%) andparticipants’ performance. We observed arobust negative correlation of r = - 0.75, t(16)= 4.60, p < 0.0003); as the proportionalsimilarity between the numbers increased,participants’ performance decreased.

Non-parametric tests revealed asimilar pattern. We used a binomial test tomeasure the reliability of each monkey’sperformance in each test condition (one vfour, two v three, and three v four--see Figure3). Results revealed that the number ofmonkeys reliably choosing the larger amountfor the one v four, two v three, and three vfour comparisons, respectively, was 5/6, 2/6,0/6. Thus, despite substantial individualdifferences, monkeys’ performance againappeared to be ratio-dependent. Monkeys’performance was better the large theproportional difference between the amounts.

Discussion

Like chimpanzees (Beran, 2001),rhesus monkeys (Hauser et al., 2000) andhuman infants (Feigenson et al., 2002),capuchin monkeys reliably chose the greaterof two discrete quantities of food objects inour sequential presentation task. Even on theirfirst test condition of one vs. two, participantschose the larger reward reliably above chance.Capuchins continued to perform above chanceat other discriminations, performing reliablyabove chance when discriminating one vs.four and two vs. three, but not three vs. four.

In contrast to previous studies withinfants (Feigenson et al., 2002) and rhesusmacaques (Hauser et al., 2000), however,capuchins’ performance in this experiment


seemed to depend not on the set-size of the

Figure 3. Number of trials (out of 10) in which individualmonkeys chose the larger quantity of discrete food objects foreach ratio condition in Experiment 1. Asterisks indicate thatperformance was significantly above chance (Binomial Test,p < .05).

two numerical comparisons, but instead onthe ratio between the quantities presented. Asthe ratio between the two quantitiesapproached 1:1, participants’ performancedecreased. Participants performed best on aone vs. four discrimination and worst on athree vs. four discrimination. Note that thispattern cannot be due to the overall number ofitems present in the display, as participants’performance on the one vs. fourdiscrimination was reliably better than on thetwo vs. three discrimination, even thoughboth of these comparisons involve five raisinsin total.

Our results therefore contrast withthose of human infants (Feigenson et al.,2002) and rhesus monkeys (Hauser et al.,2000) where performance was limited by thenumber of items in each bucket (the set size),rather than the ratio between the quantities.Our results are, however, consistent withrecent research by Beran (2001, 2004) inwhich chimpanzees’ tendency to choose thelarger of two quantities of M&M’s was foundto be dependent on ratio, not set size,regardless of whether the number of objects inas set was inside or outside the object tracking

limit. This ratio signature suggests that analogmagnitudes might underlie performance inthis task. An account in which only objecttracking was hypothesized, in contrast, wouldhave predicted equivalent performance acrossthe different comparisons until the set sizelimit was reached, at which point performanceshould have fallen to chance. The fact that ourmonkeys were only marginally above chancein the three vs. four comparison could beindicative of a set size limit of three incapuchins. However, the fact that theyperformed well above chance in the one vs.four comparison argues against thispossibility. If capuchins were limited torepresenting only three items in each cup,then performance in three vs. four and one vs.four comparisons should have beenequivalent. Nevertheless, the effect of ratio inthe small number range is clear in thisexperiment and is inexplicable on an objecttracking account. Thus, our data areconsistent with the use of analog magnitudesto represent the quantities.

Given that capuchins apparently useanalog magnitudes to quantify discreteobjects, it is of interest to ask whether andhow their quantification abilities extend toentities that cannot be enumerated usingobject indexes. Can capuchins enumeratenon-objects , like substances? Previousresearch in both infants (Huntley-Fenner etal., 2002) and adults (vanMarle & Scholl,2003) suggests that the object trackingmechanism is sensitive to the object status ofentities to be tracked. That is, both infants andadults have relative difficulty trackingnonsolid substances compared to discreteobjects. Our difficulty enumeratingsubstances is evident in natural language;English, for example, employs differentsyntax and morphology for objects andsubstances, using count noun syntax forobjects and mass noun syntax for substances.


As suggested by the term “count noun,”discrete objects can be counted, whilesubstances cannot. Thus, it is appropriate tosay “three bottles”, but not “three sands”.Conversely, while it is acceptable to say“some sand”, it is not acceptable to say “somebottle”. Children become sensitive to thisdistinction at a very early age, andsubsequently use it to infer the appropriatereferent of a count or mass noun (Soja, 1992;Soja, Carey, & Spelke, 1991; see also Hall,1996; Prasada, 1999). Given the stark contrastbetween how humans appear to processobjects and substances in these variousdomains, we wished to ask whether the sameis true for other primate species.

