n41
TYPOLOGICAL STUDIESIN LANGUAGE (TSL)
A conrpanion
series
ro thejournal.,STUDIES
IN LANGUAGE.
Hortorary Editor: JosephH. Greenberg
GeneralEditor: T. Giv6n
Editorial Board:
Alton Becker (Michigan)
WallaceChafe (Berkeley)
BernardComrie (LosAngeles)
Slott
lglancey (Oregon)
Ge-rard.Diffloth(Chicago)
R.M.W. Dixon (Canberra)
JohnHaiman (Winnipeg)
Kenncth[.lale (Canrbtidge,
Mass.)
Bernd}leine (Kciln)
paul Hopper (Binghamton)
Margaretiangdon (SanDiego)
ChailesLi (SantaBarbara) -'
JohannaNichols (Berkeley)
Andrew pawley (Aucklani)
Fransplank iKonsranz)
Gillian Sankoff lphiladelphia)
Dan Slobin (Berkelev)
SandraThompsonil-os Angeles)
Volunresin this seriesrvill be functionallyand typologicaily
oriented,cove_
ringspecifictopicsin Ianguage
by collectingtogetherdtta from a widevariety
of languages
andlanguage
typologies.
The orientationof the volumeswill be
substantiverather than formal. with the aim of investigating
universals
of
human languagevia as broadly defined a data baseas possible,leaning
towardcross-linguistic,
diachronic,developmental
and live-discourse
data.
The seriesis. in spiritaswell asin facr,a coniinuationof the traditioninitiated
by C. Li (Word Order and Word Order Change, Sbject and. Topic,
l echatrisrrts
for SyntacticChange)andcontinuedby T. G iv6n (Discourseand
Syrrta.r)and P. Hopper (Tenseand Aspect:BetweenSemanticsand pragma_
lic.t,r.
NOUN CLASSESAND
CATEGORIZATION
Proceedings
of a Symposiumon
Categorizationand Noun Classification,
Eugene,Oregon, October 1983
editedby
COLETTE CRAIG
Universityof Oregon
Volume 7
Colette Craig (ed.)
JOHNBENJAMINSPUBLISHINGCOMPANY
Amsterdam/Philadelohia
NOUN CI,,4SSE'SAND CATEGORIZATION
1986
vi
TABLE OF CONTENTS
KAREN ADAMS
Numeral Classifiersin Austroasiatic
COLETTE G. CRAIG
JacaltecNoun Classifiers:A Study in Languageand Culture .
241
263
PREFACE
IV SEMANTIC AND PRAGMATIC F\]NCTIONS OF CLASSIFIERS295
J. PETER DENNY
The SemanticRole of Noun Classifiers
297
PAUL J. HOPPER
Some DiscourseFunctionsof Classifiersin Malay . . . . . . .
309
A.J. BECKER
The Figurea ClassifierMakes:Describinga particularBurmese
Classifier
327
PAMELA DOWNING
The Anaphoric Use of Classifiersin Japanese
345
HISTORY AND ACQUISITION OF CLASSIFIERSYSTEMS
MARIANNE MITHUN
The Convergenceof Noun ClassificationSystems
MARY S. ERBAUGH
Taking Stock: l'he Developmentof ChineseNoun Classifiers
Historically and in Young Chilqren .
SCOTT DELANCEY
Toward a History of Tai ClassifierSystems
KATHERINE DEMUTH, NICOLAS FARACLAS, LYNELL
MARCHESE
Niger-Congonoun classand agreementsystemsin language
acquisitionand historicalchange .
453
LANGUAGE INDEX
473
LIST OF CONTRIBUTORS
479
379
399
437
of a symposium
on Categorization
Thisvolumecontainsthe proceedings
heldon theUniversityof OregonCampusin October
andNounClassification
1984,underthe sponsorship
of the LinguisticsDepartment,thc Psychology
Department,and the CognitiveScienceProgram.
I had initiallyapproached
T. Givon for somecommentson a draft of a
paperI had writtenon noun classification
in JacaltecMaya. His main comment was "Interesting!Let's havea conferenceand talk about it!" In his
inimitablefashion,he took the list of references
at the endof the paperand
drewa preliminarylistof participants.
That'showthesymposium
on categoriwereinvitedin
zationand noun classification
cameto be. The participants
an effort to bring variousdisciplinestogether- linguistics,psychology,
philosophy,anthropology- and to gathera representative
sampleof the
classifierlanguages
of the world.
andeffectively
I wishto thankall thosewho participated
enthusiastically
in the work of the symposium.
participants
Not all symposium
contributeda
paperfor the volume,and the volumeincludesa numberof paperswhich
werenot presentedat the symposium.
Theseadditionsweresoughtin order
to makethe volumeascomprehensive
aspossibleon the matterof linguistic
A completelist of the contributoffifollows.
classification.
gratefulto T. Givon for his supportiveenergyand to all
I am especially
the contributorsfor their cooperationin assembling
the volume.
ColetteG. Craig
THE CONVERGENCE OF NOUN CLASSIFICATION SYSTEMS
MARIANNE MITHUN
SUNYAlbanY
dtverseas
as typologicallyand geographically
A varietyof languages,
Mandarin(seeErbaugh)and Jacaltec(seeCraig) sharestrikinglysimilar
oneanotherbothin theirreperTheyresemble
of nounclassification.
systems
or
Is thissimilarityaccidental,
them.
in
their
uses
of
and
toiresof classifiers
of
operation
categorial
as
the
forces,
such
underlying
doesit reflectcertain
A
different
teleologically?
ariseby chance,or
humanminds?Do suchsystems
languages,
system,alsofoundin a varietyof widelyscattered
typeof classifier
is no accident.This secondtype of system,
that suchclassification
suggests
basedon incorporatednouns,doesnot originateasan overt nounclassifier
whoseprimaryfunctionis to qualify
systemat all, but ratherasa construction
can
Overtime,the construction evolveinto onewhosesecondary
predicates,
effectis to classifynominalsovertly.The resultingsystemsharesthc primary
of Mansystems
ofthe prototypicalnounclassifying
functionalcharacteristics
darin,Jacaltec,and others.
