lable at ScienceDirect

Animal Behaviour 79 (2010) 869–875

Contents lists avai

Animal Behaviour

journal homepage: www.elsevier .com/locate/anbehav

Positive interactions lead to lasting positive memories in horses,Equus caballus

Carol Sankey a,*, Marie-Annick Richard-Yris b,1, Helene Leroy a, Severine Henry a,Martine Hausberger a,b

a EthoS, Station Biologique, Universite de Rennes 1b EthoS, ethologie animale et humaine – UMR CNRS 6552, Universite de Rennes 1

a r t i c l e i n f o

Article history:Received 12 July 2009Initial acceptance 25 September 2009Final acceptance 15 December 2009Available online 9 February 2010MS. number: 09-00470R

Keywords:Equus caballushorsehuman–animal relationshiplearningmemorypositive interactionsocial cognition

* Correspondence: C. Sankey, Station Biologique, Laet humaine, UMR 6552 CNRS, Universite de Rennes 1

E-mail address: carol.sankey@univ-rennes1.fr (C. S1 M.-A. Richard-Yris is at the Laboratoire d’ethologi

CNRS 6552, Universite de Rennes 1, Avenue du GBeaulieu, F-35042 Rennes Cedex, France.

0003-3472/$38.00 � 2010 The Association for the Studoi:10.1016/j.anbehav.2009.12.037

Social relationships are important in social species. These relationships, based on repeated interactions,define each partner’s expectations during the following encounters. The creation of a relationshipimplies high social cognitive abilities which require that each partner is able to associate the positive ornegative content of an interaction with a specific partner and to recall this association. In this study, wetested the effects of repeated interactions on the memory kept by 23 young horses about humans, after 6and 8 months of separation. The association of a reward with a learning task in an interactional contextinduced positive reactions towards humans during training. It also increased contact and interest, notonly just after training, but also several months later, despite no further interaction with humans. Inaddition, this ‘positive memory’ of humans extended to novel persons. Overall, positive reinforcementenhanced learning and memorization of the task itself. These findings suggest remarkable socialcognitive abilities that can be transposed from intraspecific to interspecific social contexts.� 2010 The Association for the Study of Animal Behaviour. Published by Elsevier Ltd. All rights reserved.

The quality of social relationships within a group has an adap-tive value (Silk 2007; Cameron et al. 2009). These relationships arebuilt on repeated interactions, and are defined by each partnerhaving expectations during subsequent encounters (Hinde 1979).Depending on the perceived valence (positive/negative) of theinteractions, the relationship will range from confidence andreassurance to fear and stress, implying differential cerebral pro-cessing of the positive and negative emotions induced (Silberman &Weingartner 1986; Demaree et al. 2005). Therefore, the creation ofa relationship implies high social cognitive abilities which requirethat each partner is able to associate the positive or negativeemotional content of an interaction with a specific partner and torecall this association.

In the wild, horses generally live in family groups composed ofan adult stallion, one to three unrelated mares and their young,although up to a third of groups may contain up to five stallions andup to 26 mares and their young (Feist & McCullough 1976; Linklater

boratoire d’ethologie animale, 35380 Paimpont, France.ankey).e animale et humaine – UMR

eneral Leclerc, Campus de

dy of Animal Behaviour. Published

2000; Boyd & Keiper 2005). The first social interactions occurring ina foal’s life are with its dam, and often consist of feed-relatedinteractions through the suckling behaviour. If some sort ofinterference occurs around that first suckling, the mother–youngrelationship may be impaired in the long term (Hausberger et al.2007b). Thus, the first and strongest social bond in a horse’s life,one uniting a foal with its dam, appears to be at least partlya feeding bond. The family group represents a rich social environ-ment surrounding the young during its development (e.g. Boyd1988; Feh 2005; Bourjade et al. 2009). Horses maintain long-termbonds with several members of their family group, but they alsointeract temporarily with members of other groups when formingherds (Waring 2003; Feh 2005). Equid social relationships are longlasting and, in some cases, lifelong (Linklater 2007).

In groups of horses, social preferences are expressed by a greaterspatial proximity and frequent affiliative behaviours, such asmutual grooming (Clutton-Brock et al. 1976; Feh 1987; Sigurjons-dottir et al. 2003; van Dierendonck et al. 2004). Opportunisticobservations suggest that horses are able to recognize social part-ners (e.g. after a 1-year separation from their mares, stallionsimmediately distinguished and joined their former mates in a largeherd with many other mares present; Feh 2005). Experimentalstudies have demonstrated that horses can distinguish the calls ofnonkin group members from the calls of familiar individuals living

by Elsevier Ltd. All rights reserved.

C. Sankey et al. / Animal Behaviour 79 (2010) 869–875870

in a neighbouring group and from the calls of strangers (Lemassonet al. 2009), and that they are even capable of cross-modal indi-vidual recognition (Proops et al. 2009). These recent findingssuggest a process of social learning of vocal signatures and long-term memorization of the different degrees of social bonds main-tained among horses (Lemasson et al. 2009; Proops et al. 2009).

It has been suggested that the social cognitive skills of horses tonegotiate complex relationships may apply to human–animalinteractions and relations (Kruger 2007; Linklater 2007; Murphy &Arkins 2007; Hausberger et al. 2008). However, while the memoryof interactions is primarily associated with individual recognition,it has been shown in many species that in social contexts, theexperience an animal has with a specific person may be generalizedto other humans (poultry, Gallus gallus: Ghirlanda et al. 2002; pigs,Sus scrofa: Hemsworth et al. 1994, 1996; cattle, Bos taurus: Lensinket al. 2000; dogs, Canis lupus familiaris: Miklosi et al. 1998). Indomestic horses, there is growing evidence that horses tend toshow consistency in their reactions to different persons (Haus-berger & Muller 2002; Henry et al. 2005; Lansade & Bouissou 2008;Fureix et al. 2009).

