Segregation of Information About Emotional Arousal and Valence in Horse Whinnies

13 Pages·2016·0.99 MB·English

WWeellllBBeeiinngg IInntteerrnnaattiioonnaall
WWBBII SSttuuddiieess RReeppoossiittoorryy
2015
SSeeggrreeggaattiioonn ooff IInnffoorrmmaattiioonn AAbboouutt EEmmoottiioonnaall AArroouussaall aanndd VVaalleennccee
iinn HHoorrssee WWhhiinnnniieess
Elodie F. Briefer
Queen Mary University of London
Anne-Laure Maigrot
ETH Zürich
Roi Mandel
ETH Zürich
Sabrina Briefer Freymond
Agroscope - Swiss National Stud Farm
Iris Bachmann
Agroscope - Swiss National Stud Farm
See next page for additional authors
Follow this and additional works at: https://www.wellbeingintlstudiesrepository.org/acwp_asie
Part of the Animals Commons, Animal Studies Commons, and the Other Animal Sciences Commons
RReeccoommmmeennddeedd CCiittaattiioonn
Briefer, E. F., Maigrot, A. L., Mandel, R., Freymond, S. B., Bachmann, I., & Hillmann, E. (2015). Segregation of
information about emotional arousal and valence in horse whinnies. Scientific reports, 4.
This material is brought to you for free and open access
by WellBeing International. It has been accepted for
inclusion by an authorized administrator of the WBI
Studies Repository. For more information, please contact
[email protected].
AAuutthhoorrss
Elodie F. Briefer, Anne-Laure Maigrot, Roi Mandel, Sabrina Briefer Freymond, Iris Bachmann, and Edna
Hillmann
This article is available at WBI Studies Repository: https://www.wellbeingintlstudiesrepository.org/acwp_asie/131
Segregation of information about
OPEN
emotional arousal and valence
SUBJECTAREAS:
in horse whinnies
ANIMALBEHAVIOUR
ANIMALPHYSIOLOGY
ElodieF.Briefer1,Anne-LaureMaigrot1,2*,RoiMandel1,3*,SabrinaBrieferFreymond2,IrisBachmann2
Received &EdnaHillmann1
20October2014
Accepted 1ETHZu¨rich,InstituteofAgriculturalSciences,Universita¨tstrasse2,8092Zu¨rich,Switzerland,2Agroscope-SwissNationalStud
24March2015 Farm,LesLongsPre´s,P.O.Box191,1580Avenches,Switzerland,3KoretSchoolofVeterinaryMedicine,RobertH.SmithFacultyof
Agriculture,FoodandEnvironment,theHebrewUniversity,Rehovot76100,Israel.
Published
21 April2015
Studyingvocalcorrelatesofemotionsisimportanttoprovideabetterunderstandingoftheevolutionof
emotionexpressionthroughcross-speciescomparisons.Emotionsarecomposedoftwomaindimensions:
emotionalarousal(calmversusexcited)andvalence(negativeversuspositive).Thesetwodimensionscould
Correspondenceand
beencodedindifferentvocalparameters(segregationofinformation)orinthesameparameters,inducinga
requestsformaterials
trade-offbetweencuesindicatingemotionalarousalandvalence.Weinvestigatedthesetwohypothesesin
shouldbeaddressedto horses.Weplacedhorsesinfivesituationselicitingseveralarousallevelsandpositiveaswellasnegative
E.F.B.(elodie.briefer@ valence.Physiologicalandbehavioralmeasurescollectedduringthetestssuggestedthepresenceofdifferent
usys.ethz.ch) underlyingemotions.First,usingdetailedvocalanalyses,wediscoveredthatallwhinniescontainedtwo
fundamentalfrequencies(‘‘F0’’and‘‘G0’’),whichwerenotharmonicallyrelated,suggestingbiphonation.
Second,wefoundthatF0andtheenergyspectrumencodedarousal,whileG0andwhinnydurationencoded
*Theseauthors valence.Ourresultsshowthatcuestoemotionalarousalandvalencearesegregatedindifferent,relatively
independentparametersofhorsewhinnies.Mostoftheemotion-relatedchangestovocalizationsthatwe
contributedequallyto
observedaresimilartothoseobservedinhumansandotherspecies,suggestingthatvocalexpressionof
thiswork.
emotionshasbeenconservedthroughoutevolution.
Expressionofemotionsandperceptionofemotionalstatesplayanimportantroleinsocialspecies.Indeed,
emotionexpressioninformsindividualsabouttheprobableintentionofbehaviorsofothersandtherefore,
regulates social interactions1. Vocal expression of emotions has been extensively studied in humans
(‘‘affectiveprosody’’2,3).However,humanvoicealsodependsonsocio-culturalandlinguisticconventionsthat
canactasconfoundingfactorsinthestudyofaffectiveprosody4.Inmostnon-humananimals,vocalizationsare
assumedtobeunderlowervoluntarycontrolthaninhumans.Animalvocalizationsshouldthusreflectemotions
moredirectlythanhumanvoice5,andcouldserveasvaluablemodelsforstudiesonhumanaffectiveprosody.
Theycouldalsoserveasidealnon-invasivetoolstoassessemotionsinanimals,inwhichthesubjective,conscious
componentofemotionscannotbeaccessed6.