1
lncorporation
Noun incorporationis a lexicalprocesswherebya noun stemand verb
to form a derivedverb stem,asin AustralianNgandi
stemare compounded
throw' -, -galgeykdg-'tospit' (Heath 1978)
+
-geyk-f,a-'to
-gal- 'saliva'
do not refer;theyqualnouns
in othercompounds.
like
lncorporatednouns,
The
noun root -ryal-in '
verbs.
host
ify, narrowingthe scopeof their
it
gatgeyk(a-doesnot refer to a specificblob of saliva; qualifiesthe typeof
throwing predicatedby the host verb. lncorporatednounsare thus not
specificity,or number'They
markedfor suchdistinctionsas definiteness,
they
althoughsemantically
of
the
clause,
role
as
arguments
haveno syntactic
verbs
They
host
of
their
locaiions
or
mayqualifythe patients,instruments,
are thus unmarkedfor case.
380
MARIANNE MITHUN
1.I . Stagel: New Compoundsfor new concepts
The processof incorporationdevelopsin language
in a predictableway
(N,Iithunms.b).lt beginsasa strategyfor creatingnamesfor newcategories
of actionsor states.A nounandverb arejoinedto form a singlelexicalitem
denotingan institutionalized,
unitaryconcept.Not surprisingly,
the resulting
verbalcompoundsin a languagetypicallyreflectthe culturalconcernsof its
speakers.ln somecases,thesemay be relativelyuniversal,suchas Ngandi
-,rii- 'thought'+ -0a-'hear'-, &yii7a-,to think', or the compoundfrom
Caddo,a Caddoanlanguage
of Oklahoma,consisting
of -nak-,fire'* :a'nih_
'nrake'..*' -nqk'a'nih-'to build a fire' (Chafe 1977).ln other cases,the
compoundsare more area-or culture-specific,
as in the compoundsfrom
Cayuga,a Northern Iroquoianlanguagespokenin Ontario, where-a'kr_
,for snowto melt', _frka_
'snow'+ -.tfl.twe'
'to becomewet'-, -a'kranawe'
or
'peg' + -ol 'to stand'-' -hkaot'to tap trees(by insertinga woodenpeg)'.
L2. Stagell: The manipulalionof case
Onceestablished,
the processmay be furtherexploitedfor the manipu_
lation of casc relationswithin clauses.When a noun is incorporated,its
vacatedsyntacticrole (subject,directobject,absolutive
etc.)maynot disap_
pcar. but ratherabsorban otherwiseobliqueargument.A compoundof this
t),pe is the Ngandi -nraga- 'neck' + -gulk- ,cut' -,, -msgagalk- .to hang
someone'.The verb remainstransitive,but the possessor
of the neck,more
significanlthan tlte neck itself, is promotedto absolutivestatusand refer_
encedin the pronominalprefix.The neckis backgrounded
by incorporation
to qualifying
status.
(l)
Ngandi(Australian,Heath 1978)
Baru-ga-maga-gulk-!-i
ni-wolo ntyul_p-yug.
3pl/Ms-sub-neck-cut-aug-past-prf
Ms-that Ms_abrgnl_nom_abs
'They hangedthat aboriginal.,
Compoundsof thistype maybe intransitiveaswell. Whenthe subiectof an
intransitiveverb is incorporated,an obliqueargumentmaybe promotedto
the vacatedsubject(or absolutive)role. The Ngandicompound_ganan_
'ear' + -dam- 'be closedup' -> ganamdqm,bedeaf is of this type. The
promoted possessoris referencedwithin the pronominal prefix, as before:
ga-ganamdam'lam deaf'.
SYSTEMS 38I
OFNOUNCLASSIFICATION
THECONVERGENCE
1.3. StageIII: The arrangementof information in discourse
In some languagesthe processmay evolve still further into a mechaSuch
nism for backgrounding known information within discourseaffixes
pronominal
with
obligatory
polysynthetic,
are
typically
languages
within the verb, as in the Ngandi gaganamdam'1 am deaf or
battgamagagulkdi'they hanged/rlm" Independentnoun phrasesare used
within clauses'
to identify referents,but are not grammaticallynecessary
verbsto nouns
proportions
of
higher
much
typically
exhibit
languages
Such
like English.Not surprisingly,
(asmanyas 10:1)in discourse
than languages
muchinformationis generallycarriedwithin the verbby meansof modifying
affixes.
When an entity is first introducedinto discourse,it is usuallyidentified
by a full independentnoun phrase.In subsequentdiscourse,the noun phrase
may be omitted, sincepronominalaffixesrefer to the entity in question.
however,the pronounaloneis not sufficientto qualifyverbsof
Sometimes,
the entity' now old
wide scope.Yet a separatenoun phrasere-identifying
information,would sidetrackthe attentionof the listener.The solutionis
in themselves,
incorporation.Incorporatednouns,not salientconstituents
presence
is sufficient to
yet
their
flow
of
information,
the
obstruct
do not
thusprovides
incorporation
third
type
of
verb.
this
narrowthe scopeof the
old informationwithin discourse.In the Muna meansof backgrounding
durukd sentencebelow, the water is first introducedby a separatenoun,
then,in the next clause,it is incorporated.
(2)
Mundurukf (Tupi stock,Brazil,Sheffler1978)
ip
baseya'a he
Ti
dojot puye, o'timog
in
water bring when they-water'placethey basin
'When they broughtwater,they placedit in a basin.'
Comparethe positionsof the noun ohQn'ata'(root -hQna't-)'potato' in the
Cayugaconversationbelow. In Cayuga,as in most of theselanguages,constituentsare orderedaccordingto their relativeimportanceto the predication. New, crucialinformationappearsearly, while old or lesssignificant
informationappearslater.
(3)
A.
B.
