Semantic, Typicality and Odor Representation

13 Pages·2005·0.21 MB·English

Chem.Senses 30:37–49,2005 doi:10.1093/chemse/bjh255
Semantic, Typicality and Odor Representation: A Cross-cultural Study
C. Chrea1, D. Valentin1,2, C. Sulmont-Rosse´3, D. Hoang Nguyen4 and H. Abdi5
1Centre des Sciences du Gouˆt, UMR 51–70, Dijon, France, 2University of Bourgogne, Dijon,
France,3Unite´ MixtedeRecherchesurlesAroˆmes,INRA-ENESAD,Dijon,France,4Universityof
Technology ofHo Chi Minh,HCM City, Vietnam and5TheUniversity ofTexas atDallas,Dallas,
TX, USA
Correspondencetobesentto: Christelle Chrea,Centre desSciencesdu Goˆut, UMR51–70, 15rueHuguesPicardet, 21000Dijon,
France.E-mail: [email protected]
Abstract
This study investigated odor-category organization in three cultures by evaluating (i) the relationship between linguistic and
perceptualcategorizationand(ii)theexistenceofaninternalstructureofodorcategories.Inthefirstexperiment,threegroups
of30participantsfromAmerican,FrenchandVietnameseculturesperformedasortingtask.Thefirstgroupsorted40odorants
onthebasisofodorsimilarity,thesecondgroupsorted40odornamesonthebasisofnamesimilarityandthelastgroupsorted
40odornamesonthebasisofimaginedodorsimilarity.Resultsshowedthatodorcategorizationwasbasedonperceptualor
conceptualsimilarityandwasinpartindependentofwordandimaginedcategorizations.Inthesecondexperiment,another
groupof30participantsfromeachcultureratedthetypicalityoftheodorantsfor11odorcategories.Resultsshowedthatsome
odorantswereratedasmoretypicalthanothers.Moreover,thetypicalitygradientpredictedtheodorspaceobtainedintheodor
sortingtaskinaconsensualwayamongthethreecultures.Theseresultssuggestthat,asforothercategories,odorcategoriesare
basedonperceptualsimilaritiesratherthanonsemanticcues.Moreoverodor-categorystructuremighthaveacorerepresen-
tation which might becommon todifferentcultureswith boundaries which might bemoreculturallydependent.
Keywords: categorization, culture, linguistic, olfaction, typicality gradient
Introduction
Perceptual categorization has generated considerable inter- review, see Chastrette, 2002). But a linguistic study carried
est in cognition and perception; most of it focused on the outbyDupire(1987)showedthattheN’Dut(anAfricaneth-
visual modality (for a recent review, see Murphy, 2002). nicgroupfromSenegal)useafewbasictermsinordertoclas-
In this visual framework, color categories were considered sify the overall odor environment in only five categories
an ideal domain in which (cid:1)language-cognition(cid:2) research (i.e.scented,milk-fishy,rotten,urine-like,acidic).Inanother
coulddemonstratetheeffectoflinguisticcategoriesonnon- studycarriedoutintwocultures(GermanandJapanese)with-
verbalcognitiveprocesses(BrownandLenneberg,1954;Berlin outanyabstracttermstodescribeodors,Schleidtetal.(1988)
and Kay, 1969; Rosch, 1973). The main results of this re- foundthatodormemoriescansimilarlybeclassifiedintoafew
search were that (i) colors tend to be universally clustered categoriesacrossthesetwocultures(i.e.civilization,foodand
andafewbasictermsexpressuniversalfeaturesofcolorper- drink, nature, man, remainder). These categories are some-
ception; and (ii) colors are not all equivalent in a category: what similar to the categories previously found by Dupire
some colors are more representative of the color category, (1987)inthatfecalodors,humanodorsandmaterialinde-
and are said to be more typical of this category than other compositionareclassifiedasunpleasant,whereasodorsfrom
colorsare.Herewewanttoknowwhethertheapproachused vegetationareclassifiedaspleasant.Thesetwostudiessuggest
in the study of color categorization could be extended to thatodorsmightbestructuredincategoriesratherthanper-
odorcategorization.Odorclassificationhasbeeninvestigated ceived as(cid:1)unitary perceptual events(cid:2), asproposed by Engen
sincethenineteenthcenturybyresearchersfromvariousfields and Ross (1973). However, these two studies involve the
such as botanic, biology, chemistry and perfumery. But in classificationof(cid:1)odorantobjects(cid:2)andthusareconcernedwith
contrasttothevisualmodality,nouniversalprimarysensa- linguisticcategorizationofodors,ratherthanodorcategor-
tionhasyetemergedfromanyofthesestudies(forarecent izationperse.
ChemicalSensesvol.30no.1ªOxfordUniversityPress2005;allrightsreserved.
38 C.Chreaetal.
In fact, only a few studies have been carried out to inves- groupofNewGuinea),wherelanguagelacksbasictermsto
tigatetheexistenceandconsistencyofodorcategoriesperse. describecolors.PreviousworkbyDubois(2000)suggeststhat
Ueno (1993) found that Japanese and Sherpa participants odor categories might be structured according to internal
agreed on how to sort 20 artificial Japanese aromas on propertiessuchastypicality.Toexaminefurthertheexistence
the basis of their perceptual similarity, with the exception of such a typicality gradient in odor categories, we first ex-
that the Japanese classification revealed a (cid:1)fishy(cid:2) category tractedelevencategorynamesfromtheodorsortingdatacol-
thatdidnotemergeintheSherpaclassification.Theauthors lectedinChreaetal.study.Wethenaskedanothergroupof
hypothesizedthatthisdifferencemightbeduetothefactthat participants from each culture to rate the typicality of the
fishodorsarerarelyencounteredbySherpapeople.(Sherpa odorantsfortheseelevenodorcategoriesona7-pointscale.
isanethnicgroupofNepal).Morerecentlyinthesamevein, Wefinallyevaluatedwhetherwecanpredicttheodorspace
Chrea et al. (2004) asked participants of American, French resultingfromthesortingtaskbythetypicalityratings.
andVietnameseculturestosortfreely40everydayodorants.