A promising approach to study animal emotions is through their two main dimensions (‘‘dimensional
approach’’7);1)emotionalarousal(excitation,e.g.calmversusexcited)and2)emotionalvalence(negativeor
positive, e.g. sad versus happy)8. Vocal correlates of emotional arousal have been relatively well studied (see
reviews6,9).Vocalizationsusuallybecomelonger,louder,andareproducedatfasterrates,withhigherandmore
variablefrequencieswhenarousalincreases6.Bycontrast,thereisconsiderablylessknowledgeonvocalindicators
of valence (i.e. differentiating between negative and positive state)6. Several types of vocalizations have been
showntoindicateeitherpositiveornegativeemotions10–12.However,changesinvocalparametersaccordingto
emotional valence have rarely been investigated6. Another limitation of the research on vocal expression of
emotionsinanimalsisthatveryfewstudiesinvestigatedvocalcorrelatesofbothemotionalarousalandvalence
inthesamespecies(butsee13–15).Inaddition,theemotionalsituationsthathavebeenusedoftendifferinboth
their valence and their arousal, and the effects of these two dimensions on vocal parameters are not tested
independently.Therefore,morestudiesareneededtoinvestigatehowemotionsareencodedinvocalizations,
andtoadvanceourunderstandingofthephylogeneticcontinuityofvocalcorrelatesofemotionalarousaland
valencebetweenhumansandotheranimals.
SCIENTIFICREPORTS |4:9989|DOI:10.1038/srep09989 1
www.nature.com/scientificreports
We hypothesized that, in order to be effectively encoded within triggered by our situations (i.e. indicative of emotional arousal)29. In
vocalizations,cuesto emotional arousalandvalencecould be segre- theabsenceofwell-establishedvalenceindicators,weinferredtheval-
gatedindifferentacousticfeaturesorintemporallydistinctvocalseg- enceofoursituationsfromknowledgeofthefunctionofemotionsand
ments(segregationsofinformation16).Segregationofinformationsuch on horse behavior. We focused our analyses on whinnies, which are
asindividualidentityandgroupmembership,bodyweightorsizeand producedinbothseparation(negativevalence)andgreeting(positive
condition,socialstatusandhormonalstate,existsinthevocalizationsof valence) contexts to maintainorregaincontactwith affiliates or off-
severalspecies(e.g.rockhyrax,Procaviacapensis17;bandedmangoose, spring21,22. We tested the hypothesis that the emotional arousal and
Mungosmungo18).Alternatively,arousalandvalencecouldbeencoded valenceexperiencedbyhorsesareindicatedbyparticularvocalprofiles.
inthesameparameters,inwhichcaseatrade-offwouldexistbetween Ourvocalanalysessystematicallyrevealedtwofundamentalfrequencies
thetwodimensions,resultinginonedimensionbeingmoreaccurately in all whinnies, suggesting biphonation, a rare phenomenon among
communicatedthantheother(e.g.identityandqualityinfallowdeer, mammals that had not been described in previous studies in horse
Damadama,sexuallyselectedvocalizations19). vocalisations21,23,24.Wethusfirstruledoutpotentialalternativeexplana-
Totestwhethertheencodingofthetwoemotionaldimensionsis tionstobiphonation,beforetestingtheeffectofemotionalarousaland
inaccordancewiththesegregationsofinformationhypothesis16,orif valence on vocal parameters, including those related to the two fun-
theyareencodedinthesameparameters(trade-offhypothesis19),we damental frequencies. Additionally, in order to confirm underlying
investigated vocal correlates of emotional arousal and valence in emotions,wetestediftheresultingemotionalarousallevels,aswellas
domestichorses,Equuscaballus.Horsesareverysocialanimalsthat, thepresupposedvalenceofthesituations,wereaccompaniedbyphysio-
inthewild,liveinharems(stallion,femalesandfoals)orinbachelor logical and behavioral changes measured during the tests30,31. We
bands(youngoroldstallionswithoutaharem)20.Vocalexpressionof defined the parameters that changed according to increased arousal
emotions should benefit horses by regulating social interactions levelsasreliablecuestoarousal.Similarly,wedefinedtheparameters
within groups. Horse vocalizations have been rarely investigated. thatchangedconsistentlyfromnegativetopositivevalenceasreliable
This species produces several types of calls; whinnies, nickers, cuestovalence29.
squeals, blows, snores, snorts, roars, and groans21,22. Squeals have
been shown to contain information about dominance rank23, and Resultsanddiscussion
whinnies about sex, body size and individuality24. A recent study We tested 20 privately owned horses of various breeds and age,
showedthattheoverallstructureofwhinniesdiffersbetweennega- housed in five different farms (3–5 horses per farm; Table 1). We
tiveandpositivesituations25.However,thisstudydidnotspecifically designedfoursituationspotentiallyelicitingdifferentlevelsofemo-
testwhichvocalparametersareaffectedornotbyvalence. tional arousal and characterized by negative or positive valence,
Wecombinednewframeworksrecentlyadaptedfromhumanto whichwerelikelytotriggerwhinnies.Thesesituationsinvolvedsepa-
animalresearchtoanalyzevocalizations(source-filtertheory26)and ration(supposedlyofnegativevalence20)andreunion(supposedlyof
emotions8.Weplacedhorsesinonecontrolsituationandfoursocial positivevalence32)witheitherallgroupmembers(supposedlyhigh
situations,triggeringvariouslevelsofemotionalarousalanddiffer- emotionalarousal)oronlyonegroupmember(supposedlymoderate
entvalence.Heartrateisawell-establishedindicatorofphysiological emotional arousal). In the negative situation ‘‘All-Leave’’, all the
stress27,whichislinkedtoemotionalarousalduringsituationsassoc- other horses from the farm (2–4 horses depending on the farms;
iatedwithbothpositiveandnegativevalence28.Wethusmeasured hereafter ‘‘group members’’) were removed, while the subject
heartratetodeterminethearousalofoursituations,andassumed waskeptinitshomeboxorpaddockalone.Inthepositivesituation
that variations in heart rate were linked to underlying emotions ‘‘All-Return’’,allthegroupmemberswerebroughtbacktowardsthe
Table1|Characteristicsofthehorsesusedintheexperiment;breed,sex(F5female;G5gelding),age(in2013),bodyweight,aswellas
thenumberofwhinniesthatwereanalyzedforeachhorse.