Cayuga(lroquoian.Ontario.Henryp c )
('What are you going to plant this year?')
akwt: onqhQ' osahe'td'ohqn'ati'q hni',
A:y(!
too
corn bean
it-seemsall
Potato
'Oh, everything,I guess,corn, beans,and potatoestoo;
3S:
SYSTEMS
THE CONVERGENCEOF NOUN CLASSIFICATION
MARIANNEMITHUN
B.
gyakway(thwahsp:',thr1hs a:y(:'
kwahs i:s6:'a
just
much-ishwill-we-all-plant for
it-seems
we'll plant quite a lot, becauseit seemslike
o:ne b:kehs kand:'
oh(n,ata'.
now sure
it-expensive potato
potatoesare reallyexpensive
now.'
Swqhgn'ataygthw( hne:' t'ike tshike:nhth?
you-all-potato-plantthen ? when-it-summer
'Did you potato-plantlastsummer?'
Thq' tlfine:', thrihs to:kihs akasta:nohw6'nhe'.
no there-the for
sure
past-it-drop-throw
'No, in fact we didn't, because
it rainedso much.
A ketshahni'k a:yghpn'atd:tkeh.
past-I-fear would-it-potato-rot
I wasafraidthey would potato-rot.'
Whether old informationis representedby an incorporatednoun or
simply by a pronominalprefix is determinedprimarilyby the verb. Some
verbs,particularlythoseof relativelywide semanticscope,whosemeaning
is significantlyaffectedby their patients,tend to incorporatewhen€verpossible.Others,especially
thosewith narrowerscope,neverincorporateat all.
Although the primary function of the incorporatednoun is to qualify the
hostverb, it hasa secondary
€ffect.It retainsthe entity in questionwithin
the arenaof discourse.
A distinguishingfeature of this third type of incorporation is its high
productivity.Sincenearlyany entity can becomeold information,a large
numberofnounsareincorporated.
Severalfactsindicate,however,thatboth
the creationof new incorporatingforms and the manipulationof largenumbers of such forms for stylistic purposesare relatively difficult operations.
Suchincorporation
is oneofthe lastlinguisticskillsto beacquiredby children.
(SeeMithun ms.aon Mohawk.) It is one of the first to disappearin language
death.(SeeMithunandHenryms.on Cayuga,Comrie1981:251
onChukchi,
Weltfish 1937:viand Parks1976:250on Pawnee.)The ability to useincorporation for backgrounding
entitiesin discourseis one of the mostsalientdifferencesbetweenespeciallyadmiredspeakersof incorporatingIanguages
and averageor marginalspeakers.
This difficultylimits to someextentthe
numberof suchlexicalitemsthat are createdand used.
As severalgrammarianshave noted, incorporatednounstend to be
relativelygeneric.Suchnounsare not only more usefulqualifiers,but also
3I't3
fewerin numberthanthe specificnounstheyinclude.Osborne(1974)notes
of ArnhemLand, conthat Tiwi, for example,anotherAustralianlanguage
tainsat leastfourteenseparatefree nounsfor typesof wild honey,but only
noun,a genericterm whichincludesallof them.Thisleads
oneincorporable
At first mention,
systemof nominalclassification.
to a covertbut pervasive
discourse,
phrase.
In
subsequent
an entityis identifiedby a full, specificnoun
a more generic noun stem is incorporatedin its steadto narrow the scope
of verbs.
2.
The developmentof oYertclassification
astheirindependent
nounsthusdo not identifyarguments'
Incorporated
counterpartsdo, but rather simply narrow the scopeof their host verbs,
implying that a particularclassof patients,instruments,or locationsis
effect:theovert
involved.Overtirne,thisoperationcandevelopa secondary
nominals.
of
classification
2.1. Settingthe stage
svstem'
of a classificatory
Threefactorssetthestagefor thedevelopment
is the
second
The
nouns.
incorporated
of
qualifying
function
is
the
The first
to
the
ditTiculty
partly
due
and
function
qualifying
factthat, partlydueto this
of the processitself,only relativelygenericnoun stemsare incorporated'
body
containingincorporated
of compounds
The third involvesthe Presence
parts.
It is well knownthat termsreferringto body partsare amongthe most
frequentlyimcorporatednouns.(See,for example,Sapir1911) This is not
activitiesinvolvethem, suchas
surprising,sinceso many institutionalized
arefrequentlyusedto
etc.Type I compounds
tooth-brushing,
face-washing,
expresssuch activities.More importantly,speakersare normally more
of a body Partthan on
interestedin the effectof an eventon the possessor
the
body part and result
can
background
Type
II
incorporation
part
itself.
the
asseenabove'
case
role'
primary
possessor
to
a
in a promotionof the
any
language, nor are
in
Proper namesare not normally incorporable
nounsreferringto particularhumanbeings.They are not sufficientlyclassificatory to narrow the scope of verbs in a useful way A verb like 'to
is unlikelyto find its way into the lexicon.Most incorPoraling
Susan-see'
however,do havecompoundslike thosebelow.
languages,
38,1
(1)
MARIANNE MITHUN
a. Ngandi
baru-waIga-kar-i -p ula
3pl/Ms-body-chase-past
pnct-dual
'they pursuedhim'
Sentences
containingsuchverbsopenthe door to an ambiguousinterpretation. What is the relationship
betweentheincorporated
nounandtheexternal
noun phrase?Is the 'body' possessed
or is it a hyponym?
(4)
b. Ngandi
Baru-walga-kar-i-pula Vi-wolo 4iyul-e-yug.
they bodychasedhim
that
Aboriginal
'They pursuedthat Aboriginal'sbody.'
(or 'They pursuedthat Aboriginal'?)
TH E CONVERGENCEOF NOUN CLASSIFICATION
SYSTEMS
gu-na-ji-fi-u11
gu-bottle-gamgar0'-yug ni-bun-gu-tini.
GU-that-kind-o-abs GU-bottle-big-abs
3Ms-ryarer-eat-p
(i.e.
that kind (in) the big bottle
hard liquor).'
At this point, the systemhasbecomeovertly classificatory.The incorporated
noun signalsthe class (type of patient) to which the external noun phrase
bclongs.