The next experiment tested capuchins’ability to choose the larger of two portions ofa continuous substance--banana puree. Forpurposes of comparison to Experiment 1, weagain ran an initial one vs. two comparison,followed by three further comparisonconditions that were matched in ratio to theprevious experiment--one vs. four, two vs.three, and three vs. four.

Experiment 2Method

Participants. The same 6 monkeys fromExperiment 1 were tested.

Materials. We conducted the experiment inthe same enclosure as before, the only difference beingthat instead of yogurt raisins, the quantities consisted ofamounts of banana puree. The puree was madeimmediately before each session by blending two ripebananas with about 1/8 cup of water until it was asmooth, pourable consistency.

Design and Procedure. The design andprocedure was identical to Experiment 1 with thefollowing exceptions. Instead of placing raisins one byone into the cups, equal-sized scoops (one scoop wasone-half of a miniature plastic egg, approximately 1.5”in diameter and 1” long) of banana puree were drawnone at a time from a small container and visibly pouredinto the cups from a height of about 5 cm. Eachpouring event took approximately 2 seconds. As with

discrete quantities, there were no explicit controls for‘duration of presentation’ or ‘rate of presentation’ andso they were generally confounded with overallamount. However, as before, they were less reliablecues than overall amount since there were again sometrials in which the monkeys looked away duringpresentation causing the rate to vary and overallduration to increase for that quantity. Each scoop, for agiven quantity, was poured directly on top of the last sothat the final quantity consisted of a single portion.Once both quantities had been poured, the tray wasimmediately placed on the upper rails and the monkeywas allowed to choose one of the quantities. As before,the experimenter avoided cuing the monkeys bylooking directly ahead while lifting the tray. A choicewas made when the monkey knocked one of the cupsover and scooped up the banana puree. As before, thequantities were presented sequentially with side (left orright) and order (larger number placed first or second).

After completing an initial one vs. zerotraining phase, all monkeys began with the one vs. twoTest condition, followed by the remaining TestConditions (one vs. four, two vs. three, and three vs.four) in the same order they received them in the firstexperiment. Additional one vs. two Training sessionswere presented between each test condition until themonkey got 80% correct in a single session. Trainingsessions were identical to the one vs. two test conditionexcept the tray was pulled away if the monkey reachedfor the smaller quantity.

Results

All participants performed well abovechance (Mean = 90%) across the first twosessions of their initial one vs. zero trainingphase. All 6 participants therefore moved ontothe first one vs. two test. Participantsperformed above chance (58 +/- 8%) on thisfirst test condition, choosing two pouredbanana scoops over one scoop (t(5) = 2.71, p< 0.02). However, despite that fact thatparticipants were above chance on this one vs.two substance condition, their performancewas reliably worse than that of the one vs.two object condition of Experiment 1 (58%vs. 75%; t(5) = 2.99, p = 0.03).

Participants then moved on to theadditional comparisons. Average percent


correct was 77 +/- 12% (CI = 67-86%), 68 +/-18% (CI = 54-83%), and 66 +/- 8% (CI = 59-72%) for one vs. four, two vs. three, and threevs. four, respectively (see Figure 2).Participants’ performance was significantlyabove chance for all conditions (one vs. four,t(5)= 5.39, p < 0.002; two vs. three, t(5) =2.45, p < 0.03; three vs. four, t(5) = 4.91, p <0.002). A repeated measures ANOVA withcondition (one vs. four, two vs. three, andthree vs. four) as a within-subject factorrevealed no significant effects. Although thetrend did not quite reach significance (F(2,10)= 2.01, p = .18, η 2

G = .10), the averagepercent correct increased as the ratio betweenthe quantities got larger. Non-parametric testsrevealed the same ratio-dependent pattern ofperformance. We again used binomial tests tomeasure the reliability of each monkey’sperformance in each test condition (one vs.four, two vs. three, and three vs. four--seeFigure 4). Results revealed that the number ofmonkeys reliably choosing the larger amountfor the one vs. four, two vs. three, and threevs. four comparisons, respectively, was 3/6,2/6, 1/6. An additional correlation analysisconfirmed this pattern; as in Experiment 1, weagain observed a highly significant negativecorrelation between ratio and performance (r= -.57, t(34) = 4.04, p < 0.0003). When theproportional similarity between the numbersincreased, participants’ performancedecreased. Thus, as with discrete objects,monkeys’ performance again appeared to beratio-dependent; the larger the ratio betweenthe substance quantities, the better theperformance.