Farm Horse Breed Sex Age(years) Weight(kg)* Numberofwhinnies
1 1 SwissPony G 20 338 4
2 EnglishThoroughbred G 31 416 21
3 SwissHalfbred F 23 569 35
4 SwissHalfbred G 7 502 22
2 1 SwissHalfbred F 16 498 8
2 SwissHalfbred F 15 594 7
3 SwissHalfbred F 7 536 2
4 IrishSportHorse F 23 526 2
3 1 Akhal-Teke F 21 404 9
2 DartmoorPony F 9 267 12
3 CamargueHorse F 14 358 18
4 QuarterHorse F 12 499 24
4 1 FrenchSaddlebred G 23 593 29
2 WelshPony G 12 477 25
3 SwissHalfbred F 7 498 5
5 1 SwissHalfbred F 10 525 0
2 SwissHalfbred G 10 508 10
3 EnglishThoroughbred F 22 516 0
4 SwissHalfbred G 19 403 32
5 ComtoisHorse G 6 560 2
*Horsebodyweightwasestimatedfollowing64.
SCIENTIFICREPORTS |4:9989|DOI:10.1038/srep09989 2
www.nature.com/scientificreports
predators.Ontheotherhand,reunionwithgroupmemberstriggers
greetingvocalizationsbetweenaffiliatedpairsofhorses21,22,andthis
situationcouldenhancefitnessbylesseningexposuretopredators.
Furthermore, the small numbers of individuals per farm (3 to 5
horses)inourstudyensuredstrongfamiliarityandpotentiallyclose
bonds between horses. The valence of the situations was thus
assumedtobenegativeforthesituationsinvolvinggroupmembers
leaving(All-LeaveandCompanion-Leaves),neutralfortheControl
situation (Control), and positive for the situations where group
members were coming back (All-Return or Companion-Returns).
Wethenverifiedthatthearousallevelsandvalenceofoursituations
were accompanied by consistent physiological and behavioral
changesmeasuredduringthetests(seeSupplementaryResults).
Vocalstructureofwhinnies.Whinniesarethelongest,loudestand
mostcommonhorsevocalization.Threepartshavepreviouslybeen
described; 1) the ‘‘introduction’’, which is tonal and high in
frequency; 2) the ‘‘climax’’, which is a long, often frequency and
amplitude modulated part; and 3) the ‘‘end’’, which is low in
frequencyandamplitude,andcomposedofapulse-trainstructure
(Figure2;AudioS1)24.Accordingtothesource-filtertheoryofspeech
production34,vocalizationsaregeneratedbyvibrationsofthevocal
folds (source, determining the fundamental frequency), and are
subsequently filtered by the supralaryngeal vocal tract (filter,
producing amplified frequencies called ‘‘formants’’35). In order to
extract both source- and filter-related vocal parametersof whinnies,
Figure1|Heartrateasafunctionoftheemotionalsituations.Heartrate weassumedthatvocalproductioninhorsesisfundamentallysimilarto
(rawvalues)foreachoftheexperimentalsituations;boxplot:the humans and to other mammals on which this framework has been
horizontallineshowsthemedian,theboxextendsfromthelowertothe successfullyapplied26.Intotal,weextracted19source-andfilter-related
upperquartileandthewhiskersto1.5timestheinterquartilerangeabove vocal parameters as well as intensity and duration measures (see
theupperquartileorbelowthelowerquartile;circlesindicateoutliers. Table 2 for abbreviations, and Supplementary Methods for analysis
Sameletters(a,b,c)indicatethatsituationsdidnotsignificantlydiffer description)from267whinniesproducedby18horses(2horsesdid
(linearmixed-effectsmodelscomparedwithlikelihood-ratiotests;log notproducewhinnies;Table1).