As in Type III incorporation, once the argument has been identified,
the incorporated noun alone is sufficient to qualify succeedingverbs. In the
Cayuga text below, the stem -nq'ta- 'baked goods' is used to qualify verbs
in place of the noun phrasetekdhswa'ne:t'pie'.
(7)
'l'he incorporatednoun for 'body' can becomea usefulmeansof qualifying
a verb pcrtainingto the physicalaspectof a personor animal without
backgrounding
the individual.
(5)
Cayuga
Akhey'atihsd:k ng:ky( ak(:kweh.
lbodysotghther that
woman
'l missedthat woman.'
Incorporationof termsfor thingssuchas'body'and'mind' alsosetthe stage
for the simultaneous
appearance
of an incorporatedclassifierand an overt
externalnoun phrase.
Onceincorporation
of thethreetypesdescribed
abovehasbecomeestablishedin a Ianguage,
the processmaydeveloponestepfurther.Certainverbs
of wide scopeoccur lessand lessfrequentlyby themselves,
and more frequcntly witlr particularincorporatednouns.At a certainpoint, thescwcll
establishedcompoundlexical items begin to be usedwith externalnoun
phraseswhichfully identifythe argumentimpliedby the qualifyingnoun. In
thc Ngandisentencebelow.the incorporatednoun -bun- in the compound
-broi-'\\,ater'+ -r[- 'eat'-, 'drink', cooccurswith full nounphrasesspecifying the typeof liquidconsumed.
((,
Ngandi
Gu-na-ji-ri
gu-bata-jtrnbaka'-wii-ug gu-ja1k-yur1
GU-that-kind-lrnmGu-con-can-having-absGU-water-abs
'He drinks that kind of liquid in the can (i.e. beer)and
Cayuga
N4:' ki' kyq:' hya: aekhnina'ti:k
it-is just this then would-we-two-bakedgoods-eat
'Before anything else, we should eat
akatn'atiqt
tekdhswa'ne:t.
l- b aked go o tl s-put in ov en tw o-it- lay ered-is
the pie I baked.
Kwahs ws'hd tshq: akatn'atagtd:kwgh.
just
now
only l-baked goods-put in oven-un-have
I j ust took it out of t he oven. '
In Mundurukri, thc classificatorynounsare alsocompoundedwith cxternal nouns.
(8)
2.2. The appearance
of overt classification
3ft5
Muncl u r uki ( Cr olt s cit e( l in Com odo l9l3l)
ws2ra2ruJ'a2cu4 dtxlcklco2co2-cn2 o' 1-su2-enJ-ro2-iott
pig-meat
she-it-ruecl-caus-come
another
'Anothcr brought some pig meat
pd.
i3-6nt-yoy2 am2. O'3-su2-ant-pipa ip'
stick
it-rneal-brkc for
she-it-tneat-put
on
to bake. She put it on a stick.'
As in Type III incorporation, although the primary function of an tncorporated noun is to clualifythc verb, it alsoservesto retain the entity it implies
within the arena ofdiscourse. Although it is not strictly speakingreferential,
and thus not anaphoric, it keeps referenceclcar by narrowing the scope of
the verb.
..ttJ6
M AR IAN N EMIT H UN
2.2.1. The semanlicbasisof classification
Althoughtheclassification
hasbecomeovertat thispoint,it stilloperates
on a semanticbasis,irnplyingcertainfeaturescharacteristic
of the referent
of the noun phrase,not certainlexicalfeaturesof a particularnoun. The
propertiesimpliedby the classifier
may or may not be overtlyexpressed
in
the noun phrase,so in a sense,the classifiers
may or may not be redundant
in Denny'ssense(thisvolume.)Comparethe Caddosentences
below.
(9)
Caddo(Caddoan.Oklahoma,Chafep.c.)
Kapi: kan(d:ni'ah.
coffee liq ui d-buy-past
'He bought(liquid)coffee.'
Kapi: dAn:fitri'ah.
coffeep ow der-buy-past
'He bought(ground)coffee.'
The separatenoun phrasemay add only deicticdistinctions
or someother
rnodificationof thc refcrent. ln the Caddo sentencebelow, the external
subjectis a dcmonstrative.
(10) Caddo
Di: tisdy'ah.
tis house-be
'This is a house'
Iu the Cayugasentence
below,the externalnoun phraseis just a number.
(ll)
Cayuga
Kdi niuttkwiytig'.
e
fottr so-l-children-hav
'l havefour children.'
2.1.2. l'he questionof categorylevel
In fact, any incorporablenoun.representing
anycategorylevel,can be
usedclassificatorily
in languages
whereincorporationof thistype is productive. Considerthe Cayugasentencebelow,for example.
(12) Cayuga
Ohen'atafte: akhdn'ald:k.
it-p otato-ro tlen past-l-potato-eat
'I ate a rotten potato.'
TH E C ONVERCENCEO FNO UNCLASSI FI CATI O NSYSTEM S387
The external noun phrase ohqn'atdtkq: is morphologically a verb. Verbs
function regularly as nominals in Cayuga, but they cannot be incorporated.
(lt is possible to nominalize a verb by means of a suffix, then incorporate
the result, but such an operation is cumbersomeand done only for nameworthy entities.) Yet the verb -k 'eat' incorporatesits patient if at all possible.
This presentsa conflict: the noun phraseshould but cannot be incorporated.
lncorporation of the noun stem alone providesa solution. Suchincorporution
is strictly speaking, classificatory:the set of rotten potatocs is includcd in
the setof all potatoes.This classificationis not quite the sameasthe prototypical classificationof languageslike Mandarin and Jacaltec, however. -Ihe
incorporated noun representsa basic level category, and the external noun
phrase,ohen'atdtk€:'rotten potato', a subordinateone. In the prototypical
systems.classifiers:rre hypernyms.
Structurallyequivalentconstructionsdo expressthis generic(supcrordinate) - basic level relationship. Sentenceslike the Cayuga one below are
common.
