43
-22-
Thisproblemarisesformanypartsotherthanatoms,ofcourse. Somecomponents
canserveaspartsofmanydifferentthings:thinkofallthedifferentobjectsthathave
gears. Itissometimesthecasethatanobjectcanbetwokindsofthingatthesame
time—apianoisbothakindofmusicalinstrumentandakindoffurniture,for
example—whichresultsinwhatissometimescalledatangledhierarchy(Fahlman,
1979). Tangledhierarchiesarerarewhenhyponymyisthesemanticrelation. In
meronymichierarchies,ontheotherhand,itiscommon;point,forexample,isa
meronymofarrow,awl,dagger,fishhook,harpoon,icepick,knife,needle,pencil,pin,
sword,tine;handlehasanevengreatervarietyofholonyms. Sincethepointsand
handlesinvolvedaresodifferentfromoneholonymtothenext,itisremarkablethatthis
situationcausesaslittleconfusionasitdoes.
FunctionsandPredication
Theterm‘function’hasservedmanypurposes,bothinpsychologyandlinguistics,
soanyonewhousesitisobligatedtoexplainwhatsensetheyattachtoitinthiscontext.
Afunctionalfeatureofanominalconceptisintendedtobeadescriptionofsomething
thatinstancesoftheconceptnormallydo,orthatisnormallydonewithortothem. This
usagefeelsmorenaturalinsomecasesthaninothers. Forexample,itseemsnaturalto
saythatthefunctionofapencilistowriteorthefunctionofaknifeistocut,buttosay
thatthefunctionofacanaryistoflyortosingseemsabitforced. Whatisreally
intendedhereareallthefeaturesofnominalconceptsthataredescribedbyverbsorverb
phrases. Nominalconceptscanplayvarioussemanticrolesasargumentsoftheverbs
thattheyco-occurwithinasentence:instruments(knife-cut),materials(wool-knit),
products(hole-dig;picture-paint),containers(box-hold),etc.
Theredoesnotseemtobeanobvioustermforthistypeofdistinguishingfeature.
TheyresemblethefunctionalutilitiesoractionpossibilitiesthatGibson(1979)called
‘affordances.’ Gardner(1973),borrowingatermfromJeanPiaget,spokeof
‘operativity’;operativeconceptsareacquiredbyinteractionandmanipulation,whereas
figurativeconceptsareacquiredvisually,withoutinteraction.Lackingabetterterm,
functionwillserve,althoughthepossibilityshouldnotbeoverlookedthatamoreprecise
analysismightdistinguishseveraldifferentkindsoffunctionalfeatures.
Theneedforfunctionalfeaturesismostapparentwhenattemptingtocharacterizea
conceptlike{ornament,decoration}. Anornamentcanbeanysizeorshapeor
composition;partsandattributesfailtocapturethemeaning. Butthefunctionofan
ornamentisclear:itistomakesomethingelseappearmoreattractive. Atleastsince
Dunker(1945)describedfunctionalfixedness,psychologistshavebeenawarethatthe
usestowhichathingisnormallyputareacentralpartofaperson’sconceptionofthat
thing. Tocallsomethingabox,forexample,suggeststhatitshouldfunctionasa
container,whichblocksthethoughtofusingitforanythingelse.
Therearealsolinguisticreasonstoassumethatathing’sfunctionisafeatureofits
meaning. Considertheproblemofdefiningtheadjectivegood. Agoodpencilisonethat
writeseasily,agoodknifeisonethatcutswell,agoodpaintjobisonethatcovers
completely,agoodlightisonethatilluminatesbrightly,andsoon. Astheheadnoun
44
-23-
changes,goodtakesonasequenceofmeanings:writeseasily,cutswell,covers
completely,illuminatesbrightly,etc. Itisunthinkablethatallofthesedifferent
meaningsshouldbelistedinadictionaryentryforgood. Howshouldthisproblembe
handled?
Onesolutionistodefine(onesenseof)goodas‘performswellthefunctionthatits
headnounisintendedtoperform’(Katz,1964). Agoodpencilisonethatperformswell
thefunctionthatpencilsareintendedtoperform;agoodknifeisonethatperformswell
thefunctionthatknivesaresupposedtoperform;andsoon. Thissolutionputsthe
burdenontheheadnoun. Ifanobjecthasanormalfunction,thenoundenotingitmust
containinformationaboutwhatthatfunctionis. Thenwhenthenounismodifiedby
good,thefunctionalfeatureofthenoun’smeaningismarked‘+’;whenitismodifiedby
bad,thefunctionalfeatureismarked‘-’. Ifanobjecthasnonormalfunction,thenitis
inappropriatetosayitisgoodorbad:agoodelectronissemanticallyanomalous. If
somethingservesseveralfunctions,aspeakerwhosaysitisgoodorbadcanbe
misunderstood.