transformedheartratevaluescontrolledforsex,ageandbodyweightofthe Previous studies on whinnies only measured the fundamental
horses,orderofthesituations(AllorCompaniontestsfirst),dayof frequency in the highly tonal introduction part (easily identifiable
experiment,individualandfarmidentity).Basedontheseresults, fundamental frequency), and argued that this parameter is not
situationsmarkedwith‘‘a’’receivedanemotionalarousallevelof0, always identifiable in the climax and end parts24. However, after
situationsmarkedwith‘‘b’’receivedanarousallevelof1,includingAll- carrying out detailed analyses, we noticed that the climax part is
Returnsituation,andtheAll-Leavesituation(markedwith‘‘c’’)was always (i.e. 100% of whinnies) composed of a second, lower fun-
consideredofarousallevelof2.Resultingemotionalarousallevels(0–2) damental frequency and its corresponding harmonics (Figure 2;
andvalence(Negative,NeutralandPositive)correspondingtothe AudioS1),suggestingbiphonation(i.e.presenceoftwoindependent
situationsareindicatedbeloweachbox(arousallevel/valence). fundamentalfrequenciesthatarenotintegermultiples36).Asbipho-
nationisararephenomenoninmammalsandhadnotbeenprev-
subjectfollowingtheAll-Leavesituation.Thetwoothernegativeand iously described in horses, we first carried out a detailed vocal
positivesituationsweresimilartothefirstones,exceptthatonlyone analysistoruleoutalternativeexplanationstobiphonation,before
groupmember(the‘‘Companion’’,andnotallgroupmembers)was testingtheeffectofemotionalarousalandvalenceonthestructureof
takenawayfromthesubject(‘‘Companion-Leaves’’)andthenwalked whinnies.Wemeasuredthetwofundamentalfrequenciesthrough-
back towards it (‘‘Companion-Returns’’). The four situations were out the introduction and climax parts. These frequencies are not
comparedtoacontrolsituation(‘‘Control’’),duringwhichthesub- clearlyvisibleintheendpartofthewhinnies,whichischaracterized
jectandallothergroupmemberswereintheirhomeboxorpaddock byamorechaotic-likepulse-trainstructure,withpulsespotentially
andwerenotmanipulated. correspondingtothevibrationsofthevocalfolds37,suggestingadrop
Theactualemotionalarousalleveltriggeredbyoursituationswas in fundamental frequency (Figure 2). The lower fundamental fre-
assessedfromthehorseheartrate,measuredduringthetests29.The quency(399.22699.39 Hz,range552–1050 Hz,n5267whin-
analysisofheartrateasafunctionofemotionalsituationsrevealed nies),startingatthebeginningoftheclimax,ishereafterreferredas
three emotional arousal levels; 0 for Control and Companion- ‘‘F0’’.Wereferredtothehigherfundamentalfrequency(1543.266
Returns, 1 for Companion-Leaves and All-Return, and 2 for 326.45 Hz,range5493–3012 Hz,n5260whinnies),startingatthe
All-Leave(Figure1;heartrate,arousallevel0:43.93611.58beats/ beginningofthewhinny,as‘‘G0’’(Figure2)36.
min; arousal level 1: 50.27 6 16.96 beats/min; arousal level 2: In order to ruleout alternative explanations to biphonation, we
56.32623.46beats/min;seeMethodsformoredetailsandstatistics). firstverifiedthatF0doesnotsimplyresultfromaregistertransition
Thevalenceofthesituationswasinferredfromknowledgeofthe (i.e.abruptchangeinfundamental frequency)38,39.Inthiscase,G0
functionofemotionsandofhorsebehavior.Positiveemotionsresult andF0shouldnotoverlap.However,inhorsewhinnies,G0,which
from encounters with rewarding stimuli that enhance fitness. By startsatthebeginningofthewhinny,canbeobservedandmeasured
contrast,negativeemotionsaretriggeredbypunishingstimulithat throughouttheintroductionandclimax,evenafterF0appearsatthe
threaten fitness8. Horses are highly social animals and separation beginningoftheclimax(Figures2and3).G0andF0thusoverlap
fromconspecificsisthusstressfulforthem20,33.Thissituationwould, over 79.40 6 20.52% of the call on average ((Dur-DurIntro/Dur);
inthewild,potentiallythreatenfitnessthroughgreaterexposureto n5267whinnies;TableS1).Secondly,weverifiedthatF0andG0are
SCIENTIFICREPORTS |4:9989|DOI:10.1038/srep09989 3
www.nature.com/scientificreports
Table2|Abbreviationsforthevocalparameters.
Abbreviation Parameter
Dur(s) Durationofthewhinny
DurIntro(s) Durationoftheintroductionpart
G0start(Hz) FrequencyvalueofG0atthestartofthewhinny
G0max(Hz) MaximumG0frequencyvalueacrossthewhinny
G0mean(Hz) MeanG0frequencyvalueacrossthewhinny
F0start(Hz) FrequencyvalueofF0atthestartoftheclimaxpart
F0max(Hz) MaximumF0frequencyvalueacrossthewhinny
F0mean(Hz) MeanF0frequencyvalueacrossthewhinny
TimeF0max(%) PercentageofthetotalwhinnydurationwhenF0ismaximum
AmpVar(dB/s) Cumulativevariationinamplitudedividedbythetotalwhinnyduration
AMrate(s-1) Numberofcompletecyclesofamplitudemodulationpersecond
AMextent(dB) Meanpeak-to-peakvariationofeachamplitudemodulation
Q25%(Hz) Frequencyvalueattheupperlimitofthefirstquartilesofenergy
Q50%(Hz) Frequencyvalueattheupperlimitofthesecondquartilesofenergy
Q75%(Hz) Frequencyvalueattheupperlimitofthethirdquartilesofenergy
F1mean(Hz) Meanfrequencyvalueofthefirstputativeformant
F2mean(Hz) Meanfrequencyvalueofthesecondputativeformant
F3mean(Hz) Meanfrequencyvalueofthethirdputativeformant
F4mean(Hz) Meanfrequencyvalueofthefourthputativeformant
Figure2|Spectrogramsoftwowhinnies.(a)and(b)spectrograms(above),oscilliogams(below)and100Hzcepstral-smoothedspectra(right;
frequencyversusamplitude)oftwowhinniesproducedbydifferenthorses.Whinny(a)containsthethreetypicalparts;theintroduction,theclimaxand
theend,whereputativeformants(F1-F4)couldbemeasured.Whinny(b)containsonlytheintroductionandtheclimax,whichisfrequencymodulated.
Theendpartisnotpresentandputativeformantscouldnotbemeasuredinthiskindofwhinny.Forbothwhinnies(aandb),F0(lowerfundamental
frequency)andG0(higherfundamentalfrequency)areindicatedonthespectrogramsandcepstral-smoothedspectra,aswellastheharmonics
(i.e.multiples)ofF0(Hn(F0))andofG0(Hn(G0)).Asshownonthecepstral-smoothedspectra,theharmonicsofF0andG0donotoccuratthesame
frequencies.Thesewhinniesareavailableasaudiofiles(AudioS1).