Positive human–animal relationships have often been used toinvestigate animal cognition (Davis 2002). Using a close relation-ship with their subjects, Boysen (1992) and Pepperberg (2000)have extended the boundaries of what is known about the mentalabilities of chimpanzees, Pan troglodytes, and African grey parrots,Psittacus erithacus, respectively. However, once the relationship hasbecome established, one may wonder what memory would be keptif the interactions were to cease and a prolonged separation wereto occur.

Here, we used the human–horse relationship to investigate theability of horses to build a relationship on the basis of short- andlong-term memory of the interactions. Thus, we tested whethera series of positive interactions could create and influence thehuman–animal relationship in the long term, even after a pro-longed separation. We used the horse training context whichrequires daily human–horse interactions, and provided food duringthe training sessions to make training a pleasurable experience forthe animals. Horses’ behaviour during training confirmed that theuse of food rewards led them to perceive training sessions aspositive interactions (see Results). We hypothesized that horses areable to build a bond and keep a long-term memory of the rela-tionship, on the basis of repeated interactions. In addition, wetested whether horses, in this interspecific context, were alsocapable of building a general perception of humans on such a basis.

METHODS

Animals and Experimental Groups

We studied 20 Anglo-Arabian and three French Saddlebredhorses (15 females and eight geldings). All horses were born andhoused at the ‘Station Experimentale des Haras Nationaux’ (SEHN),Chamberet, France. When 4–6 days old (in April–June 2006), foalswere led with their dams to a large pasture where they stayedtogether until weaning at the age of 6 months. From weaning,horses were kept at pasture in randomly composed small groups(three groups of six and one group of five). Three weeks afterweaning, they were pushed along fenced pathways into four largeindoor stalls for winter; each weaning group was kept in an adja-cent but separate stall. At the age of 7 � 1 months, all malesunderwent a surgical castration procedure. Our training pro-gramme started in the spring when the horses were approximately1-year old, and all yearlings were then released in a large pasturenear the training arena. During the winter period followingtraining, they were led back to their winter stables in randomly

composed groups of five or six individuals. Before and during thisexperiment, additional human interventions were limited to haydistribution by caretakers twice a day during the winter periods.

Horses were randomly assigned to two treatment groupsaccording to their sex and sire: horses trained with positive rein-forcement (PR; N ¼ 11) and controls, trained with no reinforcement(C; N ¼ 12). The positive reinforcement was a food reward, whichconsisted of a few hand-given grain pellets. As the experiment wasconducted in spring and summer, horses from both groups werekept together in a large pasture and were not given any supple-mentary feed.

Learning and Remembering

Training procedureThe aim of our training programme was to train yearlings to

remain immobile without being held in response to a simple vocalcommand: ‘reste!’ (i.e. French for ‘stay!’). Once this was achieved,the horses underwent various handling or veterinary proceduresusing the same vocal command. All horses were trained untilcompletion of the training programme (Table 1). Before the start oftraining, all yearlings were habituated to being haltered and led inhand. Rewards were never used during this initial habituation.

Horses were trained 5 min/day, 5 days per week. During the tworemaining days, they all stayed together in a pasture. On trainingdays, the horses were pushed along fenced pathways from theirpasture to a large indoor pen. They were then individually allowedto enter an adjacent training arena (10.5 � 15 m), separated fromthe pen by large doors, which prevented the horses from touchingor seeing each other during training. During this experiment,horses would spontaneously approach the doors leading to thetraining arena and the experimenter only had to open the doorsand let one horse in at a time. Training was performed by a singleexperimenter (woman, long dark hair). She dressed differentlyevery day but always wore the same green coat. At the beginning ofeach training session, the experimenter opened the door of thetraining arena and waited for one horse to enter spontaneously. Itwas haltered and a lead rope was attached. Training then began fora 5 min session, after which the horse was released in an outdoorpen. At the end of the day, all the horses were released in theirpasture.

During all training sessions, the horses’ behaviour was contin-uously recorded using a digital voice recorder equipped witha microphone, to verify whether horses perceived positive rein-forcement training as a positive interaction (e.g. positive/affiliativebehaviours: sniffing, licking; versus negative/defence behaviours:biting, kicking or ‘falling down’ on the experimenter, which consistsof the horse leaning heavily against, and letting itself fall down on,the experimenter while she was picking up a foot).

The experimental training programme involved learning toremain immobile in response to a vocal command and thenaccepting various handling or veterinary procedures. Training wasdivided into steps of increasing complexity as described in Table 1.The horse had to fulfil the performance criterion of each step to getto the next one, that is, it had to succeed three times consecutivelyin the given step. For example, a horse had to remain immobile onorder while the experimenter brushed its whole body three times(trials) consecutively (step 12), before moving on to the nexthandling procedure (feet picking: steps 13–21). Horses were notlimited in the number of steps they could complete successfullywithin a training session, which simply ended after 5 min. Eachtime they remained immobile for the required duration (steps 1–6)and accepted the handling procedure (steps 7–41), horses of the PRgroup received hand-given pellets as a reward and carried on withthe training programme, whereas horses from the C group were

Table 1Description of the steps comprising the yearlings’ training programme

Steps Description

Steps 1–5and 7–41

When the experimenter stopped walking, she placed the leadrope on the horse’s neck before giving the vocal order to stay. Tosucceed, the horse had to maintain immobility for the followingtimes

Step 1 For 5 sStep 2 For 10 sStep 3 For 30 sStep 4 For 45 sStep 5 For 1 minStep 6 For 1 min (but no lead rope attached)Step 7 While the experimenter brushed its neck and shoulder

(30 s/side)Step 8 While the experimenter brushed its back and croup (30 s/side)Step 9 While the experimenter brushed its flank and belly (30 s/side)Step 10 While the experimenter brushed its forelegs and hindlegs