Multidimensionalscalinganalysesofthesedatashowedthat Experiment 1: Relationship between linguistic
four common clusters (i.e. sweet, floral, bad and nature) and odor categorization
emerged for the three groups of participants. In addition,
Chreaetal.founddifferencesatafinerlevel,whichmayhave Materials andmethods
beenduetodifferencesinfoodhabits.Forinstance,winter-
green—whichisusedascandyorsodaflavorintheUSA— Participants
was put in the (cid:1)sweet(cid:2) cluster by most of the American Three groups of 30 participants from each culture were
participants while it was put in the (cid:1)medicine(cid:2) cluster by recruited from the University of Texas at Dallas (USA),
Frenchparticipantsandinthe(cid:1)floral(cid:2)clusterbyVietnamese theUniversityofBourgogneatDijon(France)andthePoly-
participants. Those results are in agreement with Ueno’s technicInstituteofDanang(Vietnam).Groupswerecompar-
results and suggest that odor perceptual categorization able in gender and age distribution, both across tasks and
dependsinpartonfamiliarityorfrequencyofexposurewith across cultures. The participants were born and raised in the
specific odors. countryoftheexperiment.Noneoftheparticipantswerein-
BothUeno(1993)andChreaetal.(2004)supportthehy- formed of the real aim of the experiment. In the word con-
pothesisderived from linguisticstudies that odorsmight be dition, to ensure that participants performed the task on
organizedinconsensualcategoriesacrossdifferentcultures. wordratherthanodorsimilarity,participantsweretoldthat
These studies, however, did not explore—contrary to what the experiment aimed at studying language representation.
wasdoneforcolors—thesimilaritybetweenlinguisticandper- In the two other conditions, the participants were told that
ceptualcategories,nordidtheyexploretheorganizationofthe the experiment aimed to investigate odor perception.
perceptualcategoriesthemselves.Theaimofthepresentstudy
istoaddressthisissuebyexpandingChreaetal.’swork. Stimuli
In the first experiment, we evaluated the relationship Odorants.Weused40odorantsselectedfromaninitialsetof
between linguistic categorization as reported by Dupire 56odorantsamplesprovidedbySentosphe`re(Paris,France).
(1987) and Schleidt et al. (1988), and odor categorization To select these odorants, we used familiarity rating scores
per se as reported by Ueno (1993) and Chrea et al. (2004). collected in a previous study (Ly Mai, 2001). The odorants
To achieve this goal, we compared the perceptual categor- wereselectedsothat17ofthemwereequallyfamiliar inall
izationofodorantstimuli(odorconditionreportedinChrea threecultures,sixwereratedasmorefamiliarinFrance,six
et al.) with two linguistic categorizations of odor names. In wereratedasmorefamiliarintheUSAand11wereratedas
thefirstlinguisticcategorization,weaskedparticipantsfrom more familiar inVietnam.The evaluation ofthe rating was
American,FrenchandVietnameseculturestosortthenames obtainedfromatwo-wayanalysisofvariancewithcultureas
of the odorants from the Chrea et al. study on the basis of the between-participant independent variable and odor as
theirsimilaritywithouttellingtheparticipantsthatwordsre- the within-participant independent variable. The odorants
ferredtoodornames(wordcondition).Inthesecondlinguistic weremicroencapsulatedandpresentedin2cmhighpunched
categorization, we asked another group of participants— plastic flasks randomly coded by a two-digit code number.
from the same three cultures—to imagine the odors from Participants were asked to shake the flask before opening
theodornamesandtosortthemonthebasisofthesimilarity it and to bring it up close until the opened flask was about
oftheirimaginedodors(imaginedcondition).Inthesecond 1centimeterundertheirnose.Participantswereinstructedto
experiment, we were interested in evaluating whether we smell the odorants by breathing normally, without sniffing.
can find an odor-category internal structure similar to the They could manipulate the stimuli freely with no time limi-
one described by Rosch (1973) for color categories. Rosch tation.Inordertoreduceolfactoryadaptation,participants
showedthatcolorcategoriesarestructuredalongatypicality wereaskedtowaitfor15sbetweentwoodorants.Although
gradientinthatsomeexemplarsarebetterandmorerepresen- a longer interval delay is generally recommended between
tativethanothers,eveninculturessuchastheDani(anethnic two trials to reduce the cross-adaptation phenomenon
Semantic,TypicalityandOdorRepresentation 39
(Ko¨ster, 1971), we had to deal with the natural speed at ing a non-parametric alternating least-squares scaling algo-
which the participants wanted to perform the task. We ob- rithm(ALSCAL;seee.g.BorgandGroenen,1997).Forall
served, however,thatparticipantsspontaneouslytookmore threeculturesandallthreesortingconditions,three-dimensions
time between two samples when they felt it necessary. If a wereselectedasthemostappropriateMDSsolution(thestress
participant perceived no odor when smelling an odorant, valuesoftheMDSsolutionswererespectively0.16intheodor
heorshedidnotperformthetaskforthatodorant. condition,0.13inthewordconditionand0.17intheimagined
Wordandimaginedconditions.Foreachculture,odorlabels condition for France; 0.16, 0.14 and 0.17 for the USA; and
wereobtainedfromamultiple-choiceidentificationtaskper- 0.18,0.16and0.18forVietnam).Thisanalysiswascompleted
formedbytheodorsortinggroup.Theaimofthistaskwasto by a hierarchical cluster analysis (HCA) performed on the
selectthemostconsensuallabelattributedtoeachodorantin resultsoftheMDSanalysis.Herewepresentonlytheresults
ordertoobtainlabelswhichmadesenseforagivenculture. oftheHCAs.