To explore performance acrossconditions and across Experiments 1 and 2,we performed a repeated-measures ANOVAwith experiment (object and substance) andcomparison (one vs. four, two vs. three, andthree vs. four) as within subject factors.There was no main effect of experiment

(F(1,5) = 0.056, p = 0.82, η 2G = 0.008).

Overall, participants performed similarlywhen faced with objects and substances. OurANOVA did, however, reveal a significanteffect of comparison (F(2,10) = 7.83, p =0.009, η2

G = 0.48). Across both studies,participants demonstrated a reliable pattern ofperforming best on the one vs. fourcomparison (81 +/- 12%; CI = 74-87%),slightly worse on the two vs. threecomparison (67 +/- 15%; CI = 58-75%), andworst on the three vs. four comparison (61 +/-10%; CI = 56-67%). As the ratio between thetwo items approached 1:1, participants’performance got worse. There was nointeraction between experiment and condition(F(2,10) = 2.36, p = 0.14); participantsdemonstrated the same pattern of performanceacross both Experiments 1 and 2.

Figure 4. Number of trials (out of 10) in which individualmonkeys chose the larger quantity of substance for each ratiocondition in Experiment 2. Asterisks indicate thatperformance was significantly above chance (Binomial Test,p < .05).

Discussion

Capuchins are able to quantify andcompare different amounts of a continuoussubstance just as well as they quantify thesame amounts (i.e., ratios) of discrete objects.Capuchins selected the larger of the two


quantities significantly more often thanchance across all ratio comparisons. As in thefirst experiment, performance was ratio-dependent--the bigger the proportionaldifference, the more likely participants wereto select the larger amount. Monkeys, likehuman infants (vanMarle, 2004), canrepresent quantities of a nonsolid substance,compute the total amount of substance in eachquantity, and compare these representations.One potential concern with this interpretationis that the total amount of substance in eachquantity was perfectly confounded with the‘number of pouring events’. Thus, monkeyscould have selected the larger quantity bycounting the ‘number of pouring events’instead of computing the total amount ofsubstance in each quantity. Although a validconcern, the fact that performance wasreliably above chance in the three vs. fourSubstance condition, but not the three vs. fourObject condition, argues against thispossibility. If they had been counting thenumber of hiding events, then one wouldexpect them to show equivalent performancein both cases.

General Discussion

The experiments reported hereexamined capuchin monkeys’ ability toquantify objects and substances. Consistentwith previous results with nonhuman primates(Beran, 2001; Hauser et al., 2000) and humaninfants (Feigenson et al., 2002; vanMarle,2004), capuchin monkeys are able to quantifydiscrete quantities of food and when given achoice between two quantities, they reliablychoose the larger amount. The present studyextended these earlier results by comparingcapuchin monkeys’ ability to quantify discreteobjects versus continuous substances.Previous work with infants (Huntley-Fenneret al., 2002; vanMarle, 2004) and adults

(vanMarle & Scholl, 2003) suggests thattracking substances is more difficult thantracking discrete objects. The present findingsare somewhat in accord with these previousfindings; on their first one vs. two testsession, monkeys performed reliably better atdiscriminating objects than substances. Thatsaid, unlike infants and adults, our capuchinsperformed above chance at both object andsubstance discriminations. In this sense, ourfindings are also consistent with recentresearch comparing infants’ ability to quantifysubstances and objects in a similar objectchoice task (vanMarle, 2004). In that study,10- and 12- month-old infants were able torepresent and compare the magnitudes of twohidden portions of food substance, but theirability to do so was limited compared to theirdiscrete quantification abilities. Thus,although processing substances appears to besomewhat more difficult than processingdiscrete objects, both human and nonhumanprimates may have some way of representingamount of “stuff” in addition to discretenumbers of objects. Of course, given that‘amount of stuff’ was confounded with otherdiscrete cues (e.g., ‘number of pours’) inExperiment 2, our results with capuchinsshould be viewed with caution. We predicthowever, that capuchins would still showsuccessful (and ratio-dependent) performanceif tested under conditions that ruled outdiscrete cues to “amount of stuff”.