(30 s/side)Step 11 While the experimenter brushed its tail for 1 minStep 12 While the experimenter brushed its whole body for 1 min

(30 s/side)Step 13 And gently let the experimenter pick up its front feetStep 14 And gently let the experimenter pick up its back feetStep 15 And gently let the experimenter pick up its front feet and hold

them (5 s/foot)Step 16 And gently let the experimenter pick up its back feet and hold

them (5 s/foot)Step 17 And gently let the experimenter pick up its front feet and hold

them (10 s/foot)Step 18 And gently let the experimenter pick up its back feet and hold

them (10 s/foot)Step 19 And gently let the experimenter clean its front feet (10 s/foot)Step 20 And gently let the experimenter clean its back feet (10 s/foot)Step 21 And gently let the experimenter clean its four feet (10 s/foot)Step 22 While the experimenter placed a surcingle on its backStep 23 While the experimenter placed a surcingle on its back and left

it for 5 sStep 24 While the experimenter placed a surcingle on its back and

passed the belts under the bellyStep 25 While the experimenter placed a surcingle on its back and

tightened the beltsStep 26 While the experimenter placed a surcingle on its back and

attached the beltsStep 27 While the experimenter placed and attached the surcingle and

then walk two steps with the surcingle attachedStep 28 As in step 27, but five stepsStep 29 As in steps 27 and 28, but 10 stepsStep 30 While the experimenter fitted tendon boots on its forelegsStep 31 While the experimenter fitted protection boots on its hindlegsStep 32 The horse had to remain immobile while the experimenter

fitted tendon boots on its forelegs and left them for 10 s eachStep 33 While the experimenter fitted protection boots on its hindlegs

and left them for 10 s eachStep 34 While the experimenter lifted its tailStep 35 While the experimenter inserted a thermometer in its rectumStep 36 While the experimenter inserted a thermometer in its rectum

and left it for 10 sStep 37 While the experimenter applied a vapour spray on its neck and

shouldersStep 38 While the experimenter applied a vapour spray on its back and

croupStep 39 While the experimenter applied a vapour spray on its flanks and

bellyStep 40 While the experimenter applied a vapour spray on its four legsStep 41 While the experimenter applied a vapour spray on its whole

body

C. Sankey et al. / Animal Behaviour 79 (2010) 869–875 871

never rewarded and simply carried on with the next steps. A trialwas considered a fail when the horse moved before the requiredduration or before the end of the handling procedure or when itshowed defensive behaviours. Between each trial (succeeded orfailed), the horse was led in hand for 15 s around the training arena.Punishment was never used in this experiment, nor was negativereinforcement.

At the end of training, each horse had learnt to remain immobileon order, with the lead rope simply placed over its neck, andaccepted the various handling procedures included in the trainingprogramme.

Tests of long-term rememberingTo test for possible long-term memory, each winter stall group

was successively driven to the pen adjoining the training arena(used in the initial training programme). Each horse was thenindividually allowed to enter the training arena. After two human–animal relationship tests (see below), the horse was haltered andthe memorization test began. Two memorization tests wereconducted, one at 6 months and one at 8 months after the last horsecompleted the training programme. The first test was with theoriginal familiar trainer (dressed differently compared with theinitial training) and the second with an unknown experimenter(woman, short dark hair, dressed differently from the familiarexperimenter).

These tests took place as follows: first, the experimenter led thehorse around the training arena for 15 s before stopping. She thenplaced the lead rope on the animal’s neck, ordered it to stay, tooka step back and remained immobile for 1 min. She then performedthe six handling procedures the animals had been trained to accept(i.e. brushing the whole body, picking the four feet, attaching thesurcingle and walking with the horse while it made 10 steps,applying tendon boots on four legs, taking temperature with ananal thermometer and applying a vapour spray on the whole body).If the horse failed to remain immobile in response to the vocal orderduring a given procedure, the experimenter took the lead rope andled the horse for another 15 s before stopping again and ordering itto stay while she carried on with the remaining procedures. Noreward was provided during the tests.

Generalization: novel situation/novel personAfter the end of our training programme, the experimental

station required that all horses were dewormed using an oral paste.This procedure was performed by one of the station’s technicians(male, short dark hair), who was newly employed and thereforehad had no contact with our experimental horses. This proceduretook place in the same training arena the horses had originally beentrained in. Each horse was individually allowed in the arena; thetechnician fitted it with a halter and a lead rope, placed the leadrope on the animal’s neck and ordered it to stay. He then proceededwith the deworming procedure by inserting the syringe containingthe oral paste in the horse’s mouth and pressed to release the paste.Again, no reward was provided during this procedure.

Evaluation of the Human–Animal Relationship

To assess the effect of our experimental training programme onthe horses’ perception of humans and therefore on the quality ofthe horse–human relationship, all horses were individually testedin the training arena (10.5 � 15 m) using two tests commonly usedin the literature (described in Henry et al. 2005; reviewed inWaiblinger et al. 2006; Hausberger et al. 2008).

(1) Motionless person test: during this test, the experimenterstood motionless in the centre of the training arena. A secondexperimenter opened the door to let one horse into the arenabefore walking out of the arena. The horse was allowed to run freein the arena for 5 min. The experimenter standing in the middle ofthe arena did not move and did not look at the horse, nor did she tryto interact with the animal. The latency to approach the experi-menter was recorded as well as the distance to the experimenterand the behaviour of the animal, which were recorded every 10 susing the scan sampling method (Altmann 1974). For the horses

***6

5

4

e (h

)

C. Sankey et al. / Animal Behaviour 79 (2010) 869–875872

that never approached the experimenter, a maximum latency of300 s (5 min) was assigned.