When the participants had completed the sorting task (see Toselectthenumberofclustersyieldedby theHCAs, we
below),theywereaskedtore-smelleachodorantandtofind examinedthedendrogramsforlargechangesinlevel.Specif-
the nameof its odor among a list of 90 labels: the 40labels ically we used an approach similar to the scree test (the
given by Sentosphe`re to the odorants and 50 additional (cid:1)elbow-test(cid:2))usedinprincipalcomponentanalysis.Foreach
labels frequently given in a previous free identification task leveloftheclusteranalysis,wecomputed(usingSASPROC
of the 40 odorants (Ly Mai, 2001). For each odorant, the CLUSTER) the root-mean-square standard deviation cor-
label associated with the highest frequency of citation was respondingto this level. We selectedthenumberofclusters
selected.Whenthisfrequencywas£10%,thelabelprovided correspondingtothelargestdifferencebetweentwoconsecu-
by Sentosphe`re was selected. This occurred approximately tive levels (Milligan and Cooper, 1985). We found, for
equally in the three cultures (6, 10, and 11 respectively for France, five clusters in all three sorting conditions. For
France, USA and Vietnam) and was expected because in theUSAwefoundfourclustersintheodorandimaginedcon-
each culture, some odorants were unfamiliar. The 40 labels ditionsandsixinthewordcondition.Finally,forVietnamwe
selectedineachculturearepresentedinTable1.Eachlabel foundfourclustersintheodorcondition,andfiveclustersin
was written down on a card in capital letters in the native thewordandimaginedconditions.Figure1showsasimplified
language of each culture. representation of the clusters yielded by the HCAs in the
three conditions for the three cultures. In this figure, each
Procedure cluster is labeled by a few generic terms, which represent
Participantssatatatabletoperformthetask.Stimuli(odorsor the most frequent descriptors given by the participants.
printed label cards) were presented to the participants
arranged on the table in random order. Participants were
askedtosmell/readthestimuliandtosortthemonthebasis Global comparison of cluster memberships across
oftheirsimilarity.Intheimaginedcondition,participantswere the threeconditions
instructedtoimaginetheodorsofthe40odorlabelsandsort We first looked at the simplified dendrograms and focused
thewordsonthebasisoftheimaginedodorsimilarity.Inall ontheclustermembershipstocomparethethreesortingcon-
conditions, participants could sort the stimuli into as many ditions within a culture. Three similar findings in all three
groupsastheywished,andeachgroupcouldcontainasmany culturesemergedfromFigure1.Thesefindingsaredescribed
stimuli as the participants wished. After completion of the below.
sortingtask,participantswereaskedtoprovideafewwords Similarityinthemacrostructureanddifferencesinthemicro-
todescribeeachofthegroupstheyhadformed. structure.Inallthreesortingconditions,threecommonclus-
ters emerge. A first cluster includes mostly fruit items,
Experimentalconditions asecondonemostlyfloweritems,andalastonemostlyani-
In the odor condition, the task was conducted in a well- mal and musty items. Besides those three common clusters,
ventilatedroomunderredlightinordertomaskevidentdif- someitemsaregroupeddifferentlyinthethreesortingcon-
ferencesinthecoloroftheplasticflasks.Inthewordandthe ditions(e.g.clove,gingerandnutmegforFrance,catpeeand
imagined conditions, the task was conducted in a standard civet for the USA, mushroom and moldy for Vietnam).
classroom in daylight. Specificclusterscommontowordandimaginedconditions.A
(cid:1)food(cid:2)or(cid:1)fatty(cid:2)cluster—includingmainlynuts,milkandbut-
teritems—appearsinthewordandimaginedconditions,but
Results
notintheodorcondition.Onthesameline,a(cid:1)cleaners(cid:2)clus-
We started by deriving pairwise similarity estimates by ter—including mainly chemical and cosmetic products—
counting the number of times two items were sorted into appears for the French and American groups in the word
thesamegroupoveralltheparticipantsineachcultureand andimaginedconditions,whilecleaningproductsaresorted
each sorting condition. These co-occurrence matrices were withthefloweritemsintheodorconditionforbothofthese
submittedtoamultidimensionalscalinganalysis(MDS)us- cultures.