AsurprisingconsequenceofthisformulationisthatanobjectthatisnotanXcanbe
saidtobeagoodXifitperformswellthefunctionthatX’snormallyperform. For
example,callingaboxachairdoesnotmakeitone,yetapersonwhositsonaboxcan
say‘‘Thisboxisagoodchair’’andsoindicatethattheboxisperformingwellthe
functionthatchairsareexpectedtoperform. Suchasentencewouldbeunintelligibleif
thefunctionthatachairnormallyserveswerenotincludedaspartofthemeaningof
chair.
Intermsofthepresentapproachtolexicalsemantics,functionalinformationshould
beincludedbypointerstoverbconcepts,justasattributesareincludedbypointersto
adjectiveconcepts. Inmanycases,however,thereisnosingleverbthatexpressesthe
function. Andincaseswherethereisasingleverb,itcanbecircular. . Forexample,if
thenounhammerisdefinedbyapointertotheverbhammer,bothconceptsareleftin
needofdefinition. Moreappropriately,thenounhammershouldpointtotheverbpound,
becauseitusuallyplaysthesemanticroleofinstrumentandisusedforpounding;the
verbhammerisaconflationofitssuperordinatehitandtheinstrumentusedtodoit. The
semanticroleofnounslikehammer,wallpaper,orboxtendtobethesamewhereverthey
occurinsentences,independentoftheirgrammaticalrole. Thatistosay,inbothJohnhit
themuggerwithahammerandThehammerhithimonthehead,thesemanticroleof
hammeristhatofaninstrument.Similarly,woolisasemanticmaterialineachofthe
followingsentences:Sheknittedthewoolintoascarf,Sheknittedascarfoutofthewool,
andThiswoolknitswell. Thisconsistencyinmappingontothesamesemanticrole
independentlyofsyntaxisnotafeatureofallnominalconcepts,however:whatisthe
functionofappleorcat?
Althoughfunctionalpointersfromnounstoverbshavenotyetbeenimplementedin
WordNet,thehyponymichierarchyitselfreflectsfunctionstrongly. Forexample,aterm
likeweapondemandsafunctionaldefinition,yethyponymsofweapon—gun,sword,
club,etc.—arespecifickindsofthingswithfamiliarstructures(Wierzbicka,1984).
Indeed,manytanglesinthenounhierarchyresultfromthecompetingdemandsof
38
-24-
structureandfunction. Particularlyamongthehumanartifactstherearethingsthathave
beencreatedforapurpose;theyaredefinedbothbystructureanduse,andconsequently
earndoublesuperordinates. Forexample,{ribbon,band}isastripofclothonstructural
grounds,butanadornmentonfunctionalgrounds;{balancewheel}isstructurallya
wheel,butfunctionallyaregulator;{cairn}isapileofstonesthatfunctionsasamarker;
etc. Functionalpointersfromthesenominalconceptstotheverbalconcepts{adorn},
{regulate},{mark},etc.couldeliminatemanyofthesetangles. Atthistimeitisnot
obviouswhichrepresentation(ifnotboth)hasthegreaterpsycholinguisticvalidity.
Thedetailsareobviouslycomplicatedanditishardtofeelthatasatisfactory
understandingofthesefunctionalattributesofnominalconceptshasyetbeenachieved.
IfsupportforthecontinueddevelopmentofWordNetisforthcoming,theexerciseof
addingpointersfromnounstotheverbsthatexpresstheirfunctionsshouldleadtodeeper
insightintotheproblem.
Antonymy
Thestrongestpsycholinguisticindicationthattwowordsareantonymsisthateach
isgivenonawordassociationtestasthemostcommonresponsetotheother. For
example,ifpeopleareaskedforthefirstwordtheythinkof(otherthantheprobeword
itself)whentheyhear‘‘victory,’’mostwillrespond‘‘defeat’’;whentheyhear‘‘defeat,’’
mostwillrespond‘‘victory.’’ Suchoppositionsaremostcommonfordeadjectival
nouns:happinessandunhappinessarenounantonymsbecausetheyderivefromthe
antonymousadjectiveshappyandunhappy.