SCIENTIFICREPORTS |4:9989|DOI:10.1038/srep09989 4
www.nature.com/scientificreports
Figure3|Negativeandpositivewhinnies.(a)and(c)spectrograms(above)andoscilliogams(below)ofwhinniesproducedduringthenegative
situationsbytwodifferenthorses;(b)and(d)spectrograms(above)andoscilliogams(below)ofwhinniesproducedduringthepositivesituationsbythe
sametwohorses(differenthorsesthanforFigure2).F0(lowerfundamentalfrequency)andG0(higherfundamentalfrequency)areindicated,aswellas
thefirstharmonics(i.e.multiples)ofF0(H1(F0))andofG0(H1(G0)).Positivewhinnies(b)and(d)areshorterindurationandhavealowerG0
(startandmean)thannegativewhinnies(a)and(c).Thesewhinniesareavailableasaudiofiles(AudioS2).
notharmonicallyrelated(i.e.notintegermultiplesofeachother).For horses).ThissuggeststhathorseswithahigherG0donotnecessarily
instance,F0couldbeasub-harmonicofG0thatappearsfollowinga have a higher F0 and vice versa. G0 and F0 contours measured
bifurcation (i.e. change in regime from normal phonation, where in 71 whinnies, in which we had been able to extract these
vocal folds are synchronized, to sub-harmonic regime, where one contoursthroughouttheentireintroductionandclimax,weresignifi-
vocalfoldismovingfasterthantheotherandhavingtwiceormore cantly correlated for 63 of them (positively correlated in
theperiodoftheother40).Iftheywereharmonicallyrelated,G0and 56whinniesandnegativelycorrelatedin7whinnies).Whenthecor-
F0 should have been positively correlated with each other both relation was significant (p , 0.034), r2 ranged from 0.06 to 0.92
betweenhorses(i.e.betheproductofoneanother;G05n*F0), (Spearman’s rank correlation: r25 0.51 60.23, n 5 63 whinnies).
aswellasovertimewithineachwhinny(i.e.havethesamefrequency F0 and G0 are thus neither the product of one another, not fully
overtime(‘‘contour’’);r2betweenG0andF0contourapproaching1)41. correlated over time (r2 close to 1), indicating that they are not
G0was4.3962.24timeshigherthanF0onaverage(n5260whin- harmonicallyrelated.
nies).TheaverageG0meanandF0meanwerenotcorrelatedbetween Next,weruledoutvariousspectrogramartifacts(aliasing,clipping
horses (Spearman’s rank correlation: r2 5 0.002, p 5 0.88, n 5 18 and reverberation), which could be mistaken for biphonation or
SCIENTIFICREPORTS |4:9989|DOI:10.1038/srep09989 5
www.nature.com/scientificreports
othernon-linearphenomena(e.g.subharmonicsanddeterministic Our analyses of vocal parameters as a function of the emotions
chaos36).Aliasingoccurswhenthevocalizationcontainsfrequency triggeredbytheexperimentalsituationsrevealed12parametersthat
componentsaboveonehalfofthesamplingfrequency.Thesecom- wereinfluencedbyemotionalarousal(Table3;seeTableS1forraw
ponentsappearasimagefrequenciesinthespectrogramandcanbe values).Dur(duration),G0start,G0max,G0mean(G0-relatedpara-
confoundedforbiphonation.Inourcase,thisartifactcanberuled meters),F0start,F0max(F0-relatedparameters),Q24%,Q50%and
out as oursampling frequency was 44100 Hz, which is more than Q75% (energy quartiles) all increased with arousal levels, while
twicethehighestfrequenciesobservedinhorsewhinny(meanhigh- AMextendandAmpVar(amplitude-relatedparameters)decreased
est frequency measured onaspectrum 51367762947, range 5 (see Table 2 for abbreviations and definitions). F0mean was also
9153–21157 Hz,n52whinniesperhorse,18horses(36whinniesin affectedbyarousal,butdidnotchangeconsistentlywithincreasing
total)). Clipping results from severe overloading and produces arousallevels(level0.1,2).F4mean(fourthputativeformant)
abruptchangesinthespectrogram,whichmayresemblebifurcations. tended to increase, but not significantly, with arousal. Ten para-
Yet,wecontrolledforoverloadingbyadjustingtherecordinglevel meters were influenced by valence (Table 3; see Table S1 for raw
duringourrecordings,andbyexcludingfromouranalyseswhinnies values).Dur,G0start, G0max,G0mean,F0max, Q24%,Q50% and
that were overloaded (see oscillograms in Figures 2 and 3). Q75% all decreased from negative to positive valence, while
Finally, reverberation is due to the resonance properties of the AMextend and AmpVar increased. F0start and AMrate tended to
environment and appears as an artificial sound prolongation36. decreasefromnegativetopositivevalence,buttheseeffectswereonly
Thisartifactcanalsoberuledoutinourcase,asthesecondfrequency marginallysignificant(0.05,p,0.06).Theotherparameterswere
(F0)alwaysappearsatthebeginningoftheclimaxpart,andnotatthe neitheraffectedbyarousal,norbyvalence(Table3).