40 C.Chreaetal.
Table1 OdorantsprovidedbySentosphe`reandlabelsforthewordandimaginedsortingtaskconditions
Odor Abbreviation Frenchlabel Americanlabel Vietnameselabel
Amber AMB amber babypowder babypowder
Anise ANI anise licorice anise
Apricot APR apricot apricot apricot
Blackcurrant BLA blackcurrant raspberry lemon
Butter BUT cheese butter butter
Catpee CAT catpee catpee catpee
Cinnamon CIN cinnamon cinnamon cinnamon
Civet CIV feces feces feces
Clove CLO clove clove toothmedicine
Cookies COO caramel icing cookies
Detergent DET detergent detergent detergent
Eucalyptus EUC eucalyptus nosemedicine eucalyptus
Ginger GIN ginger ginger ginger
Hazelnut HAZ hazelnut hazelnut hazelnut
Honey HON honey honey honey
Jasmine JAS jasmine honeysuckle jasmine
Lavender LAV lavender lavender lemongrass
Leather LEA leather leather leather
Mango MAN mango mango mango
Melon MEL melon watermelon jack-fruit
Milk MIL butter popcorn milk
Mothball MOT mothball bathroomcleaner insecticide
Moldy MOL moldy earth moldy
Mushroom MUSH mushroom mushroom mushroom
Musk MUS musk musk musk
Nutmeg NUT nutmeg nutmeg plastic
Orangeblossom ORA orangeblossom orangeblossom flowerofgrapefruit
Peanut PEA peanut peanut peanut
Pine PINE pine pine pine
Pineapple PIN pineapple pineapple pineapple
Rose ROS rose rose rose
Soap SOA soap soap soap
Strawberry STR strawberry strawberry strawberry
Tea TEA tea beefjerky tea
Truffle TRU truffle cheese fishsauce
Vanilla VAN vanilla vanilla vanilla
Violet VIO violet violet violet
Walnut WAL walnut pecan traditionalmedicine
Wintergreen WIN nosemedicine wintergreen mint
Woody WOO earth cedar woody
LabelsinboldarethoseprovidedbySentosphe`rewhenthepercentageidentificationwas<10%.
Semantic,TypicalityandOdorRepresentation 41
a A stronger hierarchy in the word space. A higher heterogen-
eityinclustersizeinthewordconditioncomparedtothetwo
othersortingconditionsappears.Someclustersincludeonly
twoorthreeitems,whereasothersincludeupto12–20items
andshowsomeobvioussub-divisions.Forinstance,asseen
ANI
AMB CIN CLO inFigure1,theFrench(cid:1)floral–spicy(cid:2)clusterissubdividedin
ROS NUT (cid:1)floral(cid:2)and(cid:1)spicy(cid:2)sub-clustersatafinerlevelofthedendro-
APR BUT
LAV EUC WAL
BLA CAT gram. The Vietnamese (cid:1)vegetation–fruity(cid:2) cluster is subdi-
SOA WIN
STR CIV vided in (cid:1)floral(cid:2) and (cid:1)fruity(cid:2) sub-clusters.
MOT MAN
MEL TRU GIN
ORA Thesefirstobservationsindicatethatodorsarenotcategor-
DET MEDICINE VPAINN MHOONL PLIENAE izedinthesamewayasthenamesassociatedwiththeodors.
MUS
COO MUSH TEA Inaddition,imaginedcategorizationseemstobeclosertothe
JAS
VIO MIL HAZ WOO word categorization than to odor categorization.
SWEET PEA
FLORAL NATURE
BAD Distances betweenthe clusterpartitionsamong thethree
Odor condition
sorting conditions
Toexaminemorepreciselythelevelofsimilaritybetweenthe
threesortingconditionsandthethreecultures,wecomputed
distancesbetweentheclusterpartitionsyieldedbytheHCAs.
The distance we usediscalled thesymmetricdifference dis-
CAT tance (sometimes called the (cid:1)Hamming distance(cid:2)). The sym-
AMB BUT
APR CIV
MUS MIL metricdifference distance is adistance defined between sets
MEL MOL
PINE MUSH DET of objects; it corresponds to the number of elements which
MAN
EUC TRU SOA
PIN belongtoonlyoneset(seee.g.Carre´,1979,p.7).Inthepres-
JAS LEA MOT
LAV BLA FOOD WOO WIN entwork,thedistancebetweentwoclustersisthenumberof
STR
ROS CLEANERS odorswhicharepresentinoneclusterandnotintheother.
VIO FECES Smallervaluesofthedistanceindicatethattwoclustersshare
MATERIAL
COO
a large number of items, while a large value indicates that
ANI HON
ORA HAZ two clusters include different items.
CIN PEA Table2reportsthesedistances.Afirstpointtonoteisthat,
GIN WAL
in France and in the USA, the cluster partition in the odor
NUT
CLO FRUIT condition is equally distant from the partitions in the word
VAN and the imagined conditions. In contrast, for Vietnam, the
TEA cluster partition in the odor condition is much closer to
FORAL thepartitionintheimaginedconditionthantothepartition
SPICY in the word condition. A second point is that distances are
Word condition smaller between cluster partitions among the three cultural
groups within the odor condition than between cluster par-
titionsinthewordandodorconditionswithinaculture.This
resultsuggeststhattheconsensusbetweenthethreecultural
groupsincategorizingodorsisquiterobustbecausethiscon-
AMB CAT sensusisstrongerthantheconsensuswithinacultureincat-
APR
MUS MEL BUT MOT egorizing odors and odor names.