One obvious question to ask iswhether the same mechanisms thought tounderlie discrete quantification (an analogmagnitude mechanism and an object trackingsystem) also underlie monkeys’ ability toquantify continuous substances. Our presentresults support the view that monkeys may beusing analog magnitudes in both cases. First,monkeys were successful at discriminatingthe two quantities regardless of whether thefood quantities were sets of discrete


individuals or substances. Second, capuchins’performance was dependent on the ratiobetween the two numbers to be discriminated;such ratio-based performance is a classicsignature of an analog magnitude mechanism.

For these reasons, our results suggestthat nonhuman primates can use analogmagnitudes to represent small sets of discreteobjects. Moreover, given that the objecttracking system is both sensitive to whetheran entity is an object or a portion ofsubstance, and has difficulty tracking entitiesthat move like nonsolid substances (i.e.,extend/contract, disintegrate, etc.), it seemsunlikely that the capuchins were using thismechanism to quantify the substances inExperiment 2. Therefore, the similarity inperformance across Experiments 1 and 2 ismore consistent with the notion that thecapuchins were using analog magnitudes torepresent the quantities in both cases.

Our results add to an increasinglypuzzling picture about which types ofmechanisms underlie nonverbal quantityrepresentations in different tasks. On onehand, both preverbal infants (Feigenson et al.,2002) and rhesus monkeys (Hauser et al.,2000) show clear set size limitations in anordinal choice task, consistent with an objecttracking account. In contrast, chimpanzees(Beran, 2001, 2004), orangutans (Call, 2000),and now capuchins show clear ratio-dependent performance, consistent with ananalog magnitude account (see also Lewis etal., 2005, for a similar findings in a slightlydifferent task with lemurs). Is there a way torectify these sets of findings? One possibilityconcerns the way in which the two sets ofstudies were conducted. Both Feigenson et al.(2002) and Hauser et al. (2000) collected onlya single data point for each participant. Suchbetween-subject analyses make it difficult toobserve possibly subtle differences inperformance across different ratio

comparisons. Previous research withchimpanzees (Beran, 2001, 2004), in additionto the present studies, used a more sensitiveparadigm by testing individual participants onmultiple trials and multiple comparisons. Inboth cases, this approach revealed differencesin performance that were dependent on ratio.It is possible, then, that rhesus monkeys (andpossibly human infants) would show ratio-dependent performance if tested on multipletrials of the same numerical comparison in anordinal choice task. A second possibilityconcerns the role of multiple trials indetermining which nonverbal number systembegins to operate. It is possible that seeingmultiple numerical comparisons togethersomehow engages the analog magnitudesystem in a way that single presentationswould not. This alternative makes theprediction that capuchin monkeys tested withsingle trials would perform much like infantsand rhesus monkeys, demonstrating a set-sizelimit rather than ratio-dependent performance.

Since these experiments represent aninitial attempt to observe successful quantity-based responding in a primate species whosequantitative abilities have been relativelyneglected (but see Judge et al., 2005), theyface some limitations. First, in order to findstrong evidence that capuchins are reallyusing analog magnitudes rather than objectindexes to quantify objects and substances, itis necessary to test quantities that exceed thesupposed set size limit but are of adiscriminable ratio. For example, based on thepresent results, if capuchins are using analogmagnitudes in our task, then they should beable to reliably choose the larger quantitywhen presented comparisons of three vs. sixraisins. In fact, we attempted to test this exactcomparison, but due to methodologicaldifficulties, the results were not interpretable.Specifically, we found that: (a) the monkeysbecame satiated quickly (from obtaining such


a large number of raisins), which resulted inlowered motivation and failure to complete afull testing session, and (b) the amount oftime necessary to hide six raisins apparentlyexceeded our monkeys’ attention span whichresulted in their being increasingly morelikely to stop participating as the session wenton. Future studies will attempt to obtain thiscritical data by having monkeys completeshorter sessions across separate testing days.