Semanticoppositionisnotafundamentalorganizingrelationbetweennouns,butit
doesexistandsomeritsitsownrepresentationinWordNet. Forexample,thesynsetsfor
manandwomanwouldcontain:
{[man,woman,!],person,@...(amaleperson)}
{[woman,man,!],person,@...(afemaleperson)}
wherethesymmetricrelationofantonymyisrepresentedbythe‘!’pointer,andsquare
bracketsindicatethatantonymyisalexicalrelationbetweenwords,ratherthana
semanticrelationbetweenconcepts. Thisparticularoppositionechoesthroughthekin
terms,beinginheritedbyhusband/wife,father/mother,son/daughter,uncle/aunt,
brother/sister,nephew/niece,andevenbeyond:king/queen,duke/duchess,actor/actress,
etc.
Whenallthreekindsofsemanticrelations—hyponymy,meronymy,and
antonymy—areincluded,theresultisahighlyinterconnectednetworkofnouns. A
graphicalrepresentationofafragmentofthenounnetworkisshowninFigure2. There
isenoughstructuretoholdeachlexicalconceptinitsappropriateplacerelativetothe
others,yetthereisenoughflexibilityforthenetworktogrowandchangewithlearning.
22
-25-
Figure2. Networkrepresentationofthreesemanticrelations
amonganillustrativevarietyoflexicalconcepts
group
family
person
relative
brother
sister
natural
object
body
arm
leg
substance
organic
substance
flesh
bone
hyponymy
antonymy
meronymy
41
-26-
AdjectivesinWordNet
ChristianeFellbaum,DerekGross,andKatherineMiller
(RevisedAugust1993)
WordNetdividesadjectivesintotwomajorclasses:descriptiveandrelational.Decriptive
adjectivesascribetotheirheadnounsvaluesof(typically)bipolarattributesandconsequently
areorganizedintermsofbinaryoppositions(antonymy)andsimilarityofmeaning
(synonymy). Descriptiveadjectivesthatdonothavedirectantonymsaresaidtohaveindirect
antonymsbyvirtueoftheirsemanticsimilaritytoadjectivesthatdohavedirectantonyms.
WordNetcontainspointersbetweendescriptiveadjectivesexpressingavalueofanattribute
andthenounbywhichthatattributeislexicalized. Reference-modifyingadjectiveshave
specialsyntacticpropertiesthatdistinguishthemfromotherdescriptiveadjectives. Relational
adjectivesareassumedtobestylisticvariantsofmodifyingnounsandsoarecross-referenced
tothenounfiles. Chromaticcoloradjectivesareregardedasaspecialcase.
Alllanguagesprovidesomemeansofmodifyingorelaboratingthemeaningsof
nouns,althoughtheydifferinthesyntacticformthatsuchmodificationcanassume.
Englishsyntaxallowsforavarietyofwaystoexpressthequalificationofanoun.
Forexample,ifchairaloneisnotadequatetoselecttheparticularchairaspeakerhasin
mind,amorespecificdesignationcanbeproducedwithadjectiveslikelargeand
comfortable. Wordsbelongingtoothersyntacticcategoriescanfunctionasadjectives,
suchaspresentandpastparticiplesofverbs(thecreakingchair;theoverstuffedchair)
andnouns(armchair,barberchair).Phrasalmodifiersareprepositionalphrases(chairby
thewindow,chairwithgreenupholstery)andnounphrases(mygrandfather’schair).
Entireclausescanmodifynouns,asinThechairthatyouboughtattheauction.
Prepositionalphrasesandclausalnounmodifiersfollowthenoun;genitivenounphrases
andsinglewordmodifiersprecedeit.
Nounmodificationisprimarilyassociatedwiththesyntacticcategory‘‘adjective.’’
Adjectiveshaveastheirsolefunctionthemodificationofnouns,whereasmodificationis
nottheprimaryfunctionofnoun,verb,andprepositionalphrases.Adjectiveshave
particularsemanticpropertiesthatarenotsharedbyothermodifiers;someoftheseare
discussed. Thelexicalorganizationofadjectivesisuniquetothem,anddiffersfromthat
oftheothermajorsyntacticcategories,nounandverb.