GIN ANI MIL CIV
PIN
NUT LAV MOL
BLA Prediction of perceptualcategorization bylinguistic
CLO DET
STR HAZ
WIN MAN PEA LEA categorization
CIN EUC WAL MUSH Toevaluatewhetherwecanpredictthespaceobtainedinone
TEA PINE TRU
sortingconditionfromthespaceobtainedinanothercondi-
COO JAS FATTY WOO
HON CLEANERS ORA FOOD tion, we performed a series of linear regressions on the
VAN ROS BAD
VIO
SPICY SOA
SWEET
FRUIT Figure1 Compositionoftheclustersforallthreesortingconditionsin(a)
FLORAL France,(b)USAand(c)Vietnam,respectively.Clusterswereselectedonthe
Imagined condition basisofthefirstlargechangeinlevels.Whenasecondlargechangeoccurred
inthedendrogram,sub-clustersaremarkedwithadashedline.
42 C.Chreaetal.
b c
AMB CLO
SOA BUT NUT ANI AMB JAS APR
ROS TRU TEA MIL DET CAT PINE BLA
MODREOATT WCPCEAAIVATL PLEIAUNVCE BAPLPINAR OLCARLOVA HTCRAIVUZ MSTOEOAAT MCSOTEROL
LEA HON GIN MEL ANI LEA MUS MIL
MAN HAZ JAS STR WIN PEA NUT PIN
MUS MOL COO CIN MOL MUSH ROS
VIO BAD WOO VAN EUC WAL WOO BUT
FLORAL MUSH CIN GIN BAD VIO HON
NATURE WIN FLORAL NATURE MAN
VAN
SWEET SWEET
Odor condition Odor condition
CIN
ANI APR
GIN AMB
APR CIN MAN
AMESDOMUEOACTBT BOCCDAIVTILY CBTMAOEUNIALOTI MMBSPTLAIENRALN HVNCLEOALUAONNT OLRVJAAROIOSVAS MVSCDOALEOAONTT HCTBMROOUIULONT CCAIVT MORVJAROUIOSASS MHBSPTLAIENRAZL MICMRMOUOS-LOHRGANISMS
FUNCTIONS TRU MOL FLOWER WAL TEA FECES
CLEANERS MUSH HAZ MUS COLOR PEA EUC GIN
WAL LEA
SNACKS PINE PINE LAV
PEA WOO NUT FATTY WOO WIN
WIN FOOD
FRUIT CHEMICAL
NUT VEGETATION
SPICE
FRUIT/NUT
NATURE
Word condition Word condition
AMB CIN ANI
DET GIN BUT VAN
MSECCOUAOIVACTT MNCWLUELUISANOTH WHTTMREAAIAULZL MHCASOOTAPRRONN ASVJAOMIOSAB CBOUOT TCLCREAIVAUT WPNEIUUNACTEL
BATHROOM MOL PEA BLA ROS APR MIL CLO HAZ
MUS PIN ORA BLA HON DET TEA
PINE FOOD MEL ANI MAN VAN MOT WOO
WOO JAS CIN PIN MOL
LAV STR FATTY BAD MUSH
SPICY MUS MEL CHEMICAL PEA
NATURE ORA GIN
LAV WIN FRUIT PLANT
VIO MUSTY
ROS FLORAL
ESSENTIAL OILS
SWEET
Imagined condition Imagined condition
Figure1 Continued.
EuclidiandistancematricesresultingfromtheMDSanalyses determinationcoefficientsshowthattheodorspaceisbetter
inthethreesortingconditions.Table3reportsthecoefficients predicted by the imagined space than by the word space in
ofdeterminationandtheirsignificance.Theregressionanaly- all three cultures, but the difference of variance explained
ses are all highly significant for all three cultures. But it is bythetwopredictorsissmall.Besides,thewordspacepredicts
worthnotingthatthishighsignificancemaybeduetothelarge theimaginedspacemuchbetterthanitpredictstheodorspace.
numberofdegreesoffreedomoftheregressionanalysis.The These findings suggest that asking participants to imagine
Semantic,TypicalityandOdorRepresentation 43
Table2 Valuesofthesymmetricdifferencedistance(1)betweenthe Nevertheless, our results showed a contradictory example
clusterpartitionsinthethreeexperimentalconditionswithinacultureand withnutodors,sortedinthe(cid:1)bad(cid:2)clusterbyalmostallpar-
(2)betweentheclusterpartitionsinthethreecultureswithintheodor
ticipants in the three cultural groups in the odor condition.
condition
Indeed, the resultsshowed that even participantssuccessful
(1)Culture Odorversus Odorversus Wordversus
inidentifyingnutodorsinthemultiple-choiceidentification
word imagined imagined
task sorted them in the (cid:1)bad(cid:2) cluster. In contrast, nut items
France 193 208 199 were mostly sorted with fruit items in the word condition.
TheUSA 238 235 267 Finally,wefoundthattheodorcategorizationwasequally
distant from the word and the imagined categorization for
Vietnam 228 191 165
FranceandtheUSA,whileforVietnam,odorcategorization
(2)Odor Franceversus France TheUSA wasclosertoimaginedcategorizationthantowordcategor-
condition theUSA versusVietnam versusVietnam
ization.TheresultsfortheVietnamesegroupareconsistent
136 183 211 withastudyofSugiyamaetal.(2003),whoaskedJapanese
participants to perform a pair similarity judgement task in
eitheranodor,linguistic,orimaginarycondition.Usingpro-
crustean distances to compare the three MDS solutions,
theseauthorsalsofoundagreatersimilaritybetweenpercep-
tualandimaginaryspacesthanbetweenperceptualandlin-
Table3 Determinationcoefficients(R2)andtheirstatisticalsignificance guisticspaces.Aplausibleexplanationfortheseresultsmay
betweenthethreeexperimentalconditionsforallthreecultures(n=760) be that in both studies odorants were not very familiar to
participants and thus difficult to identify. The distance be-
France USA Vietnam
tween odor categorization and odor names categorization
Odorversus 0.12,P<0.0001 0.14,P<0.0001 0.06,P<0.0001 mightthereforebeduetotheinadequacybetweentheodor-
word
ant source names and the participants’ mental representa-
Odorversus 0.19,P<0.0001 0.17,P<0.0001 0.23,P<0.0001 tions of the odors. However, in contrast with our study,
imagined
the odorants used in the Sugiyama et al. study were manu-
Imaginedversus 0.33,P<0.0001 0.53,P<0.0001 0.22,P<0.0001 facturedinthecountryoftheparticipants.Also,becauseno
word indication was reported on the participants’ability to iden-
tifytheodors,itisdifficulttoevaluatetheactualfamiliarity
of Sugiyama et al.’s participants with the odors.