(13) C ayu ga
So:wd:s ukhnihskwag'.
dog
l-domesticanimal-haye
'l have a (pet) dog.
The grammatical system does not differentiate category labelsaccordingto
level. This may not be an accident.The notion of level is not alwaysas clear
cut as one might hope. ln some cases,the part test discussedby Tversky is
revealing. In other cases,parts do not take us as far, becausefcwer parts
are i nvol ved. For English spcakcr sat lcast , it scc|] r sclcar t hat 'f r uit 'r cpr csents a superordinat ecat egor y,nouns like'applc' and'banana' r epr cscnt
basiclevel categories,sharing some attributes but differing in parts like the
core, while 'pippin' and 'delicious'representsubordinatecategories,sharing
partswhile differing in other ways.What about a noun like 'mcat'? It typically
hasfew characteristicparts, and in some casesnot an especiallycharacteristic
shape.Are pork, chicken, and beef basiclevel categories,or are they subordinate?The answerprobably dependson the experienceof particularspeakers, context, and the nature of the instances.A person who never cooks.
takes little interest in food, and routinely encoutersmeat in the form of a
neutral colored lump, may consider 'meat' a basic level category. A cook
looking at a picture of a roast chicken or bacon, on the other hand, would
be highly unlikely to identify these simply as'meat'. Real world taxonomies
llrs
M AR IAN N EM I' IH U N
SYS-TEMS
THE CONVHRGENCEOF NOUN CLASSIFTCA-IION
are not as clear cut with respectto category level as might be hoped. Incorporated classifiersthus do not obscurea distinctionof level, but rather reflect
an aspecl of their reality.
Category level is furthermore subject to changeover time. The Cayuga
noun o/rnydhsa''cucurbit' ('squash/gourd/melon/cumcumber')originally
represented a basic level category. Now the category includes other
categoricswhich are obviously relatively recent, as shown by the morphologically transparency of their names and by the nature of their referents:
oltnvphsakihteh'watermelon'(literally 'raw cucurbit'), o hnyghs'dqweh'hubbard squash'(fiterally 'prototypical cucurbit'), ohny6hskwrie''cucumber'(literully'cucurbit covered with bumps'), and otsi'tkwd:' ohny|hsa' 'pumpkin'
(l i tcra lly ' y ellow c uc u rb i t' ).
form of the verb root :tr- 'drag'.The externalnominalk'atrfhta'aboveis
redundant.To tell someonethat I own a car, I shouldsay:
considered
2..1.The development of the classifiers
All classificatorystemsbegin life as nouns. Some,especiallythose which
had a generic senseto begin with, retain this sensewhen used as classifiers,
likc tlre Cayuga -rtal]6kw-'domestic animal', or the Muduruku -Cn 'meat' .
More often, however, classifiersoriginate as relatively concrete nouns
of narrow scope.When used as independentnominals, they retain their concr!-tc. spccific sense,When used as classifiers,thev assumea more generic
sensc,like the Caddo tisa-'house'-, 'building', or the C ayuga-na'ta-'6read'
-r 'baked goods'. The useof the samestem to representcategoriesof different levcls reveals an interesting aspect of this inclusion relation. Speakers
do not use general classificatorynouns to classifytheir concrete cognates.
Ca1'ugasentenceslike those below are fine,
(14) Cayuga
Skiti
ake'lrehtdg'.
skidoo l-vehicle-have
'I have a skidoo.'
Similar constructionscan be used for trucks, busses,sleds,and motorcycles,
but not for cars.
( l 5 ) Cay uga
* K'atrehta' ake'trghtdg'
car
l-vehicle-have
*' I hav e a c a r.'
The stem for'car' is the same as the classifierfor vehicles,a nominalized
3IJ9
(16) Cayuga
Ake'trchtde'
I-vehicle-have
'l havea car.'
and their free cognates
Althoughpresumablythe incorporatedclassifiers
servedifferent functions,their combinationis consideredunacceptably
r€dundant.
phenomenon
occurswhenthelink ofcognationisbroken,
An interesting
personior
speaking
the nameof a deceased
to
tabus
against
Due
however.
languages
many
Australian
resemble
the
name,
evcnwordswhichsomehow
Typically,free nouns
havea particularlyhigh levelof lexicalreplacement.
of compounds'
the constituents
arereplaced,but their boundcounterparts,
The resultsis a low matchbetween
mayremainintactwithinthe compounds.
Oncethe cognate
nounsand their free semanticcounterparts.
incorporated
and
a moregeneral
with
reference
a
noun
specific
separate
rclationbetween
is
apParently
acceptable.
classifieris gone,their combination
incorporated
Hcrth notesthat the combiningform of the Ngandinoun root for'water'
cquivalent
for 'liquids')is -bun-,whilethesemantically
(andalsotheclassifier
jQfk.
which
the
sentence
below,
nouns
is
He
cites
in
separate
root occurring
containsboth.
(17) Ngandi
Gu-ja1k-yur1 ba'ga-bun-gu-ni
G U-water-atx3pl-sub-waler-eat-pcon
'and they drank water'
Oates (1964) reports that in Gunwinggu, another Australian languageof
ArnhemLand, the combiningform of the term for 'water'(and 'liquid') is
-bo:-,while the free noun is gugga.This combinationis alsoacceptable.
(18) Gunwinggu(Australian,Oates1964)
Gugu gabo:manganwater it-liquid-fall
'Water is falling.'
equivalentrootsmaycooccur,providedthat the free
As in otherlanguages,
noun.
to theliteralsenseofthe incorporated
information
adds
nounDhrase
M AR IAN N EM IT H U N
(19) Gunwinggu
gunbo:weyn.
laho:gun
I-li4uid-consume GUN-wcter-much
'I drank muchwater.'
The sameis true of Cayuga.The sentences
beloware perfectlyacceptable.
(20) Cayuga
Ohnekaka'(
skhn€kgha'
it-liquid-delicious-infr
will-I-liquid-consume
'l guessI'll havesomepop.'