(2) Approach-contact test: the experimenter stood at a distanceof 2 m from the animal and started slowly walking towards itsshoulder at an angle of 90�, looking at the horse in front of her. If thehorse moved away, the experimenter repeated this procedure untilthe end of the test. If the horse accepted the approach and allowedthe experimenter to touch its shoulder, the latency to contact wasrecorded. The maximum allowed time to touch the horse was 60 s.If the horse did not allow the experimenter to touch it during theallocated time, it was assigned a maximum latency of 60 s.

These two tests were performed four times during this experi-ment: (1) just before training (day before starting the trainingprogramme); (2) just after finishing the training programme (dayfollowing the successful completion of the last step, variableaccording to the horse); (3) after a delay of 6 months, with thefamiliar experimenter; and (4) after a delay of 8 months, with anunfamiliar person.

Variables and Statistics

During all training sessions, the behaviours of horses directedtowards the experimenter were recorded continuously (‘positive’behaviours: sniffing and licking the experimenter; ‘negative’behaviours: biting, kicking and ‘falling down’ on the experimenter).

During the memorization tests, we recorded success/failure torespond to the vocal command, duration of immobility andacceptance or not of handling procedures. During the motionlessperson test, the experimenter recorded the latency for the horse toapproach her, as well as the horse’s behaviour and distance to her,which was sampled every 10 s using the scan sampling method(Altmann 1974). In the approach-contact test, she recorded thetime required to touch the animal’s shoulder.

Analyses were carried out using nonparametric statistical testswith a significance threshold at 0.05. The Mann–Whitney U testwas used to compare all independent samples (i.e. experimentalgroups). The Wilcoxon signed-ranks test was used to comparerelated samples (i.e. comparison of the behaviour of the sameindividual at different times, for example before and after training).Success rates in the memorization test were compared betweengroups using Fisher’s exact test.

RESULTS

We were able to verify that the use of a food reward as a rein-forcement during training led PR horses to perceive trainingsessions as positive interactions: they displayed more ‘positive’behaviours towards the experimenter, such as sniffing and licking(Table 2), while negative reactions occurred almost only in controls.Horses trained without reinforcement expressed four to six timesmore ‘negative’ behaviours, such as biting, kicking and ‘fallingdown’ on the experimenter (Table 2).

Table 2Mean number of occurrences �SE of ‘positive’ and ‘negative’ behaviours directedtowards the experimenter during training

PR C Statistics

U P

Positivebehaviours

Sniffing 95.5�14.4 57.5�8.5 34.5 <0.05Licking 0.36�0.2 0.0�0.0 48.0 <0.05

Negativebehaviours

Biting 7.8�3.6 44.5�5.8 6.0 <0.001Kicking 0.5�0.3 5.4�1.9 22.5 <0.01Falling down 1.5�0.8 4.1�1.0 34.0 <0.05

PR: positive reinforcement; C: controls. Mann–Whitney U test: NPR ¼ 11, NC ¼ 12.

Horses trained with no reinforcement defecated more oftenduring training sessions than horses trained with positivereinforcement (XPR � SE ¼ 1� 0:3, XC � SE ¼ 2:6� 0:6;Mann–Whitney U test: U ¼ 28.5, NPR ¼ 11, NC ¼ 12, P < 0.05).

Learning and Remembering

Positive reinforcement enhanced learning and especiallymemory. PR horses required less time to complete training thancontrols (XPR � SE ¼ 3:75� 0:08 h, XC � SE ¼ 5:24� 0:24 h;Mann–Whitney U test: U ¼ 1, NPR ¼ 11, NC ¼ 12, P < 0.001; nooverlap between treatment groups; Fig. 1).

Six months after the end of initial training, most PR horses(8/11 versus 5/12 for C; NS) still responded to the vocal order. Theyalso remained immobile for longer than controls(XPR � SE ¼ 55:8� 2:2 s, XC � SE ¼ 38:0� 6:3 s; Mann–WhitneyU test: U ¼ 36, NPR ¼ 11, NC ¼ 12, P < 0.05) and still accepted mostof the handling procedures (XPR � SE ¼ 4:0� 0:3, XC � SE ¼ 1:7�0:3 of the six handling procedures; Mann–Whitney U test: U ¼ 13,NPR ¼ 11, NC ¼ 12, P < 0.001). Differences were particularly visiblefor procedures such as feet picking, where nine of the 11 PR horsesversus 0 of the 12 controls recalled the order and spontaneouslygave their four feet (Fisher’s exact test: P < 0.0001), or fittinga surcingle, where 10 of the 11 PR horses succeeded and five of the12 controls did (Fisher’s exact test: P < 0.01).

These results were confirmed 8 months after training with anunknown person, when PR horses appeared to respond to the vocalorder better than controls (success rate in PR: 10/11; in C: 6/12;Fisher’s exact test: P < 0.05) and they maintained immobility forlonger in response to that order (XPR � SE ¼ 58:2� 1:8 s,XC � SE ¼ 46:2� 5:7 s; Mann–Whitney U test: U ¼ 38, NPR ¼ 11,NC ¼ 12, P < 0.05). Again, PR horses accepted more handlingprocedures than controls (XPR � SE ¼ 4:5� 0:2,XC � SE ¼ 3:1� 0:4 of the six handling procedures; Mann–Whit-ney U test: U ¼ 23.5, NPR ¼ 11, NC ¼ 12, P < 0.01), with a persistingdifference in the acceptance of feet picking (acceptance rate in PR:5/11; in C: 0/12; Fisher’s exact test: P < 0.05). Horses’ performancein the memorization tests did not differ when tested with thefamiliar trainer or with an unfamiliar experimenter.