odorstendstoincreasejustalittlethepredictionofapercep-
tual categorization by a linguistic categorization of odor
names. Finally, there is a better prediction of a linguistic Experiment 2: Internal structure of odor
categorization by another linguistic categorization than a categories
perceptualcategorizationbyoneofthetwolinguisticcategor-
In the previous experiment we showed that the perceptual
izations.
organizationofodorsdidnotmatchthelinguisticorganiza-
tion of the names associated with odors. In this experiment
weexplorefurtherthemechanismsunderlyingperceptualor-
Discussion
ganizationbyinvestigatingtheexistenceofaninternalstruc-
TheresultsofExperiment1suggestthatperceptualcategor- ture in odor categories such as a typicality gradient.
ization is in part independent of linguistic categorization. Moreoverwewanttoevaluatewhetherthisperceptualorgan-
Whereaswordstendtobeclassifiedonthebasisofbiological ization is culture dependent or consensual among the three
taxonomy,odorstendtobecategorizedonthebasisoftheir cultural groups.
perceptual similarity. However, we cannot completely rule
out the hypothesis that odor categories may be affected
bysemanticassociations.Indeed,fruitandflowerodorants, Materials andmethods
identifiedasfruityorfloralbymostoftheparticipantsinthe
three cultural groups, were grouped in the same way in the Participants
wordandimaginedconditions.Wemightsupposethatiden- Agroupof30participantswasrecruitedfromtheUniversityof
tificationoftheodorsduringthesortingtaskhelpedpartici- TexasatDallas(USA),theUniversityofBourgogneatDijon
pantstofindsomelexicalcriteriatoformtheirgroups.This (France)andthePolytechnicInstituteofDanang(Vietnam).
result is in concordance with the suggestion of Chastrette Groupswerecomparableingenderandagedistributionacross
et al. (1988) that for odors such as (cid:1)fruity(cid:2) or (cid:1)floral(cid:2), odor the cultures. The participants were born and raised in the
classification is driven more by the semantic classification countryoftheexperiment.Allwerenaivetothepurposeof
of the odorant sources than by the perceptual similarities. theexperimentandwerenotfamiliarwitholfactorytesting.
44 C.Chreaetal.
Stimuli sensusinthemembershipofthesortingtaskandthecultural
ThestimuliwerethesameasintheodorconditionofExperi- consensusinthetypicalityratings.Forinstance,melon,pine-
ment1.Odorants were coded by a random three-digitcode apple and strawberry—which are rated as very typical of
number. fruit and candy categories by the three groups of partici-
pants—arealsocommontoallthreeculturesinthe(cid:1)sweet(cid:2)clus-
Procedure ter.Inthesameway,civet,catpeeandmoldy—whicharerated
The participantswerepresented withthe40odorsinaran- as very typical of animal and musty categories by the three
domizedorder.Aftersmellinganodorant,participantswere groupsofparticipants—arecommontoallthreeculturesin
asked to rate the typicality of its odor for 11 categories, the(cid:1)bad(cid:2)cluster.Besidesthisagreement,someculturaldiffer-
namelyfruit,flower,candy,cleaner,animal,musty,bakery, encesarealsoobviousforsomeodorants.Forinstance,win-
cosmeticproduct,spice,medicineandnature.Thesecategor- tergreenwasputinthe(cid:1)sweet(cid:2)clusteronlyintheUSA,and
ieswereselectedonthebasisofthedescriptorsprovidedby wasratedasmoretypicalofthecandycategoryintheUSA
the participants in the odor sorting task. The participants (mean score 4.16) than in France (1.50) or in Vietnam
gave their answers on 7-point scales labeled at each end of (2.70). Along the same lines, mango was put in the (cid:1)sweet(cid:2)
the scale (e.g. (cid:1)How typical is this odor of a fruit odor? cluster only in Vietnam, and was rated as more typical of
Not typical at all/very typical(cid:2)). To ensure that the partici- thefruitcategoryinVietnam(4.20)thaninFrance(2.90)or
pants understood the notion of typicality, they were given intheUSA(2.74).