endofthewhinnyorwhentwowhinniesoverlap. When a parameter was significantly affected by both emotional
In addition to the previously mentioned spectrogram artifacts, arousalandvalence,weusedamodelselectionprocedurebasedon
sidebands (or side frequencies) that are non-harmonically related Akaike’sinformationcriterion(AIC),toidentifywhichofarousalor
tothefundamentalfrequencycanappearonthespectrogramsym- valencebestexplainedthevariationineachparametervalue29.We
metricallyoneachsideofthefundamentalanditsharmonics,asa used AIC adjusted for small sample sizes (AIC ), because AIC
C C
resultofstrongandrapidamplitudemodulations42,43.However,G0 converges to AIC as the sample size increases and should thus be
andF0couldstillclearlybeobservedinwhinniesafterwemodified usedbydefault45.Themodel(includingarousalorvalence)withthe
thembyartificiallyremovingtheamplitudemodulation(FigureS1). lowest AIC can be considered as the best model46. This model
C
Similarly,sidebandsoneachsideofthefundamentalfrequencyand selectionprocedurerevealedthatthevariationinF0max,AmpVar,
itsharmonicscanappearinnarrow-bandspectrogramsasaresultof Q24%,Q50%andQ75%wasbetterexplainedbyarousalthanvalence
rapid frequency modulation44. Yet, in this case, such sidebands (Table4).Conversely,thevariationinDur,G0start,G0max,G0mean
should always be apparent at equal distance above and below andAMextendwasbetterexplainedbyvalencethanarousallevels
G0 and its harmonics. The absence of this clear pattern in most (Table4).ForF0max,AmpVar,G0maxandAMextend,thediffer-
whinnies(Figures2and3),aswellasthelackofappearanceofrapid encebetweentheAIC valuesofthetwomodels(DAIC )waslower
C C
frequency modulations when changing the bandwidth of spectro- than3,indicatingthatthemodelsincludingarousalandvalencewere
grams (narrow-band to broad-band)44, allowed us to also rule out competitive.Themodelincludingarousalhad78%and62%chance
thishypothesis.Therefore,thereisstrongevidencesuggestingthatF0 to be the best model for F0Max and AmpVar, respectively. The
andG0arenotartifacts. model including valence had 81% and 73% chance to be the best
OurresultsshowthatF0andG0arenotgeneratedbyartifacts,nor modelforG0maxandAMextent,respectively.ForQ25%,theDAIC
C
aretheyharmonicallyrelated,suggestingrealbiphonationinwhin- waslowerthan4,indicatingthatthemodelincludingvalencehadless
nies.G0waspresent,inadditiontoF0,inallwhinnies,indicatingthat supportbythedatathanthemodelincludingarousal,whichhad88%
thepresenceoftwofundamentalfrequenciesisacommonfeatureof chancetobethebestmodel.
this type of call. Whinnies hence differ from the vocalizations of Tosummarize,accordingtoourcriteria,F0start,Q50%andQ75%
other mammalian species, in which biphonation, mostly found in werereliablecuestoarousal,becausetheywerechangingconsistently
youngprimatesandcanids,appearssporadically(e.g.44%ofthecalls witharousalandwereclearlymoreaffectedbyarousalthanvalence
indhole,Cuonalpinus41;60%ofthecallsofAfricanwilddog,Lycaon (DAIC .7).Bycontrast,Dur,G0startandG0meanwereclearly
C
pictus36).Thevocalproductionmechanismsresponsibleforbipho- moreaffectedbyvalencethanarousal(DAIC .7)andwerethere-
C
nationinwhinniescouldbethefollowing;involvementofvocalfold forereliablecuestovalence.Similaranalysescarriedoutonphysio-
extensions,vortex-sheddingattheglottalconstriction,source-tract logicalandbehavioralparametersmeasuredduringthetestsshowed
couplingor,moreprobably,asynchronousvibrationpatternofthe that the emotional arousal and valence of our situations were
vocal folds36,41. Further detailed anatomical investigation of horse reflected by physiological and behavioral changes in the horses,
vocal production apparatus are required to understand the mode suggestingunderlyingemotions(seeSupplementaryResults).
ofproductionofF0andG0. Inordertoexamineclusteringamongparameters,wethencarried
out a principal component analysis (PCA), including all the vocal
Vocalcorrelatesofemotionalarousalandvalence.Afterdescribing parametersmeasuredinfiverandomlyselectedwhinniesperhorses
thevocalstructureofhorsewhinnies,wecarriedoutlinearmixed- (n59horses,i.e.horsesthatproducedatleast5whinniesinwhich
effectsmodelstotesttheeffectofemotionalarousalandvalenceon all 19 parameters were successfully measured; Table 1). The PCA
allthevocalparametersmeasured,includingthoserelatedtoF0and generated six principal components (PCs) that exceeded Kaiser’s
G0.Subjects never whinnied duringtheControl situation (neutral criterion(eigenvalues .1)and accountedfor 83%ofthevariation
valence), and the effect of valence on vocal parameters could thus intheoriginaldataset.Amongthesixvocalparametersselectedas
only be assessed by comparing situations of negative and positive good cues to horse emotions, those indicating arousal (F0start,
valence(Table3).Becausethearousallevelsandvalenceattributed Q50%,Q75%)wereclusteredinPC1,whilethoseindicatingvalence
to the analyzed whinnies were correlated (Spearman’s rank (G0start and G0mean) were clustered in PC2. Finally, Dur, which
correlation: r2 5 0.47, p , 0.0001; n 5 267 whinnies), the effects alsoindicatedvalence,loadedhighlyonPC6(TableS2).