O'trchtatki'
ake'trghtdQ'
it-vehicle-dirtyl-vehicle-have
'l havea lemon,'
Over tinre, certain of the classifiersevolveone step further. They shift
from indicatingkindsof entitiesto qualities.Considerfor exampleone of
the more commonclassifiers.The stemfor'water' often becomesa classifier
for all (potable)liquids.Liquids,especially
potableones,certainlyconstitute
a genericclass.They are also distinguishedby a more generalquality, their
liquid state.At a certainpoint the stemcomesto imply anyentityin a liquid
state. Such classificationby quality rather than kind representsthe final
extensionof classificatoryincorporation.Concretenoun stemscome to representclassesof categoriesdistinguishedby shapeand/or consistency.The
Caddo classifierfor small round objects, -ii'aft-, is also the noun stem for
'eye'.The Mundurukdclassifier
for circular/spherical
objects,a2,is the noun
'head'.That for long, rigid, cylindricalobjects,baa,is the noun for.arm'.
That for long, flexiblecylindricalobjectsis baz,the noun for,finger,.
The selectionof a particularlexicalitem to representa setof categories
is an intriguingissue.Why is the noun for water usuallyusedas the liquid
classifier?
Why did the Caddo'select'the term for'eye'insteadof, say,that
lbr'stone', or'plum' to representall smallroundobjects?Why do the Mundurukriusethe term for'arm'to representrigid, cylindricalobjects,instead
of thal for 'leg' or 'stick'?One might be temptedto arguethat the cover
terms representlhe prototypical membersof the categoriesthey standfor.
They are characterizedby the largestnumber of the most important criteria
for group membership.It could even be arguedthat their own most salient
propertiesarejust these.Water,for example,is not only a moreprototypical
liquid than molasses
or chickensoup,it alsohasfewerdistracting
non-essential properties.The choiceis in fact not that purposeful,however.
SYSTEMS
THE CONVERGENCEOF NOUN CLASSIFTCATION
39I
the most
oftendo stemfrom nounsrepresenting
Althoughthe classifiers
theyimply,their selectionis actually
membersof the categories
prototypical
constructions.
Oncecoined,
a functionof the lexicalstalusof incorporating
compoundsare learnedas lexicalunits by speakersand usedas such,not
createda new eachtime they are used.Termsfor entitiesthat appearespeexperience,
in a varietyof contexts,are likely
ciallyfrequentlyin speakers'
to appearin the largestnumberof frequentlyusedlexicalcompounds.It is
of classifiers.
that serveasthe modelfor the establishment
thesecompounds
in
more
contexts
than
water
appears
more
frequently,
cultures,
In most
frequently
used
The nounfo|water is tbusa constituentof more,
molasses.
It is not surprisingthat thisterm shouldbecomethe liquidclascompounds.
in Caddoandespecially
sifier.The useof termsfor body partsasclassifiers
in Muduruku has the sameexplanation.Suchtermsare amongthe most
frequentlyincorporatednouns.
Does the choiceof particulartermsto serveasclassifiersaffectthe ideas
generations
Apparof speakers?
concerningprototypicalityof subsequent
for 'fruit' is -ar)-, alsothe term for'berry'.
entlynot. The Mohawkclassifier
This might seemstrange,sinceall Mohawksconsumemany more apples,
oranges,and bananas,than berriesof any kind. The reasonfor the choice
wasestablished
longbeforeMohawkshad
of thisterm is clear:the classifier
comeinto contactwith apples,oranges,and bananas.When askedto give
a list of fruits in Mohawk, modern speakersinvariably begin with
'apple'.The orderof the fruitswhichfollowvariessomewhat
sewahy6:wane
to
speaker,andappearsto be determinedpartlyby frequency
speaker
from
and partly by the complexityof the Mohawkterm, but the
of appearances
term for 'berry', kdhi, appearcvery late, if at all. Specificterms for 'strawberry'and'blackberry'(literally'longkahi')generallyappearafterthosefor
'apple','banana','pear','grapes',etc.The useof a noun asa classificrdoes
when usedas a free
not evenpreventthe noun from shiftingsemantically
nounphrase.The liquidclassifier
-lnek- of theNorth€rnlroquoianlanguages
whenusedas a free
meaningof 'whiskey/liquor'
hasassumeda specialized
noun as in Caytgaohn(:ka'. When usedto designate'water',the stemis
incorporatedinto the verb 'cool''.ohnikanos'coolliquid'-) 'water'.classifiercare thusoftencognatewith termsfor the prototypcialmembersof the
they represent,but not necessarily
so, and appearto havelittle
categories
prototypicality.
on
ideas
about
effect
392
3.
MARIANNE MITHUN
Solidificationof lhe system
All incorporationdiscussedup to this point involvesproductivelexical
rules.As notedabove,however.it can be a relativelydifficultoperationto
perform,andis correspondingly
lragile.Its evolutionmuy
ut unyporn,.
""ur"
When the processlosesits productivity, speakerssimply stop
inventingnew
noun-verbcombinations,althoughthey continueto usethosealreadyin the
lexicon.Over time, normal processes
of linguisticchangeobscurethe rnternal
structure of the compounds.phonologicalchangesblur morphemebound_
aries. Semanticchangesskewthe semanticrelationshipsbetwien the
bound
constituentsand their free cognates.Lexical replacementmay eliminate
the
free cognatesfrom the lexiconaltogether.The resultis an everdiminishins
pool of increasingly
opaquerelic compounds.When the lexicalrules losJ
their produclivity,the set of incorporated
classifiers
simplybecomesclosed.
The existingclassifiers
continueto function,however,evenafter theyhave
lost their formal transparency.
Speakercontinueto use the oncecomDlex
verb stemsappropriately,evenwith respectto newnominalsin the language.
An exampleof a systemat thisstagecanbe foundin Cherokee,spoken
in North CarolinaandOklahoma.Relicformsremainof all typesof incorpo_
ration. Among theseare setsof verb stems(King 1976counts30) whose
selectionis determined by specificpropertiesof their patients, like those
below.