The unfamiliar technician spent more than twice the timerequired to administrate a deworming oral paste to controls than tothe PR horses (XPR � SE ¼ 8:0� 1:0 s, XC � SE ¼ 19:6� 8:9 s,Mann–Whitney U test: U ¼ 30.5, NPR ¼ 11, NC ¼ 12, P < 0.05),

PR C

3

2

Mea

n t

im

1

0

Figure 1. Mean time (h) required to complete training successfully �SE. PR: positivereinforcement; C: controls. ***P < 0.001.

70

60

50

40

30

20

10

0After 6 months

Familiar experimenterAfter 8 months

Unknown experimenter

Tim

e sp

ent

< 0.

5 m

fro

m t

he

exp

erim

ente

r (%

)

PRC

∗∗∗∗

∗∗

Figure 3. Mean percentage of time �SE spent at a distance of less than 0.5 m from theexperimenter during a ‘motionless person test’ performed 6 months (familiar exper-imenter) and 8 months (unknown experimenter) after the end of training. PR: positivereinforcement; C: controls. *P < 0.05; **P < 0.01; ***P < 0.001.

C. Sankey et al. / Animal Behaviour 79 (2010) 869–875 873

showing that PR horses were more able to generalize to noveltasks and humans than controls. Only three of 11 PR and six of 12 Clifted their heads to avoid contact when the technician approachedthe horse’s head with the syringe and only one horse (C group)backed up.

Human–Horse Relationship

Short-term impactPositive reinforcement appeared to have a positive impact on

the horses’ reactions to humans, even when measured outside thetraining context. Thus, before training, all horses spent over 90% ofthe time at a distance of 3 m or more from the experimenter duringthe motionless person test. After training, horses from the PR groupspent more time very close to the experimenter than controls (PR:78%; C: 27% at less than 0.5 m; Mann–Whitney U test: U ¼ 15,NPR ¼ 11, NC ¼ 12, P < 0.01; Fig. 2) and less time at a distance of3 m or more (PR: 7%; C: 36%; Mann–Whitney U test: U ¼ 33,NPR ¼ 11, NC ¼ 12, P < 0.05). None the less, horses from bothexperimental groups spent more time close to the experimenterafter training than they did before (PR: Xbefore � SE ¼ 4:3� 0:8%,Xafter � SE ¼ 78:0� 2:7%; Wilcoxon signed-ranks test:Z ¼ �2.8, N ¼ 12, P < 0.01; C: Xbefore � SE ¼ 0:0� 0:0%,Xafter � SE ¼ 27:4� 2:8%; Wilcoxon signed-ranks test: Z ¼ �2.4,N ¼ 12, P < 0.05; Fig. 2), showing that in the absence of anyconditioning strategy, mere habituation may occur.

Before the training began, the mean time required to toucha horse on the shoulder (approach-contact test) did not differbetween the experimental groups and was 44 � 4.5 s, whereasafter completion of training, PR horses were faster to acceptthe experimenter’s contact, even though all horses readilyaccepted it (mean latency to contact: XPR � SE ¼ 1:63� 0:2 s,XC � SE ¼ 4:25� 1:7 s; Mann–Whitney U test: U ¼ 23, NPR ¼ 11,NC ¼ 12, P < 0.01).

Long-term impactGeneral perception of humans. Large differences between PR andcontrol horses could be observed again from 6 and up to 8 monthsafter the end of training, despite no further human intervention(Fig. 3). PR horses approached the experimenter in half the time ofthat taken by controls, whether the experimenter was the familiartrainer (XPR � SE ¼ 71:7� 17:7 s, XC � SE ¼ 188:0� 24:0 s;Mann–Whitney U test: U ¼ 17, NPR ¼ 11, NC ¼ 12, P < 0.01)or an unfamiliar person (XPR � SE ¼ 78:9� 33:9 s,XC � SE ¼ 250:2� 28:9 s; Mann–Whitney U test: U ¼ 21, NPR ¼ 11,

∗∗∗

∗∗90

80

70

60

50

40

Tim

e sp

ent

< 0.

5 m

from

th

e ex

per

imen

ter

30

20

10

0Before After

PRC

Figure 2. Mean percentage of time �SE spent at a distance of less than 0.5 m from theexperimenter during a ‘motionless person test’ performed before and after training.PR: positive reinforcement; C: controls. *P < 0.05; **P < 0.01.

NC ¼ 12, P < 0.01). Again, they spent four times longer ata very close distance (�0.5 m: familiar experimenter:XPR � SE ¼ 59:8� 4:6 s, XC � SE ¼ 15:0� 5:6%; Mann–WhitneyU test: U ¼ 7.5, NPR ¼ 11, NC ¼ 12, P < 0.001; unfamiliarexperimenter: XPR � SE ¼ 39:0� 6:4%, XPR � SE ¼ 7:2� 3:2%;Mann–Whitney U test: U ¼ 17.5, NPR ¼ 11, NC ¼ 12, P < 0.01), andless time far (�3 m) from them (familiar experimenter:XPR � SE ¼ 30:2� 2:5%, XC � SE ¼ 71:5� 6:7%; Mann–Whitney Utest: U ¼ 7.5, NPR ¼ 11, NC ¼ 12, P < 0.001; unfamiliar experimenter:XPR � SE ¼ 45:4� 3:3%, XC � SE ¼ 75:5� 6:4%; Mann–Whitney Utest: U ¼ 20.5, NPR ¼ 11, NC ¼ 12, P < 0.01) than controls.

Moreover, PR horses were quicker to accept a human con-tact on the shoulder than controls (familiar experimenter:XPR � SE ¼ 4:9� 1:0 s, XC � SE ¼ 10� 2:5 s; Mann–Whitney Utest: U ¼ 23, NPR ¼ 11, NC ¼ 12, P < 0.01; unfamiliar experimenter:XPR � SE ¼ 2:8� 0:3 s, XC � SE ¼ 8:8� 3:1 s; Mann–Whitney Utest: U ¼ 20.5, NPR ¼ 11, NC ¼ 12, P < 0.01). Repeated positiveinteractions with humans through training thus led to a long-termgenerally positive perception of humans, thereby confirming thegeneralization abilities of horses.