TheadjectivesynsetsinWordNetcontainmostlyadjectives,althoughsomenouns
andprepositionalphrasesthatfunctionfrequentlyasmodifiershavebeenenteredaswell.
Thepresentdiscussionwillbelimitedtoadjectives.
WordNetpresentlycontainsapproximately19,500adjectivewordforms,organized
intoapproximately10,000wordmeanings(synsets).
WordNetcontainsdescriptiveadjectives(suchasbig,interesting,possible)and
relationaladjectives(suchaspresidentialandnuclear). Arelativelysmallnumberof
adjectivesincludingformerandallegedconstitutetheclosedclassofreference-
modifyingadjectives. Eachoftheseclassesisdistinguishedbytheparticularsemantic
54
-27-
andsyntacticpropertiesofitsadjectives.
DescriptiveAdjectives
Descriptiveadjectivesarewhatoneusuallythinksofwhenadjectivesare
mentioned. Adescriptiveadjectiveisonethatascribesavalueofanattributetoanoun.
Thatistosay,xisAdjpresupposesthatthereisanattribute
A
suchthat
A
(x)=Adj. To
sayThepackageisheavypresupposesthatthereisanattribute
WEIGHT
suchthat
WEIGHT
(package)=heavy. Similarly,lowandhigharevaluesfortheattribute
HEIGHT
.
WordNetcontainspointersbetweendescriptiveadjectivesandthenounsynsetsthatrefer
totheappropriateattributes.
Thesemanticorganizationofdescriptiveadjectivesisentirelydifferentfromthatof
nouns. Nothinglikethehyponymicrelationthatgeneratesnominalhierarchiesis
availableforadjectives:itisnotclearwhatitwouldmeantosaythatoneadjective‘‘isa
kindof’’someotheradjective.Thesemanticorganizationofadjectivesismorenaturally
thoughtofasanabstracthyperspaceofNdimensionsratherthanasahierarchicaltree.
Antonymy:Thebasicsemanticrelationamongdescriptiveadjectivesisantonymy.
Theimportanceofantonymyfirstbecameobviousfromresultsobtainedwithword
associationtests:Whentheprobeisafamiliaradjective,theresponsecommonlygiven
byadultspeakersisitsantonym. Forexample,totheprobegood,thecommonresponse
isbad;tobad,theresponseisgood.Thismutualityofassociationisasalientfeatureof
thedataforadjectives(Deese,1964,1965). Itseemstobeacquiredasaconsequenceof
thesepairsofwordsbeingusedtogetherinthesamephrasesandsentences(Charlesand
Miller,1989;JustesonandKatz,1991a,1991b).
Theimportanceofantonymyintheorganizationofdescriptiveadjectivesis
understandablewhenitisrecognizedthatthefunctionoftheseadjectivesistoexpress
valuesofattributes,andthatnearlyallattributesarebipolar. Antonymousadjectives
expressopposingvaluesofanattribute. Forexample,theantonymofheavyislight,
whichexpressesavalueattheoppositepoleofthe
WEIGHT
attribute. InWordNet,this
binaryoppositionisrepresentedbyreciprocallabeledpointers:heavy!filightandlight
!fiheavy.
Thisaccountsuggeststwocloselyrelatedquestions,whichcanservetoorganizethe
followingdiscussion.
(1)Whentwoadjectiveshavecloselysimilarmeanings,whydotheynothavethesame
antonym? Forexample,whydoheavyandweighty,whicharecloselysimilarin
meaning,havedifferentantonyms,lightandweightless,respectively?
(2)Ifantonymyissoimportant,whydomanydescriptiveadjectivesseemtohaveno
antonym? Forexample,continuingwith
WEIGHT
,whatistheantonymofponderous?
Tothesuggestionthatlightistheantonymofponderous,thereplymustbethatthe
antonymoflight(intheappropriatesense)isheavy. Issomedifferentsemantic
relation(otherthanantonymy)involvedinthesubjectiveorganizationoftherestof
theadjectives?
48
-28-
ThefirstquestioncausedseriousproblemsforWordNet,whichwasinitially
conceivedasusinglabeledpointersbetweensynsetsinordertorepresentsemantic
relationsbetweenlexicalconcepts. Butitisnotappropriatetointroduceantonymyby
labeledpointersbetweenthesynsets{heavy,weighty,ponderous}and{light,weightless,
airy}. PeoplewhoknowEnglishjudgeheavy/lighttobeantonyms,andperhaps
weighty/weightless,buttheypauseandarepuzzledwhenaskedwhetherheavy/weightless
orponderous/airyareantonyms. Theconceptsareopposed,butthewordformsarenot
familiarantonympairs.