ofthetwoemotionaldimensionsonvocalparameterswerenottested Ourstudyrevealedthatwhinniesproducedduringhighemotional
in the same models45. Instead, we ran one first set of models with arousalsituationshaveahigherF0(F0start)andenergydistribution
arousallevelasanexplanatoryfactor,andanothersecondsetwith (i.e.energyquartiles;Q50%andQ75%)thanthoseproducedduring
valence29. low arousal situations. These results could be explained by 1) an
SCIENTIFICREPORTS |4:9989|DOI:10.1038/srep09989 6
www.nature.com/scientificreports
es,orderofthex2d-ratiotests:withemotionalorvice-versa). p .***0.064.***.0.005.0.0008.0.058.0.0420.330.23,0.010.0.057,0.006.0.033.0.041.0.00030.830.720.780.46
modelscontrolledforsex,ageandbodyweightofthehorsrelistedinTableS1),alongwithstatisticalresults(likelihoo,,,cant(0.050.06)effects(‘‘’’indicatesanincreasepectwasnotconsistent,i.e.increasefollowedbydecreaseers. Valence NegativePositive x2(n)MeanSDMeanSD1 20.070.220.420.5624.03(267)20.010.160.030.203.43(267)232.69108.95235.71275.4219.41(260)212.5941.98144.13161.338.00(260)213.3344.44127.37141.9011.31(260)26.1298.2218.62108.473.59(267)24.8814.8576.2273.624.14(267)22.1571.746.5558.630.95(267)20.5313.881.6114.411.43(267)20.702.128.179.736.57(267)20.101.640.311.843.63(267)20.030.080.330.317.69(267)218.4056.04266.77289.644.57(267)217.3552.85251.74316.344.19(267)20.020.060.150.1913.31(267)20.4832.340.9429.530.05(97)20.8144.982.0648.850.13(97)20.5037.391.5436.950.08(131)21.7043.514.3136.930.55(106)
nearmixed-effects6SD;rawvaluesadmarginallysignifindicatesthattheeffonsoftheparamet p ,0.0030.27,0.0005,0.024,0.048,0.003,0.0100.020NC0.28.0.0060.36.0.017,0.003,0.0006,***0.700.450.57NC0.053
usallevelandvalenceonvocalparameters.Residualsofthelisfirst),dayofexperiment,individualandfarmidentity(mean#ues).Thedirectionisindicatedforthesignificant(0.05)anp.opositivevalence,whereas‘‘’’indicatesadecrease;NCi***,bold(indicates0.0001).SeeTable2forabbreviatip Arousal 12 x2(n)DMeanSDMeanSD1 20.030.0600.490.428.81(267)220.030.170.020.161.22(267)224.2140.039237.38259.7912.22(260)27.2514.703148.83147.735.09(260)4.927.891144.18121.303.90(260)16.7417.721109.6693.999.00(267)29.4210.65370.0580.586.66(267)210.019.74362.2374.805.42(267)30.2814.540.4513.311.17(267)20.781.2059.607.407.57(267)20.001.610.051.710.84(267)20.020.0420.340.315.71(267)231.8342.159287.20266.578.54(267)225.3643.994268.96263.6811.71(267)20.030.0460.170.1526.65(267)2200.6028.190.8334.430.15(97)2415.2047.439.4541.670.56(97)240.6340.991.8735.000.31(131)258.4239.3213.2244.783.75(106)
Table3|Effectofemotionalarosituations(AllorCompaniontestvalues,samplesize()andvalnparousallevelsorfromnegativetSignificantresultsareshownin 0 ParameterMeanS 2Dur0.150.6DurIntro0.010.22G0start84.69239.22G0max35.73156.72G0mean44.03137.22F0start17.0392.52F0max12.0871.02F0mean6.9859.22TimeF0max2.4815.1AmpVar2.199.52AMrate0.191.9AMextent0.080.32Q25%63.79241.72Q50%89.50273.92Q75%0.070.1F1mean3.2036.32F2mean17.3641.9F3mean4.6732.02F4mean1.3336.5
SCIENTIFICREPORTS |4:9989|DOI:10.1038/srep09989 7
www.nature.com/scientificreports
Alternatively,vocalcorrelatesofemotionalarousalcouldhavebeen
Table4|Results of the model comparisons for vocal parameters
highly conserved, while those of valence could be more species
affected by both emotional arousal and valence. The fit of the specific. This latter hypothesis is supported by the fact that none
modelsisassessedbyAkaike’sinformationcriterionadjustedfor ofthevocalcuestovalencefoundingoats(Caprahircus),inwhicha
smallsamplesizes(AICC).Thebestmodel(i.e.arousalorvalence; similarexperimentalprocedurewasused29,aresimilartothosethat
modelthatbestexplainsthevariationineachparametervalue)for wefoundinhorses.
agivenparameteristhemodelwiththelowestAICCandisindi- Ourresultssuggestthatvocalexpressionofemotionsinhorsesis
cated in bold. DAIC gives the difference in AIC between each in accordance with the hypothesis of segregation of information
C C
modelandthebestmodel.Akaike’sweight(vi)assessestherela- aboutemotionalarousalandvalenceindifferentvocalparameters.
tivesupportthatagivenmodelhasfromthedata,comparedto Indeed,F0start,Q50%,Q75%werefoundtobegoodcuestoemo-
theothercandidatemodel.SeeTable2forabbreviationsofthe tional arousal, while Dur, G0start and G0mean were found to be
parameters. goodcuestovalence.APCAincludingall19measuredvocalpara-
metersrevealedasegregationofcuestoemotionalarousal(F0start,
Parameter Arousal/Valence AICC DAICC vi Q50%,Q75%)andvalence(Dur,G0startandG0mean)indifferent
PCs,suggestingtheirindependence.Emotionalarousalandvalence
Dur A 420.35 15.23 0.00
V 405.13 0.00 1.00 are therefore encoded in different parameters, which seem to be
G0start A 3696.94 7.19 0.03 relativelyindependentofeachother.