Specific relationship. Nevertheless, the interactions during trainingled to a close partner-specific bond (Fig. 3). Thus, PR horses spentmore time close to the familiar than to the unfamiliar experimenter(Xfamiliar � SE ¼ 59:8� 4:6%, Xunfamiliar � SE ¼ 39:0� 6:4%;Wilcoxon signed-ranks test: Z ¼ 2.0, N ¼ 12, P < 0.05). They alsotended to display more affiliative behaviours, such as sniffing,with the familiar person than with the unfamiliar one, althoughthis was not quite significant (Xfamiliar � SE ¼ 12:9� 1:5%,Xunfamiliar � SE ¼ 6:5� 1:3%; Wilcoxon signed-ranks test: Z ¼ 2.5,N ¼ 12, P < 0.1). No such differences were observed in controls.Horses were thus able to build a relationship of specific bonds onthe basis of earlier repeated interactions.

DISCUSSION

The association of a reward with a learning task in an interac-tional context, which in this case was the training of young horsesby humans, induced positive reactions towards humans duringtraining. It also increased contact and interest, not only just aftertraining, but also several months later, despite no further interac-tion with humans. In addition, this ‘positive memory’ of humans

C. Sankey et al. / Animal Behaviour 79 (2010) 869–875874

extended to novel persons. Thus, it appears that horses were able tobuild and maintain a long-term memory of the relation withhumans on the basis of repeated interactions. But they were alsoable to recall the specific bond created with a specific partner(trainer). Overall, positive reinforcement enhanced learning andmemorization of the task itself.

Creating a positive learning situation appeared to benefit bothlearning and behaviour during the training sessions. This is nota new finding: in humans, hundreds of experimental studies havedemonstrated that systematic use of reinforcement can improveboth classroom conduct and the rate of learning: ‘when teachersreinforce students at a high rate, students (.) actually enjoylearning’ (Chance 1992, page 202). A phenomenon referred to in theliterature as the ‘fading affect bias’ states that emotions associatedwith pleasant events decrease in intensity less than the emotionsassociated with unpleasant events (Walker et al. 2003). From ourresults, it appears that horses are no different from humans: theybehave, learn and memorize better when learning is associated witha positive situation. Nevertheless, in this study, control animals alsolearnt fairly easily and even though training without a reward didnot have such strong positive effects on the relationship as positivereinforcement training, neither did it create a ‘negative’ memory,such as the one reported in studies focusing on a negative interac-tional context (Fureix et al. 2009; C. Sankey, M-A. Richard-Yris,S. Henry, C. Fureix, F. Nassur & M. Hausberger, unpublished data).

Using a food reward may have enhanced the animals’ motiva-tion to learn and helped to focus their attention on the task.Attentional and motivational processes are essential in any case oflearning and lower motivational levels, as observed in stereotypichorses, somehow diminish their capacity to learn a new task(Hausberger et al. 2007a).

Neurophysiological studies concerning the brain mechanisms ofrewarded behaviour have revealed a neuronal circuit consisting ofglutamatergic interconnections among the amygdala, nucleusaccumbens, and prefrontal cortex and dopaminergic afferents to allthree regions (Cardinal et al. 2002). The release of dopamine signalsthe circuit to initiate adaptive behavioural responses to a motiva-tional event, and in doing so it facilitates cellular changes thatestablish learned associations with the event (Jay 2003).

This may explain why the horses successfully associated humanpresence with their positive experience (reward), thus leading toa generalized long-term positive memory of humans. Horses clearlyexpress their positive/negative perception of humans in a general-ized way (Fureix et al. 2009), which may be based on daily inter-actions with specific partners (Hausberger & Muller 2002).Emotion-inducing interactions may lead to negative associations(C. Sankey, M-A. Richard-Yris, S. Henry, C. Fureix, F. Nassur &M. Hausberger, unpublished data) expressed in a variety of contexts(Fureix et al. 2009). In our study, horses experienced solely positiveinteractions and lived in a controlled environment without furtherinteraction with humans. It is likely that positive memories elicitedby training could be disturbed if some sort of negative interactionswere to occur.

Within their social group, horses have been shown to havestrong individual preferences for certain social partners (Feist &McCullough 1976) and mares are selective with respect to their ownfoal (Tyler 1972). Furthermore, recent work suggests that equinesare capable of discriminating between humans by using facialcharacteristics (Koba et al. 2004; Stone 2010), and that they are evenable to transfer such discrimination to photographic representa-tions of these humans (Tanida et al. 2005; Stone 2010). This studytends to support this; although positively reinforced horsesresponded consistently to the vocal command, whether given bythe familiar or the unfamiliar experimenter, they behaved differ-ently in the human–animal relationship tests. They seemed able to

discriminate between their familiar trainer and a new person, thusdisplaying different levels of bonding based on their previousexperience with different humans, similar to the gradient observedin their complex intraspecific social network (Lemasson et al. 2009;Proops et al. 2009). These findings illustrate that the cognitiveprocesses shared by domestic and wild equids provide a promisingcontinuity framework for ethologists (Linklater 2007).

While our results are consistent with previous studies showingconsiderable success in training animals to cooperate duringvarious handling or veterinary procedures using food rewards (e.g.Reinhardt 1991), one especially original aspect of this study lies inthe use of a vocal order (verbal stimuli from the experimenter);although the auditory sense is widely used in dog training (Hibyet al. 2004), the majority of horse-riding training is based on tactilesensations (pressure from bits, movements of riders’ legs, weightchange in the saddle; McCall 2007). The area of best sensitivity ofhorse hearing is broad and covers the range of the human voicebetter than does the dog’s (Heffner & Heffner 1983). This studyconfirms that horses are able to learn and memorize human wordsthat are auditory stimuli in association with a task.