Theproblemhereisthattheantonymyrelationbetweenwordformsisnotthesame
astheconceptualoppositionbetweenwordmeanings. Exceptforahandfuloffrequently
usedadjectives(mostofwhichareAnglo-Saxon),mostantonymsofdescriptive
adjectivesareformedbyamorphologicalrulethatchangesthepolarityofthemeaning
byaddinganegativeprefix(usuallytheAnglo-Saxonun-ortheLatinatein-andits
allomorphsil-,im-,ir-). Morphologicalrulesapplytowordforms,nottoword
meanings;theygenerallyhaveasemanticreflex,ofcourse,andinthecaseofantonymy
thesemanticreflexissostrikingthatitdeflectsattentionawayfromtheunderlying
morphologicalprocess.Buttheimportantconsequenceofthemorphologicaloriginof
antonymsisthatword-formantonymyisnotarelationbetweenmeanings—which
precludesthesimplerepresentationofantonymybypointersbetweensynsets.
Ifthefamiliarsemanticrelationofantonymyholdsonlybetweenselectedpairsof
wordslikeheavy/lightandweighty/weightless,thenthesecondquestionarises:whatisto
bedonewithponderous,massive,andairy,whichseemtohavenoappropriate
antonyms? Thesimpleanswerseemstobetointroduceasimilaritypointeranduseitto
indicatethattheadjectiveslackingantonymsaresimilarinmeaningtoadjectivesthatdo
haveantonyms.
Gross,Fischer,andMiller(1989)proposedthatadjectivesynsetsberegardedas
clustersofadjectivesassociatedbysemanticsimilaritytoafocaladjectivethatrelatesthe
clustertoacontrastingclusterattheoppositepoleoftheattribute. Thus,ponderousis
similartoheavyandheavyistheantonymoflight,soaconceptualoppositionof
ponderous/lightismediatedbyheavy. Gross,Fischer,andMillerdistinguishdirect
antonymslikeheavy/light,whichareconceptualoppositesthatarealsolexicalpairs,
fromindirectantonyms,likeheavy/weightless,whichareconceptualoppositesthatare
notlexicallypaired. Underthisformulation,alldescriptiveadjectiveshaveantonyms;
thoselackingdirectantonymshaveindirectantonyms,i.e.,aresynonymsofadjectives
thathavedirectantonyms.
InWordNet,directantonymsarerepresentedbyanantonymypointer,‘!fi’;
indirectantonymsareinheritedthroughsimilarity,whichisindicatedbythesimilarity
pointer,‘&fi.’ TheconfigurationthatresultsisillustratedinFigure1fortheclusterof
adjectivesaroundthedirectantonyms,wet/dry.Forexample,moistdoesnothaveadirect
antonym,butitsindirectantonymcanbefoundviathepath,moist&
fi
wet!
fi
dry.
ThisstrategyhasbeensuccessfulwiththegreatbulkofEnglishadjectives,but
particularadjectiveshaveposedsomeinterestingproblems. Amongthefewadjectives
thathavenosatisfactoryantonym,eveninanun-form,aresomeofthestrongestand
41
-29-
wet
dry
watery
damp
soggy
humid
moist
parched
arid
anhydrous
sere
dried-up
similarity
antonymy
Figure1. BipolarAdjectiveStructure
mostcolorful. Angryisanexample. . Theattribute
ANGER
isgradablefromnoangerto
extremefury,butunlikemostattributesitdoesnotseemtobebipolar. Manytermsare
similarinmeaningtoangry:enraged,irate,wrathful,incensed,furious. Butnoneof
themhasadirectantonym,either. Whenadjectivesareencounteredthatdonothave
directantonyms,theusualstrategyistosearchforarelatedantonympairandtocodethe
unopposedadjectiveassimilarinmeaningtooneortheothermemberofthatpair.Inthe
caseofangry,thebestrelatedpairseemstobepleased/displeased,butcodingangry&fi
displeasedseemstomisstheessentialmeaningofangry. (Andamicable/hostileiseven
worse.) Inordertodealwiththissituation,aspecialclusterheadedangry/notangrywas
created,withcalmandplacid(whichindicateabsenceofemotionaldisturbance)coded
assimilarinmeaningtothesyntheticadjectivenotangry. Thesignificanceofsuch
exceptionsisnotobvious,buttherecognitionthatthereareexceptionsisunavoidable.