Second, further research is necessaryto rule out temporal cues to amount thatcovaried with the dimensions of interest(number for Experiment 1, amount forExperiment 2). In particular, future studieswill profit from controlling the ‘rate ofpresentation’, ‘duration of presentation’, and‘number of pouring events’. Doing so willallow for the further investigation of theconditions under which object indexes oranalog magnitudes are employed in theabsence of these cues. It should be notedhowever, that analog magnitudes are notspecific to number representation. In fact,they were originally used to account for rats’ability to represent temporal intervals (ScalarExpectancy Theory; Gibbon, 1977) and laterproposed to underlie rats’ ability to representnumber (Meck & Church, 1983). Thus, evenif our capuchins were responding on the basisof temporal attributes of the presentationsequences (which we feel is unlikely for thereasons discussed in the methods sections),this is still consistent with the conclusion thatperformance was based on the use of analogmagnitudes. In addition, if this were the case,it would further rule out the possibility thatthey were using object indexes since objecttracking models have no provisionwhatsoever for the representation of temporalproperties of events or event sequences.

In conclusion, the present datacontribute to the ongoing debate regarding thenature of the representations that underlie

nonverbal ordinal judgments for discretequantities. Our capuchin findings also go astep further by examining whetherperformance differs depending on whether thequantified entities are discrete objects orcontinuous substances. The results support thenotion that capuchin monkeys may use analogmagnitudes to quantify small sets of discreteobjects and portions of continuous substances.Further research is necessary to elucidate theprecision with which capuchins can quantifyobjects and substances and the conditionsunder which an organism will use eitheranalog magnitudes or object indexes torepresent a given quantity. Answers to thesequestions will inform not only theories ofcomparative cognition and cognitivedevelopment, but will also contribute to thebroader question of how organisms representquantitative information in their environmentand use it to guide their behavior.

References

Beran, M. J. (2001). Summation andnumerousness judgments o fsequentially presented sets of items bychimpanzees (Pan troglodytes). Journalof Comparative Psychology, 115, 181-191.

Beran, M. J. (2004). Chimpanzees (Pantroglodytes) respond to nonvisible setsafter one-by-one addition and removalof items. Journal of ComparativePsychology, 118, 25-36.

Beran, M. J., & Beran, M. M. (2004).Chimpanzees remember the results ofone-by-one addition of food items tosets. Psychological Science, 15, 94-99.

Beran, M. J., & Rumbaugh, D. M. (2001)."Constructive" enumeration bychimpanzees (Pan troglodytes) on acomputerized task. Animal Cognition, 4,81-89.


Boysen, S. T., & Berntson, G. G. (1989).Numerical competence in a chimpanzee( Pan troglodytes ). Journal ofComparative Psychology, 103, 23-31.

Brannon, E. M., & Roitman, J. (2003).Nonverbal representations of time andnumber in nonhuman animals andhuman infants. In W. Meck (Ed.),Functional and Neural Mechanisms ofInterval Timing (pp. 143-182). NewYork: CRC Press.

Brannon, E. M. & Terrace, H. S. (1998).Ordering of the numerosities 1-9 bymonkeys. Science, 282, 746-749.

Brannon, E. M. & Terrace, H. S. (2000).Representation of the numerosities 1-9by rhesus macaques (Macaca mulatta).Journal of Experimental Psychology:Animal Behavior Processes, 26, 31-49.

Brannon, E. M., Wusthoff, C. J., Gallistel, C.R., & Gibbon, J. (2001). Numericalsubtraction in the Pigeon: Evidence fora linear subjective number scale.Psychological Science, 12(3), 238-243.

Call, J. (2000). Estimating and operating ondiscrete quantities in Orangutans(Pongo pygmaeus). Journal ofComparative Psychology, 114(2), 136-147.

Cordes, S., Gelman, R., Gallistel, C. R. &Whalen, J. (2001). Variabilitysignatures distinguish verbal fromnonverbal counting for both large andsmall numbers. Psychological Bulletinand Review, 8, 698-707.

Davis, H., & Perusse, R. (1988). Numericalcompetence in animals: Definitionalissues, current evidence, and a newresearch agenda. Behavioral and BrainSciences, 11(4), 561-615.

Dehaene, S. & Changeaux, J. P. (1993).Development of elementary numericalabilities: A neural model. Journal ofCognitive Neuroscience, 5, 390-407.

Dehaene, S. (1997). The number sense. Oxford:Oxford University Press.

Feigenson, L. & Carey, S. (2003). Trackingindividuals via object-files: Evidencefrom infants’ manual search.Developmental Science, 6, 568-584.