Although horses are generally said to have excellent memories,the limits of this ability have not yet been explored. Marinier &Alexander (1994) have shown that horses remember a maze thatthey have learnt to criterion perfectly after 1 week and Wolff &Hausberger (1996) found that horses that have succeeded ina detour task or acquired an operant response to open a chest andfind food perform the same task faster after a 1-month interval.Only recently has it been reported that horses can rememberrelatively complex problem-solving strategies for a minimum of 7years and as long as or longer than 10 years (Hanggi & Ingersoll2009). In our study, long-term memory was first tested after a 6-month interval, but considering their recently demonstrated long-term memory capacity, it is likely that human–horse encountersmay be recalled for a much longer period. The experience of humanactions in foals also has an impact over at least several weeks ormonths (Henry et al. 2005, 2007, 2009).

This study confirms that relationships are created on the basis ofrepeated interactions. Their ‘valence’ influences the quality of therelationship (Hinde 1979). Our results show that these principlesapply to interspecific interactions, as previously suggested (Waib-linger et al. 2006; Hausberger et al. 2008). Moreover, it appears thatonce established, the relationship is memorized and can be used topredict future interactions even after a long separation. Thesefindings suggest remarkable social cognitive abilities that can betransposed from intraspecific to interspecific social contexts.

Acknowledgments

We thank the Director (Laurence Wimel) and staff of the ‘Stationexperimentale des Haras’, Chamberet. Thanks to Carole Fureix, ourunfamiliar experimenter. Carol Sankey was supported by a grant ofthe ‘Haras Nationaux’ and the ‘Region Bretagne’. This study wasfunded by the COST of the ‘Haras Nationaux’. This manuscriptbenefitted from comments from Dr W. L. Linklater and twoanonymous referees.

References

Altmann, J. 1974. Observational study of behavior: sampling methods. Behaviour,49, 227–267.

Bourjade, M., de Boyer des Roches, A. & Hausberger, M. 2009. Adult-youngratio, a major factor regulating social behaviour of young: a horse study. PLoSONE, 4, e4888.

Boyd, L. 1988. Ontogeny of behavior in Przewalski horses. Applied Animal BehaviourScience, 21, 41–69.

C. Sankey et al. / Animal Behaviour 79 (2010) 869–875 875

Boyd, L. & Keiper, R. 2005. Behavioural ecology of feral horses. In: The DomesticHorse. The Evolution, Development and Management of its Behaviour (Ed. byD. Mills & S. McDonnell), pp. 55–82. Cambridge: Cambridge University Press.

Boysen, S. 1992. Pongid pedagogy: the contribution of human–chimpanzee inter-actions to the study of ape cognition. In: The Inevitable Bond: ExaminingScientist–Animal Interactions (Ed. by H. Davis & D. Balfour), pp. 250–263. NewYork: Cambridge University Press.

Cameron, E. Z., Setsaasa, T. H. & Linklater, W. L. 2009. Social bonds betweenunrelated females increase reproductive success in feral horses. Proceedings ofthe National Academy of Sciences, U.S.A., 106, 13850–13853.

Cardinal, R. N., Parkinson, J. A., Hall, J. & Everitt, B. J. 2002. Emotion and moti-vation: the role of the amygdala, ventral striatum, and prefrontal cortex.Neuroscience Biobehavioural Review, 26, 321–352.

Chance, P. 1992. The rewards of learning. Phi Delta Kappan, 73, 200–207.Davis, H. 2002. Prediction and preparation: Pavlovian implications of research

animals discriminating among humans. ILAR Journal, 43, 19–26.Demaree, H. A., Everhart, D. E., Youngstrom, E. A. & Harrison, D. W. 2005. Brain

lateralization of emotional processing: historical roots and a future incorpo-rating ‘dominance’. Behavioral and Cognitive Neuroscience Reviews, 4, 3–20,doi:10.1177/1534582305276837.

van Dierendonck, M. C., Sigurjonsdottir, H., Colenbrander, B. &Thorhallsdottir, A. G. 2004. Differences in social behaviour between latepregnant, post-partum and barren mares in a herd of Icelandic horses. AppliedAnimal Behaviour Science, 89, 283–297.

Feh, C. 1987. Etude du developpement des relations sociales chez des etalons derace Camargue et de leur contribution a l’organisation sociale. Ph.D. thesis,University of Aix-Marseille.

Feh, C. 2005. Relationships and communication in socially natural horse herds. In: TheDomestic Horse. The Evolution, Development and Management of its Behaviour (Ed.by D. Mills & S. McDonnell), pp. 83–93. Cambridge: Cambridge University Press.

Feist, J. D. & McCullough, D. R. 1976. Behaviour patterns and communication inferal horses. Zeitschrift fur Tierpsychologie, 41, 337–373.

Fureix, C., Jego, P., Sankey, C. & Hausberger, M. 2009. How horses (Equus caballus)see the world: humans as significant ‘objects’. Animal Cognition, 12, 643–654.

Ghirlanda, S., Jansson, L. & Enquist, M. 2002. Chickens prefer beautiful humans.Human Nature: an Interdisciplinary Biosocial Perspective, 13, 383–389.

Hanggi, E. B. & Ingersoll, J. F. 2009. Long-term memory for categories and conceptsin horses (Equus caballus). Animal Cognition, 12, 451–462, doi:10.1007/s10071-008-0205-9.

Hausberger, M. & Muller, C. 2002. A brief note on some possible factors involved inthe reactions of horses to humans. Applied Animal Behaviour Science, 76, 339–344.