thefollowinginstructionbeforebeginningthetask:(cid:1)Imagine
thatyouareexplainingtoanextraterrestrialwhata‘‘fruit’’
Prediction ofthe clustermembershipbythetypicality
smellis.Wouldyouchoosethisodortoillustratethisconcept
gradient
of a ‘‘fruit’’ smell?(cid:2)
Thediscriminantanalysisproducessignificantresultsforall
Thepresentationorderofcategorynameswascounterbal-
three cultures [F(44,77) = 5.63, P < 0.0001 for France,
ancedacrossparticipants,but,tofacilitatetheparticipant’s
F(33,77) = 9.10, P < 0.0001 for the USA, F(33,77) = 3.28,
task, the order was the same for all odors for an individual
P<0.0001forVietnam].Threesignificantdiscriminantfunc-
participant. The participants answered on an Apple McIn-
tions for France and the USA and two discriminant func-
tosh computer running the PsyScope data acquisition soft-
tions for Vietnam maximize the discrimination of the 40
ware (Cohen et al., 1993). For each odorant, category
odorants. Together, these linear discriminant functions ac-
namesappearedondifferentscreens,thiswasdonetoobtain
count for 97% of the variance for France, 99% for the
answers that were as independent as possible. The experi-
USA and 91% for Vietnam. Thus, in all three cultures typ-
mentalconditionswereidenticaltotheonesintheodorcon-
icality ratings predict the classification of the odors in the
dition in Experiment 1.
clusters resulting from the odor sorting task.
Table4showsthematrixofcorrelationsbetweentypicality
ratingsandsignificantdiscriminantfunctions.Wecanseethat
Results
onlyafewcommontypicalityratingscontributestronglyto
We first averaged the 11 typicality ratings of each odorant
the formation of odor clusters in all three cultures. Indeed,
across the participants in all three cultures. We then exam-
the candy typicality rating has the strongest loading on the
ined the agreement between typicality ratings and cluster
firstfunctionforallthreecultures.Thismeansthatthefirst
memberships. Finally, to evaluate further whether clusters
function discriminates the clusters mainly according to
yieldedbytheHCAswereorganizedaroundatypicalitygra-
acandytypicalitygradientinallthreecultures.Forthesecond
dient, we performed a series of discriminant analyses. This
function,mustyandanimal typicalityratings have astrong
analysisusedthe11typicalityratings topredicttheclassifi-
loading for the USA and Vietnam, while cleaner, candy,
cationofthe40odorsintotheclustersobtainedintheodor
and fruit typicality have the strongest loading for France.
condition of Experiment 1.
Finally,cosmetictypicalityhasastrongloadingonthethird
function for both France and the USA. But, as shown in
Agreement between typicalityjudgement andcluster Figure 2, the discrimination of the clusters is not identical
membership among the three cultures. For instance, the first dimension
Globallythereisaconsensusacrossthethreegroupsofpar- opposesbadandnatureclusterstosweet,medicineandfloral
ticipantstoevaluatesomeodorantsasmoretypicalofagiven clustersinFrance,whereasthisfirstdimensionopposessweet
category than others (cf. Appendix 1). For instance, in all tofloralandnatureclustersinboththeUSAandVietnam.
threecultures,cloveandgarlicareratedasmoretypicalthan
aniseorvanillaforthespicecategory.Likewise,honeysuckle
Discussion
andamberareratedasmoretypicalthanmuskorlavender
for the cosmetic product category. Moreover, for some Wefoundthatsomeodorantswereratedasmoretypicalof
items,thereisastrongagreementbetweentheculturalcon- agivencategorythanothers.Theseresultssuggestthatodors
Semantic,TypicalityandOdorRepresentation 45
Table4 Matrixofcorrelationsbetweentheeleventypicalityratingsaspredictorsandthediscriminantfunctions
Predictors France USA Vietnam
Function1 Function2 Function3 Function1 Function2 Function3 Function1 Function2
Spice (cid:1)0.10 (cid:1)0.17 (cid:1)0.42 0.06 (cid:1)0.06 0.37 (cid:1)0.10 0.38
Candy 0.51 0.44 (cid:1)0.23 0.57 (cid:1)0.59 0.14 0.77 0.36
Musty (cid:1)0.28 0.00 (cid:1)0.06 (cid:1)0.16 0.75 0.07 (cid:1)0.36 (cid:1)0.62
Cleaner 0.16 (cid:1)0.47 0.50 (cid:1)0.32 (cid:1)0.20 (cid:1)0.18 (cid:1)0.25 (cid:1)0.13
Bakery 0.16 0.32 (cid:1)0.03 0.24 0.09 0.05 0.64 0.11
Cosmetic 0.42 (cid:1)0.09 0.42 (cid:1)0.18 (cid:1)0.32 (cid:1)0.78 0.18 0.43
Nature (cid:1)0.11 0.00 0.01 (cid:1)0.16 0.33 0.28 0.05 0.49
Animal (cid:1)0.39 0.13 0.10 (cid:1)0.01 0.58 0.07 (cid:1)0.33 (cid:1)0.53
Flower 0.22 (cid:1)0.16 0.38 (cid:1)0.03 (cid:1)0.30 (cid:1)0.20 0.37 0.64
Medicine 0.16 (cid:1)0.18 (cid:1)0.35 (cid:1)0.12 (cid:1)0.14 0.32 (cid:1)0.27 0.27
Fruit 0.42 0.45 (cid:1)0.08 0.22 (cid:1)0.31 (cid:1)0.07 0.64 0.38
Figure2 Plotsofodorclustersonthediscriminantfunctionsderivedfrom11typicalityratingsfor(a)France,(b)USAand(c)Vietnam,respectively.