Feigenson, L., Carey, S., & Hauser, M. D.(2002). The representations underlyinginfants' choice of more: object filesv e r s u s a n a l o g m a g n i t u d e s .Psychological Science, 13, 150-156.

Feigenson, L., Dehaene, S., & Spelke, E.S.(2004). Core systems of number. Trendsin Cognitive Sciences, 8(7), 307-314.

Fernandes, D. M. & Church, R. M. (1982).Discrimination of the number ofsequential events by rats. A n i m a lLearning & Behavior, 10, 171-176.

Flombaum, J. I., Junge, J. A., & Hauser, M. D.(2005). Rhesus monkeys (Macacamulat ta ) spontaneously computeaddition operations over large numbers.Cognition, 97, 315-325.

Fragaszy, D. M., Visalberghi, E., & Fedigan, L.M. (2004). The Complete Capuchin:The Biology of the Genus Cebus. NewYork: Cambridge University Press.

Gallistel, C. R. & Gelman, R. (2000).Nonverbal numerical cognition: Fromreals to integers. Trends in CognitiveSciences, 4, 59-65.

Gallistel, C. R. & Gelman, R. (2005).Mathematical Cognition. In K. Holyoak& R. Morrison (Eds.) The Cambridgehandbook of thinking and reasoning(pp. 559-588). New York: CambridgeUniversity Press.

Gibbon, J. (1977). Scaler expectancy theoryand Weber’s law in animal timing.Psychological Review, 84, 279-325.

Hall, G. (1996). Naming solids and nonsolids:Children’s default construals. CognitiveDevelopment, 11, 229–264.


Hauser, M. D., & Carey, S. (2003).Spontaneous representations of smallnumbers of objects by rhesus macaques:Examination of content and format.Cognitive Psychology, 47(4), 367-401.

Hauser, M. D., Carey, S., & Hauser, L. B.( 2 0 0 0 ) . Spontaneous numberrepresentation in semi-free-rangingrhesus monkeys. Proceedings of theRoyal Society, London, 267, 829-833.

Hauser, M. D., MacNeilage, P., & Ware, M.(1996). Numerical representations inprimates. Proceedings of the NationalAcademy of Sciences, 93, 1514-1517.

Hauser, M. D. & Spelke, E. (2004). Theevolutionary and developmentalfoundations of human knowledge: Acase study of mathematics. In M.Gazzaniga (Ed.) The CognitiveNeurosciences III (pp. 853-864).Cambridge: MIT Press.

Hauser, M. D., Tsao, F., Garcia, P., & Spelke,E. S. (2003). Evolutionary foundationsof number: spontaneous representationof numerical magnitudes by cotton-toptamarins. Proceedings of the RoyalSociety, London, Series B, 270, 1441-1446.

Howell, D. C. (1997). Statistical Methods forPsychology. (4th ed.). Belmont, CA:Wadsworth Publishing Company.

Huntley-Fenner, G. (1995). The representationof objects, non-solid substances, andcollections in infancy and earlychildhood. (Doctoral dissertation,Massachusetts Institute of Technology[MIT], 1995). Dissertation AbstractsInternational: Section B: The Sciences& Engineering, 56(4-B), 2352.

Huntley-Fenner, G., Carey, S., & Solimando,A. (2002). Objects are individuals butstuff doesn’t count: Perceived rigidityand cohesiveness influence infants’representations of small numbers of

discrete entities. Cognition, 85, 203-221.

Jordan, K. E., Brannon, E. M., Logothetis, N.K. & Ghazanfar, A.A. (2005) Monkeysmatch the number voices they hear tothe number of faces they see. CurrentBiology, 15, 1034-1038.

Judge, P. G., Evans, T. A., & Vyas, D. K.(2005). Ordinal representation ofnumeric quantities by brown capuchinmonkeys (Cebus apella). Journal ofExperimental Psychology: AnimalBehavior Processes, 31, 79-94.

Kahneman, D., Treisman, A., & Gibbs, B.J.(1992). The reviewing of object files:Object-specific integration ofinformation. Cognitive Psychology, 24,174-219.

Lewis, K. P., Jaffe, S., & Brannon, E. M.(2005). Analog number representationsin mongoose lemurs (Eulemur mongoz):Evidence from a search task. AnimalCognition, 8(4), 247-252.