Hausberger, M., Gautier, E., Muller, C. & Jego, P. 2007a. Lower learning abilities instereotypic horses. Applied Animal Behaviour Science, 107, 299–306.

Hausberger, M., Henry, S., Larose, C. & Richard-Yris, M.-A. 2007b. First suckling:a crucial event for mother-young attachment? An experimental study in horses(Equus caballus). Journal of Comparative Psychology, 121, 109–112.

Hausberger, M., Roche, H., Henry, S. & Visser, E. 2008. A review of the human–horse relationship. Applied Animal Behaviour Science, 109, 1–24.

Heffner, R. S. & Heffner, H. E. 1983. Hearing in large mammals: horses (Equuscaballus) and cattle (Bos taurus. Behavioural Neurosciences, 97, 299–309.

Hemsworth, P. H., Coleman, G. J. & Barnett, J. L. 1994. Stimulus generalization: theinability of pigs to discriminate between humans on the basis of their previoushandling experience. Applied Animal Behaviour Science, 40, 129–142.

Hemsworth, P. H., Price, E. O. & Borgwardt, R. 1996. Behavioural responses ofdomestic pigs and cattle to human kind novel stimuli. Applied Animal BehaviourScience, 50, 43–56.

Henry, S., Hemery, D., Richard, M.-A. & Hausberger, M. 2005. Human–marerelationships and behavior of foals toward humans. Applied Animal BehaviourScience, 93, 341–362.

Henry, S., Briefer, S., Richard-Yris, M.-A. & Hausberger, M. 2007. Are6-month-old foals sensitive to dam’s influence? Developmental Psychobi-ology, 49, 514–521.

Henry, S., Richard-Yris, M.-A., Tordjman, S. & Hausberger, M. 2009. Neonatalhandling affects durably bonding and social development. PLoS ONE, 4, e5216.

Hiby, E. F., Rooney, N. J. & Bradshaw, J. W. S. 2004. Dog training methods: theiruse, effectiveness and interaction with behaviour and welfare. Animal Welfare,13, 63–69.

Hinde, R. 1979. Towards Understanding Relationships. London: Academic Press.Jay, T. M. 2003. Dopamine: a potential substrate for synaptic plasticity and memory

mechanisms. Progress in Neurobiology, 69, 375–390.Kruger, K. 2007. Behaviour of horses in the ‘round pen technique’. Applied Animal

Behaviour Science, 104, 162–170.Lansade, L. & Bouissou, M.-F. 2008. Reactivity to humans: a temperament trait of

horses which is stable across time and situations. Applied Animal BehaviourScience, 114, 492–508, doi:10.1016/j.applanim.2008.04.012.

Lemasson, A., Boutin, A., Boivin, S., Blois-Heulin, C. & Hausberger, M. 2009. Horse(Equus caballus) whinnies: a source of social information. Animal Cognition, 12,693–704.

Lensink, J., Boivin, X., Pradel, P., LeNeindre, P. & Veissier, I. 2000. Reducing vealcalves’ reactivity to people by providing additional human contact. Journal ofAnimal Science, 78, 1213–1218.

Linklater, W. L. 2000. Adaptive explanation in socio-ecology: lessons from theEquidae. Biological Reviews of the Cambridge Philosophical Society, 75, 1–20.

Linklater, W. L. 2007. Equine learning in a wider context: opportunities for inte-grative pluralism. Behavioural Processes, 76, 53–56.

McCall, C. A. 2007. Making equine learning research applicable to training proce-dures. Behavioural Processes, 76, 27–28, doi:10.1016/j.beproc.2006.12.008.

Marinier, S. L. & Alexander, A. J. 1994. The use of a maze in testing learning andmemory in horse. Applied Animal Behaviour Science, 39, 177–182.

Miklosi, A., Polgardi, R., Topal, J. & Csanyi, V. 1998. Use of experimenter-given cuesin dogs. Animal Cognition, 1, 113–121.

Murphy, J. & Arkins, S. 2007. Equine learning behaviour. Behavioural Processes, 76,1–13.

Pepperberg, I. M. 2000. The Alex Studies: Cognitive and Communicative Abilities ofGrey Parrots. Cambridge, Massachusetts: Harvard University Press.

Proops, L., McComb, K. & Reby, D. 2009. Cross-modal individual recognition indomestic horses (Equus caballus). Proceedings of the National Academy ofSciences, U.S.A., 106, 947–951.

Reinhardt, V. 1991. Training adult male rhesus monkeys to actively cooperateduring in-homecage venipuncture. Animal Technology, 42, 11–17.

Sigurjonsdottir, H., van Dierendonck, M. C., Snorrason, S. & Thorhallsdottir, A. G.2003. Social relationships in a group of horses without a mature stallion.Behaviour, 140, 783–804.

Silberman & Weingartner 1986. Hemispheric lateralization of functions related toemotion. Brain and Cognition, 5, 322–353.

Silk, J. 2007. Social components of fitness in primate groups. Science, 317,1347–1351, doi:10.1126/science.1140734.

Stone, S. M. 2010. Human facial discrimination in horses: can they tell us apart?Animal Cognition, 13, 51–61. 10.1007/s10071-009-0244-x.

Waiblinger, S., Boivin, X., Pederson, V., Tosi, M. V., Janczak, A., Visser, K. &Jones, R. B. 2006. Assessing the human–animal relationship in farmed species:a critical review. Applied Animal Behaviour Science, 101, 185–242.

Walker, R., Skowronski, J. & Thompson, C. 2003. Life is pleasant: and memoryhelps to keep it that way!. Review of General Psychology, 7, 203–210.

Waring, G. H. 2003. Horse Behavior, 5th edn. New York: Noyes.Wolff, A. & Hausberger, M. 1996. Learning and memorization of two different tasks

in horses: the effects of age, sex and sire. Applied Animal Behaviour Science, 46,137–143.