Theconstructionoftheantonymclustersisdiscussedinmoredetaillater. We
believethatthemodelpresentedhere—dividingadjectivesintotwomajortypes,
descriptive(whichenterintoclustersbasedonantonymy)andrelational(whichare
similartonounsusedasmodifiers)—accountsforthemajorityofEnglishadjectives. We
donotclaimcompletecoverage.
Gradation:Mostdiscussionsofantonymydistinguishbetweencontradictoryand
contraryterms.Thisterminologyoriginatedinlogic,wheretwopropositionsaresaidto
becontradictoryifthetruthofoneimpliesthefalsityoftheotherandaresaidtobe
contraryifonlyonepropositioncanbetruebutbothcanbefalse.Thus,aliveanddead
aresaidtobecontradictorytermsbecausethetruthofKennedyisdeadimpliesthefalsity
ofKennedyisalive,andviceversa.Andfatandthinaresaidtobecontraryterms
85
-30-
becauseKennedyisfatandKennedyisthincannotbothbetrue,althoughbothcanbe
falseifKennedyisofaverageweight. However,Lyons(1977,vol.1)haspointedout
thatthisdefinitionofcontrarytermsisnotlimitedtoopposites,butcanbeappliedso
broadlyastobealmostmeaningless:forexample,KennedyisatreeandKennedyisa
dogcannotbothbetrue,butbothcanbefalse,sodogandtreemustbecontraries. Lyons
arguesthatgradability,nottruthfunctions,providesthebetterexplanationofthese
differences. Contrariesaregradableadjectives,contradictoriesarenot.
Gradation,therefore,mustalsobeconsideredasasemanticrelationorganizing
lexicalmemoryforadjectives(Bierwisch,1989). Forsomeattributesgradationcanbe
expressedbyorderedstringsofadjectives,allofwhichpointtothesameattributenoun
inWordNet.Table1illustrateslexicalizedgradationsfor
SIZE,WHITENESS,AGE,VIRTUE,
VALUE
,and
WARMTH
. (Themostdifficultgradetofindtermsforistheneutralmiddleof
eachattribute—extremesareextensivelylexicalized.)
Table1
ExamplesofSomeGradedAdjectives
SIZE
WHITENESS
AGE
VIRTUE
VALUE
WARMTH
astronomical
snowy
ancient
saintly
superb
torrid
huge
white
old
good
great
hot
large
ash-gray
middle-aged
worthy
good
warm
standard
gray
mature
ordinary
mediocre
tepid
small
charcoal
adolescent
unworthy
bad
cool
tiny
black
young
evil
awful
cold
infinitesimal
pitch-black
infantile
fiendish
atrocious
frigid
ButthegradinginTable1istheexception,nottherule;surprisinglylittlegradation
islexicalizedinEnglish. Mostgradationisaccomplishedinotherways. . Agradable
adjectivecanbedefinedasonewhosevaluecanbemultipliedbysuchadverbsofdegree
asvery,decidedly,intensely,rather,quite,somewhat,pretty,extremely(Cliff,1959).
Andmostgradingisdonebymorphologicalrulesforthecomparativeandsuperlative
degrees,whichcanbeextendediflessandleastareusedtocomplementmoreandmost.
Itwouldnotbedifficulttorepresentorderedrelationsbylabeledpointersbetween
synsets,butitwasestimatedthatnotmorethan2%ofthemorethan2,500adjective
clusterscouldbeorganizedinthatway. Sincetheconceptuallyimportantrelationof
gradationdoesnotplayacentralroleintheorganizationofadjectives,ithasnotbeen
codedinWordNet.
Markedness:Mostattributeshaveanorientation.Itisnaturaltothinkofthemas
dimensionsinahyperspace,whereoneendofeachdimensionisanchoredatthepointof
originofthespace. Thepointoforiginistheexpectedordefaultvalue;deviationfromit
meritscomment,andiscalledthemarkedvalueoftheattribute.
Theantonymslong/shortillustratethisgenerallinguisticphenomenonknownas
markedness. InanimportantpaperonGermanadjectives,Bierwisch(1967)notedthat
Documents you may be interested
Documents you may be interested