McCrink, K., & Wynn, K. (2004). Large-number addition and subtraction ininfants. Psychological Science, 15, 776-781.

Mechner, F. (1958). Probability relationswithin response sequences under ratioreinforcement. Journal of theExperimental Analysis of Behavior, 1,109-122.

Meck, W. H., & Church, R. M. (1983). A modecontrol model of counting and timingprocesses. Journal of PsychologicalAnimal Behavior Processes, 9(3), 320-34.

Nieder, A. (2005). Counting on Neurons: TheNeurob io logy o f Numer ica lCompetence. Nature ReviewsNeuroscience, 6(3), 177-190.

Nieder, A., Freedman, D. J., & Miller, E.K.(2002). Representation of the Quantityof Visual Items in the Primate


Prefrontal Cortex. Science, 297, 1708-1711.

Nieder, A. & Miller, E.K. (2003). Coding ofCognitive Magnitude: CompressedScaling of Numerical Information in thePrimate Prefrontal Cortex. Neuron, 37,149-157.

Nieder, A. & Miller, E.K. (2004). AnalogNumerical Representations in RhesusMonkeys: Evidence for ParallelProcessing. Journal of CognitiveNeuroscience, 16, 889-901.

Olejnik, S. & Algina, J. (2003). GeneralizedEta and Omega Squared Statistics:Measures of Effect Size for SomeCommon Research Des igns .Psychological Methods, 8(4), 434-447.

Prasada, S. (1999). Names of things and stuff:An Aristotelian perspective. In R.Jackendoff, P. Bloom, & K. Wynn(Eds.) , Language, logic, andconcepts:Essays in memory of JohnMacnamara (pp. 119–146). Cambridge,MA: MIT Press.

Pylyshyn, Z. W. (1989). The role of locationindexes in spatial perception: A sketchof the FINST spatial-index model.Cognition, 32, 65-97.

Pylyshyn, Z. W., & Storm, R. W. (1988).Tracking Multiple Independent Targets:Evidence for a Parallel TrackingMechanism. Spatial Vision, 3(3), 179-197.

Santos, L. R., Barnes, J. L. & Mahajan, N.(2005). Expectations about numericalevents in four lemur species (Eulemurfulvus, Eulemur mongoz, Lemur cattaand Varecia rubra). Animal Cognition,8(4), 253-262.

Scholl , B. J., Pylyshyn, Z. W., & Feldman, J.(2001). What is a visual object?Evidence from target merging in

multiple object tracking. Cognition, 80,159-177.

Simon, T. J. (1997). Reconceptualizing theorigins of number knowledge: A “non-numerical” account. C o g n i t i v eDevelopment, 12, 349-372.

Soja, N.N. (1992). Inferences about themeanings of nouns: The relationshipbetween Perception and syntax.Cognitive Development, 7, 29–45.

Soja, N.N., Carey, S., & Spelke, E.S. (1991).Ontological categories guide youngchildren’s inductions of word meaning:Object terms and substance terms.Cognition, 38, 179–211.

vanMarle, K. L. (2004). Infants' understandingof number: The relationship betweendiscrete and continuous quantity.(Doctoral dissertation, Yale University,2004). Disser ta t ion AbstractsInternational: Section B: The Sciences& Engineering, 65(3-B), 1582. US:Univ Microfilms International.

vanMarle, K., & Scholl, B. J. (2003). AttentiveTracking of Objects Versus Substances.Psychological Science, 14(5), 498-504.

Whalen, J., Gallistel, C. R., & Gelman, R.(1999). Nonverbal counting in humans:the psychophysics of numberrepresentation. Psychological Science,10, 130-137.

Wynn, K. (1992). Addition and subtraction byhuman infants. Nature, 358, 749-750.

Xu, F. (2003). Numerosity discrimination ininfants: Evidence for two systems ofrepresentations. Cognition, 89, B15-B25.

Xu, F., & Spelke, E. S. (2000). Large numberdiscrimination in 6-month old infants.Cognition, 74, B1-B11.

Xu, F., Spelke, E. S., & Goddard, S. (2005).Number sense in human infants.Developmental Science, 8(1), 88-101.

How Capuchin Monkeys (Cebus apella) Quantify Objects and ...How Capuchin Monkeys (Cebus apella)...

Documents

Transcript of How Capuchin Monkeys (Cebus apella) Quantify Objects and ...How Capuchin Monkeys (Cebus apella)...