V 3689.75 0.00 0.97 Wefoundthatputativeformantfrequencieswereaffectedbybody
G0max A 3444.95 2.91 0.19 weight(decreaseinfrequencieswithweightincrease;FigureS2and
V 3442.04 0.00 0.81 TableS3).Asimilarnegativecorrelationbetweenbodyweight/size
G0mean A 3378.43 7.41 0.02 andformantfrequencieshasbeenfoundinmostmammalsstudiedto
V 3371.02 0.00 0.98
date,becauseofthestrongdependencyofvocaltractlengthonbody
F0max A 3159.55 0.00 0.78
size(e.g.goat54;koala,Phascolarctoscinereus55).However,apartfrom
V 3162.07 2.51 0.22
themeanfrequencyofthefourthputativeformant(F4mean),which
AmpVar A 1950.92 0.00 0.62
V 1951.92 1.00 0.38 tended to change with emotional arousal, these frequencies mea-
AMextent A 204.78 1.98 0.27 sured in horse whinnies were neither affected by arousal nor by
V 202.80 0.00 0.73 valence.Thiscouldbeduetothefactthatwewereabletomeasure
Q25% A 3831.99 0.00 0.88 putativeformantsonlywhentheendpartofwhinnieswaspresent
V 3835.96 3.97 0.12 (53%ofthewhinniesofeachhorseonaverage;seeSupplementary
Q50% A 3819.71 0.00 0.98 Methods).Therefore,oursamplesizewassmallerforputativeform-
V 3827.23 7.52 0.02
antfrequenciesthanforothervocalparametersandmightnothave
Q75% A 2145.33 0.00 1.00
beensufficienttodetectalinkwithemotions.
V 2131.99 13.34 0.00
Conclusions
We discovered that horse whinnies were composed of two fun-
increaseinsub-glottalpressureand/orvocalfoldtensionproducing damentalfrequencies(F0andG0),suggestingbiphonation.F0and
anincreaseinF0,and2)anincreaseinpharyngealconstrictionora the energy spectrum indicated emotional arousal, while G0 and
lesspronouncedretractionofthelarynx,resultinginahigherenergy whinny duration indicated emotional valence. The function of
distribution,withanincreaseinarousallevels2,35,47.AraiseinF0and non-linearphenomena(i.e.biphonation,subharmonicsanddeter-
a shift in energy distribution towards higher frequencies with ministic chaos36) is not clear, but these particularities could allow
increasing arousal have beencommonly observed in humans2,3, as individualstogeneratehighlycomplexandunpredictablevocaliza-
wellasinothermammals(e.g.pig,Susscrofa48;treeshrew,Tupaia tions40.Biphonationhasbeensuggestedtoenhance,amongothers,
belangeri49; squirrel monkey, Saimiri sciureus13; reviews6,9). In par- identity cues56,57. Our results show that the presence of two fun-
ticular,amongallthestudiesreviewedinBriefer6,andinwhichF0 damental frequencies can also function as a means of emotion
wasmeasured(21studies),F0consistentlyincreasedwitharousal.F0 expression,witheachfrequencyencodingoneemotionaldimension
andtheenergydistributionseemtobereliableandconsistentcuesto (i.e.arousalandvalence).Asemotionalarousalandvalenceareeach
arousalacrossmammals,andeveninbirds(zebrafinch,Taeniopygia encoded by vocal parameters that seem relatively independent of
guttata50). eachother,vocalexpressionofemotionsinhorsescorrespondsmore
Whinnies produced during positive situations were shorter in to the segregation of information hypothesis16 than the trade-off
duration(Dur)andhadalowerG0(G0startandG0max)thanthose hypothesis19.Thissuggeststhatemotionalarousalandvalencecan
produced during negative situations (Figure 3; Audio S2). The bothbeeffectivelyandsimultaneouslycommunicatedinthisspecies.
changeindurationcanbeexplainedbyshorterexpirations,resulting Vocal communication of emotions could allow horses to regulate
inshorterwhinnydurationduringpositivesituations,comparedto social interactions within groups. Further playback experiments
negativeones.However,inordertoexplainvalence-relatedchanges could test if conspecifics perceive these emotional-related changes
to G0, we would need to find the source of production of this totheacousticstructureofwhinnies24,58.Ourapproachallowedusto
identify clearly which parameters were mostly influenced by each
parameter,whichrequiresfurtherexamination.Adecreaseindura-
emotional dimension. We believe that this approach will lead to
tion between negative and positive situations, as revealed in our
study, was also found in dogs51, and squirrel monkey13. These betterknowledgeofvocalcorrelatesofemotionsinanimals,which
could help to understand the phylogenetic continuity of emotion
results are in accordance with the motivation-structural rules52,
expression between animals and humans, through cross-species
which states that calls produced during appeasing contexts are
comparisons29.
generallyofshorterdurationsthanthoseproducedduringaggress-
ivecontexts53.Alowerpitch(G0inourcase)inpositivecompared
to negative situations was highlighted in squirrel monkey13, and Methods
African elephant (Loxodonta africana15).This suggeststhatvocal Subjectsandmanagementconditions.TwentyhorsesweretestedinMayandJune
2013(Table1).Allthehorseshadbeenintheirrespectivefarmforatleast6months.
correlates of emotional valence could have been, similarly as
Foreachtestedhorse,weidentifiedthegroupmemberthat,accordingtothefarm
those of emotional arousal, conserved throughout evolution. owner,elicitedthehighestnumberofvocalizationsduringseparation
SCIENTIFICREPORTS |4:9989|DOI:10.1038/srep09989 8