(21) Cherokee
gakaneha
ganehneha
adeha
ganvneha
ahneha
(Southernlroquoain,Oklahoma,pulte 1975)
,he'sgiving
him a livingthing,
,he'sgiving
him someliquid'
,he'sgiving
him a long,rigid object,
,he'sgiving
him a flexibleobiect,
,he'sgiving
it to him. (sometiring
not containedin
one of the abovecategories)
They are usedas follows.
(22) Cherokee
Ije digohwelodi udelv
didehlohgwasgi.
newpencil
he/him-long,rigi d/gavestudent
'He gavethe studenta new pencil.'
The verbsmay be eithertransitive,asabove,or intransitive.A highproportion of the alternatingsetsdescribethe positionand/or manip-ulation
of
objects,such as 'lay', .be in water', ,be on a raisedsurface,,or .handle,,
O F NO UNCLASSI FI CATI OSYSTEM
N
S
TH E C ON VERCENCE
and not
'carry','pick up', 'pushaside',etc. This is typicatof suchsystems,
at all surprising, since the shape and consistencyof the patient involved
affectsthe stateor actionpredicated.Not all verbshaveall five
significantly
forms, since not all combinationsare pragmaticallycompatible,such as
'liguid'+ 'placein a row', or'animate' + 'cut off'.
The alternatestemsdiffer in form in suggestive
butnot perfectlysystematic ways.They are clearlylearnedas separatelexicalitems by speakers.
betraytheir origin,however.A numberofverbs
Certainrecurringsegments
-ne-. Althoughthis is not
implyingliquid patientsbeginwith the sequence
cognateto the Cherokeenoun for water,whichis ama,it is clearlycognate
to the Northern lroquoian noun stem -hnek- 'water', also usedas the liquid
Othersare more erratic,suchas an initial -n- or -v- in a number
classifier.
of verbsimplying flexiblepatientsand an initial -t- or -y- in many verbs
implyinglong and rigid patients.Assignmentof ancestralforms to these
wouldbe speculative
at best.The neutralstems,thoseusedwhen
segments
patient
belongs
to
none
the
of
othercategories,
aretypicallytheshortest
the
descended
from verbswith no incorporatednouns.
forms,presumably
Although the classifiersare no longer formally transparent,cooccurrencerestrictionsbetweenthe once complexverbsand separatenominals
remainsemanticratherthan becomingarbitrarilylexical.King pointsout,
for example,that the animateform of verbsis usedwith live animals,but
the flexible form is usedfor deadones.Threadby itself is consideredflexible,
but on a spoolit is in the neutralcategory.He notesthat a very few items
are classifiedin seeminglyunexpectedways.The termsfor'hammer'and
as neutral,althougha non-speaker
mightexpectthem to
'ax', are classified
be viewedas long and rigid. For the mostpart, however,the old classifiers
continueto funciionsemantically,
evenwith new lexicalitems.
4.
Conclusion
The classificatory
systemswhichevolvefrom noun incorporationthus
fundamental
the
characteristics
of the morefamiliarsystemsin which
share
are separatewords.One of the moststrikingsimilaritiesinvolves
classifiers
of classifiers
foundin variouslanguages.
They mayrepresent
the repertoires
primarilykinds of entities,as in Cayugaand Ngandi,they may includea
mixtureof kinds and qualities,as in Mundurukriand Caddo,or they may
consistalmostentirelyof qualities,asin Cherokee.This arrrayis the result
incorporation,but it mirrorsthe
of the nornralevolutionof classificatory
arraysfoundin prototypicalsystems,
suchasthe Mandarin(Erbaugh),Jacal-
MARIANNE MITHUN
tec(Craig),Thai (Delancey),Burmese(Becker),andAustronesian
(Adams)
onesdiscussed
elsewherein thisvolume.
At first, speakersbeginby incorporatinggenerictermsfor the categories
theywishto recognize,thosewhichwill qualifyverbsin usefulways.Animate/
inanimate,concrete/abstract,
or human/animal
distinctions
arisein thisway,
asthrough the Cayugaclassifiers-ya't-'body',:nikqhr- 'mind', -rihw-'matter, idea,fact, etc.' and-nahskw-'domesticanimal'.
As timesgoesby, someof the incorporableclassifiersbecomein a sense
even more generic, implying only general qualities, rather than kinds of
entities. The result is a mixed systemwith some terms classifyingby kind,
others by quality. Munduruk( has such a system.It includesgenericterms
suchas road, name,village,and kin, alongsideof qualifyingtermssuchas
'round objects','long,rigid objects','long,flexibleobjects',and'liquids',
etc. Caddo,furtheralongin suchevolution,containscorrespondingly
more
classifiers
for qualities.Along with suchclassifiers
as'building'and'dirt'are
evenmoregene c terms,suchas'woodenobjects',andfinally,thosewhich
denote such qualitiesas'liquid', 'granularsubstance',and'small, round
otrjects.Tlrelexicalprocess
of incorporating
newnounstemsfor classificatory
purposesis just barelystill productivein the language.
Speakers
cantry new
onesout and discusstheir merits,but they are quite awarethat they are
beinghighlyinnovative.
With the development
of evernroregenericclassifiers,
theclassification
can beconlemore systematic.
More objectsqualify as either liquid, rigid,
flexible,or animate,for example,thanaskindsof baskets,
meats,or clothing.
Feweren{itiesare unclassifiable.
Whenincorporationlosesits productivity,
furthermore,the setof classifiersis no longeropen.New classifiersreflecting
kinds of entitesare no longeraddedto the system,and thosethat remain
are presumablyused lessfrequently than the quality classifiers.The result
is a tighterbut simplersystemof classification,
asin Cherokee,whereentities
irrc simplyclassifiedas animate,liquid, long and rigid, flexible,or noneof
these.This set of distinctionsis strikinglylike that found in a numberof
prototypical classifiersystems,a set particularlyuseful for qualifyingpredicates.