46 C.Chreaetal.
within a category are not all equivalent. Moreover, the The results are consistent with previous work suggesting
results of the discriminant analysis showed that odorants that(cid:1)peopledo[perceptually]categorizeodors,butnotwith
werediscriminatedgloballyinallthreeculturesbyfourtyp- semantically cohesive general nouns(cid:2) (Engen, 1987, p. 500).
icality ratings (candy, animal, musty and cosmetic). The Thismaybeduetothefactthatodorsaredifficulttoname
strongcontributionofthesetypicalityratingswasconsistent and people’s verbal responses to them tend to be idiosyn-
withthenatureofthefourcommonclustersemergingfrom cratic (Herz and Engen, 1996). Another possibility is that
theodorsortingtask.Itshowsthattheconsensustocategor- odor categorization is more conceptual than linguistic,
izeodorsiscloselylinkedtotheconsensustojudgethetyp- andthatodorsarecategorizedaccordingtothefunctionat-
icality of odors. However, our results revealed that some tributedtotheodorantobjects.Ourresultsrevealed,forin-
variability in typicality ratings contributed also to cultural stance, that wintergreen, mango and cinnamon, which are
differences in cluster discrimination. This variability may usedindifferentcontextsinthethreecultures,weregrouped
be due to cultural differences in consumption for food and differentlybythethreeculturalgroups(formoredetails,see
cosmeticproductsandalsotoculturaldifferencesinfamiliar- Chrea et al., 2004). This is in agreement with what Dubois
itywithsomespecificodors. (1997)calls(cid:1)pragmaticfactorsofhumanactivities(cid:2)(e.g.hunt-
The question remains, why are some odors perceived as ing,cooking,domesticlife,corporalodors)whichmaybethe
more typical than others? Holley (2001) suggests that some main factors that drive the categorization and naming,
odorsmaybemoretypicalbecausetheyhaveacquiredapar- rather than common perceptual constraints.
ticularperceptualsaliency.Currenttheoriesaboutolfactory Thesecondaimofthisstudywastoinvestigatewhetheran
receptor physiology and the existence of specific anosmias internalstructuremightdriveodorcategoryformation.Our
suggest that, while most odor receptors tend to respond to resultsrevealedthatthetypicalitygradientseemstopredict
a wide range of odors, some might have more narrow sen- the structure of the odor space. Moreover, our results sug-
sitivity spectrum (Holley, 1996). Odors to which these high gest that the core representation of an odor category might
selectivity receptors are preferentially tuned might be more beuniversal,whereastheperipheryofthecategory—includ-
salientandthereforemightbejudgedasmoretypicalofacat- ing the atypical exemplars of the category—might be more
egory. Another possibility is that typicality is linked to the culturallydependent.Thisinterpretation,whichstillneedsto
familiarity with the odors within the culture: more familiar be confirmed with other population samples, would be in
odorsmaybejudgedasmoretypicalthanodorsthatareless agreement with the claim that, as suggested by Rosch
familiar.However,wefoundanexamplethatisnotconsist- (1973)forother naturalcategories,odorcategoriesareuni-
entwiththishypothesis:melon—rarelyencounteredinViet- versallyorganizedaroundsomeprototypesbuthavenorigid
namandnotincludedamongthemostcommonfruitseither boundaries. Previous work by Labov (1973) for object cat-
in the USA or in France—was nevertheless judged as the egoriesandLawlessetal.(1991)forodorcategoriesshowed
most typical fruit odor by all three groups of participants. that category boundaries are often fuzzy and can vary
Although we cannot be sure that there is not another fruit in dependingupon thecontext.Our resultsshow that, inadd-
Vietnamthathasamelon-likeodor(melonwasidentifiedas ition to contextual factors, shifts in category boundaries
jackfruitby19%ofVietnameseparticipantsinthemultiple- mightalsobeexplainedbyculturalfactors,suchasfamiliar-
choiceidentificationtask),thisexamplesuggeststhatfamil- itywithsomespecificodors,orbythefunctionattributedto
iarity is not the only determinant of typicality. the odors.
General discussion
Conclusion
Our firstobjective was to evaluate the relationship between
odorcategorizationperseandthelinguisticcategorizationof Our results provide an insight into the way odor categories
namesassociatedwithodors.Someauthorshavearguedthat areorganized.Accordingtotheuniversal–prototypicalpoint
linguistic devices are cues to access odor representations ofview,odorstendtobeencodedinmuchthesamewayas
(Dubois, 2003). Contrary to this hypothesis, our results colors. However, (cid:1)while the structure of colors in memory
showed that odor categorization does not fit the linguistic comes to resemble the structure of color names in a given
categorizationofodornamesbecauseodorsandodornames language(cid:2)(HeiderandOlivier,1972,p.338),odorsarerather
werecategorizeddifferently.Onlyforsomeodorants,suchas categorizedonthebasisofperceptualandodor-contextsimi-
fruitandflowerodorants,wasthereasimilarityofclustering larity.Anotherdifferencebetweenodorsandcolorsisthatit
observed between odor and word conditions. In addition, iseasytoconjureupamentalimageofacolorjustafterhear-
whenaskedtoimagineodorsfromtheodornames,partici- ing a color name, whereas such an association between an
pantscouldnotfreethemselvesfromsemanticcuesandcat- odornameanditsmentalimageseemsmoredifficulttoac-
egorized imagined odors mostly on the basis of the odor cess. This pattern of results suggests that the cognitive pro-
names. This suggests that participants had some difficulty cessingofodorsisnotcompletelysimilartotheprocessingof
inimaginingodorsandthusmighthavebasedtheircategor- colors and confirms the claim that odor encoding leads to
ization on semantic cues rather than perceptual cues. specific cognitive mechanisms.