The useof largesetsof classifiers,
in bothprototypicalsystems
andthose
basedon incorporation,can be relalivelydifficult process.Erbaughnotes
that childrenacquiringMandarin generallyuseclassifierssomewhatlessthan
aduhs,just as thoselearningMohawk incorporateconsiderably
lessthan
adults. Becker, commentingon Burmese,and Delancy,on Thai, both
SYSTEMS
THE CONVERGENCEOF NOUN CLASSIFICATION
395
remarkedthat adults vary widely in their skill and virtuosity in the use of
classifiers.As with incorporation,admiredspeakerstend to use larger sets
morefrequently,whenthereisachoice,thanmoreordinaryormoremarginal
speakers.
A common function attributedto the prototypicalclassifiersis an
of suchanaphora,seeDowning,
anaphoricone. (For a detaileddiscussion
thisvolume.)They functionin placeof, or in conjunctionwith, pronouns.
While incorporatednounsare not strictly speakingreferential,and thuscannot be literallyanaphoric,they do serveto retainan entitywithin the arena
of discourse.They substitutefor an externalnominal in appropriatelyqualifyingtheir host verbs,oncethe entity in questionhasbeenappropriately
identifiedby someother linguisticor pragmaticmeans.
Incorporated classifiersarise from a very different sourcehistorically
whoseprimis a construction
thanthe prototypicalfreeones.Incorporation
aryfunctionis to qualifyverbsratherthannouns.Note,however,thatDenny
is to qualify
(thisvolume)that a fundamental
functionof classifiers
proposes
verbs,settingup expectationconcerningthe scopeof the v€rbs.The two
are, in fact, similarin beingderivedfrom nounswhose
typesof classifiers
as
nounshas been significantlyreduced,in the senseof
status
categorial
and
(1984).The factthatclassificatory
incorporation
Thompson
Hopperand
predication
of
events
The
systems
aresosimilaris no accident.
freeclassifier
andstatesis a basicfunction of language.The modificationof thesepredicato deal
of entitiestheyinvolvepermitsspeakers
tionsin termsof the classes
with largequantitiesof informationin efficientways.
NOTE
I am gratefulto WallaceChafefor a numberof helpfulcommentson an earlierdraft of lhis
paperand for sharinghis knowledgcof Caddo.I alsoappreciatethc help of ReginaldHenry. of
Six Nations, Ontario, for sharinghis expertiseon his language.Cayuga,Annetle Jacobs,of
Mohawk,andLeftyWilliams,
on herlanguage,
Caughnawaga,
Quebec,for sharingh€rexpertise
classifiersin his language,Caddo
of Anadarko, Oklahoma,for discussing
REFERENCES
in Austroasiatic."
Adams,Karen.Thisvolume."NumeralClassifiers
a
Makes:Describing
volume.
"The
Figure
a
Classifier
A.J.
This
Becker,
particularBurmeseClassifier."
396
MA R IA N N EMIT H UN
THE CONVERGENCEOF NOUN CLASSIFICATION
SYSTEMS
Chafe,Wallace,1977."Caddo texts." In DouglasR. parks, ed,.Caddoan
text.IIAL-NATSVoI 2.1. Chicago:Universityof Chicagopress.27-43.
Comodo,CristinaHelenaRohwedder1981.Concord.ancia
em Munduruku.
Universidade
Estadualde Campinas,Brazil. M.A. thesis.
Cornrie,Bernard1981.Thelanguages
ofthe SovietUnjon.Cambridge:
CambridgeUnivcrsityPress.
Craig,ColetteG. Thisvolume.,,Jacaltec
Noun Classifiers:
A Studyin Languageand Culture."
DeLancey,Scott.This volume..,Towarda I{istory of Tai Classifiers
Systcnrs,"
Downing, Pamela.This volume. ,,The Anaphoric Use of Classifiersin
Japanese."
Erbaugh,Mary S.Tlrisvolume.,,TakingStock:
The DevelopmentofChinese
Noun Classifiers
Hisroricallyand in young Children.,'
Fccfing.Durbin. 1975.Cherokee-English
Dictionary.Tahlequah:
Cherokee
Nationsof Oklahoma.
Heath, Jeffrey. 1978. Ngandi grammar, texts, and dictionary. Canberra:
AustralianInstituteof AboriginalStudies.
Llopper,Pauland SandraThompson.1984..,Thediscoursebasisof lexical
categories."Language 60.4.
King. Duane. 1975.A grammarand dictionaryof the Cherokeelanguage.
Ph.D. dissertation,
Universityof Georgia.UniversityMicrofilms.
Nlithun,Marianne.ms.a.The acquisitionof polysynthesis.
paperpresented
to the AmericanAnthropologicalAssociation,Washington,D.C. 19g2.
----. ms.b."The Evolutionof Noun Incorporation."paperpresented
to the
lntcrnationalCongressof HistoricalLinguists,poznan,poland19g3.
----. and ReginaldHenry. ms. "lncipientmoribundity:the obsolescence
of
OklahomaIroquois."Paperpresentedto the Congress
of Americanists.
Vancouver:1978.
Oates,LynetteFrances.1964.,,A tentativedescriptionof the Gunwinggu
language."OceanialinguisticMonograplzs10. Sydney:University ofSydney.
Osborne,C.R. 1974.The Tiwi language.
Canberra:Instituteof Aboriginal
Studies.
Parks,Douglas.1976.A grammarof Pawnee.New york: Garlandpress.
Pufte, Wilfiam. 1975."Outline of Cherokeegrammar.,'In Feeling1975:
235--154.
Sopir, Edward. 1911."The problem of noun incorporationin Americanlcn-
guages." AmericanA nthropologistn.s. I 3.250-82.
Sheffkr, Margaret.1978." Mund.urukudiscourse."Paperson discourse.ed.
by JosephE. Grimes,119-142.
Dallas:SummerInstituleof Linguistics.
. Caddoanle,r/s.New York: G.E. Stechert.
Weltfish, Gene. 193'7
397
© Copyright 2026 Paperzz