Anda di halaman 1dari 50

Korean and Japanese Morphology from a Lexical Perspective

Author(s): Peter Sells


Source: Linguistic Inquiry, Vol. 26, No. 2 (Spring, 1995), pp. 277-325
Published by: The MIT Press
Stable URL: http://www.jstor.org/stable/4178898 .
Accessed: 09/08/2011 00:52
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

The MIT Press is collaborating with JSTOR to digitize, preserve and extend access to Linguistic Inquiry.

http://www.jstor.org

Korean and Japanese


Morphology from a Lexical
Perspective
Peter Sells
I argue that the morphology of inflected verbs and nouns in
Korean and Japanese obeys principles and constraints that are
quite different from those that operate in the syntax. I claim
that head movement is not an appropriate mechanism either to
explain possible and impossible morphological combinations,
or to account for information flow within words. This view is
supported by facts of syntactic selection, which provides evidence against the postulation of functional categories such as
I(nfl) and C(omp). I present an alternative view of the morphology and its projection into the syntax founded on the principle
of lexical integrity and the idea of unification of information,
and a theory of syntactic formation in terms of the new notion
of "combinatoric TYPE."
Keywords: functional category, head movement, Japanese, Korean, lexical integrity/lexicalism, morphology, syntax, unification

1 Introduction
This article is concerned with the correct analysis of what can broadly be called the
"inflectionalmorphology"of two languages,Japaneseand Korean. I present data from
these two languageswhose morphologyand syntax show interestingparallels, partly in
continuationof the generative traditionof consideringthem similarin relevant aspects,
andpartlybecause argumentsfromdatain one languagewill often complementarguments
from data in the other.
Since the pioneering work of Kuroda (1965) and Kuno (1973) for Japanese, and
Yang (1972)for Korean, the syntactic propertiesof various inflectionalsuffixes in these
languageshave been studiedin some detail. Illustrativeexamplesof such suffixes (underlined)are given in (1). Throughoutthis articlea following "(J)" indicatesthatthe example
The commentsof fouranonymousreviewersgreatlyimprovedthe formand contentof this article.Earlier
versionswere presentedat the West Coast Conferenceon FormalLinguistics(WCCFL)10 meetingin Tempe,
Arizona,in March1991,and in April 1991at WayneState Universityandat the Universityof California,Santa
Cruz, where the audiences provided useful discussion. I have also benefited from the suggestions of Joan
Bresnan, Young-meeYu Cho, Hye-Won Choi, Eunjoo Han, Masayo lida, Jong-BokKim, Yookyung Kim,
Steven Lapointe, Yoshiko Matsumoto,and Shigeo Tonoike. All errorsand misinterpretationsare mine.

Linguistic Inquiry, Volume 26, Number 2, Spring 1995


277-325
? 1995 by The Massachusetts Institute of Technology

277

278

PETER

SELLS

is from Japanese, where this is not obvious from the context. All other examples are
Korean.
(1) a.

omotte i-mas-u. (J)

Taroo-wa [Hanako-ga tensai-da-to]

think
Taroo-ToP [Hanako-NOMgenius-coP.PRES-COMP]

PROG-LEVEL-PRES

'Taroo thinks that Hanako is a genius.'


po-si-ess-ta.
b. Sensayngnim-unpam-ulo-uy kin yelo-lul
teacher.HoN-ToP
night-to-GEN
long journey-Acc see-HoN-PAsT-DEcL
'The teacher saw A Long Day's Journey itnto Night.'
Moreover,despite all of the advances in the realmof syntax that have been achieved
in the past 25 years, the morphologicalstatus of the inflectionalsuffixes remainsa topic
of considerablecontroversy. That is to say, from a phonologicalview, they are clearly
bound, as suffixes or possibly clitics, whereas the results in the area of syntax largely
classify each suffix as a discrete element. Recently, and partlyacceleratedby the interest
in extendingX-bartheory to functionalcategories of varioustypes (Chomsky 1986,Pollock 1989), there have been several proposals within the Japanese/Koreanliteratureto
treat some, if not all, of these suffixes as phrasal heads. For instance, Gunji (1987)
proposed the existence of PP in Japanese for examples like (2a); in most recent work,
however, PP is reserved for phrases whose English translationis a PP.1 Some researchers, includingTonoike (1991) and Whitman(1989), have proposed the existence of DP
in both Japaneseand Korean (see the Korean (2b)). In the verbal domain, the existence
of functional categories such as I and C has never been questioned (one exception is
Lee 1990).2Additionally,some authorshave suggestedthat certainsuffixes are not heads
but specifiers (Jung 1991, Lee 1993, Tonoike 1991).
b.

PP

(2) a.

NP

DP

NP

un
Hanako
ga
sensayngnim
TOP
teacher.HON
Hanako
NOM
To the extent that the morphologicaland phonologicalconstituencyof the relevantforms
has been consideredat all in the studies mentionedabove, the proposalsof Baker (1988)
have been adopted. Baker showed that head movement is a way of combiningsyntactically discrete elements into a morphologicalunit, buildingon the familiarnotion of "affix
' Gunjigeneralizedto PPpartlyin recognitionof the fact thatthereis no syntacticevidencefor a distinction
between NP and PP (in Japanese).Here I take all to be NPs; see also footnote 22.
2 Otherworks involvingor presupposingfunctionalcategories include Ahn 1988, Ahn and Yoon 1989,
Choe 1987b,Ishii 1988,Kuroda1988, Park 1990, Suzuki 1989,Tateishi 1988, 1989,Toribio 1991,Urushibara
1991,Whitmanand Hahn 1988,Whitman,Lee, and Lust 1991,Yoon 1990, 1991b,and Yoon and Yoon 1990.

KOREAN

AND JAPANESE

FROM A LEXICAL

MORPHOLOGY

PERSPECTIVE

279

movement." Thus, in particularfor the verbal morphology of Japanese and Korean,


most of the work cited above assumes successive upwardhead movement to produce
single complex words.
For the purposes of comparison, (3) shows a simplified structurefor an English
embeddedclause using the functionalcategories C, T, and Neg, which are instantiated
by the words that, did, and,not, respectively (Chomsky 1986, Pollock 1989).
CP

(3)

TP

C
NP

T
T

NegP
Neg

VP

that

Mary

did

not

swim

At this simplifiedlevel of analysis, this is a faithfulrepresentationof the relevant structural properties. Now, the Japanese translationof this English sentence appears to be
somewhatdifferentin nature. Phonologically,there are only two words, as seen in (4).
(4) Mary-ga oyog-ana-katta-to. (J)
Mary-NOM

SWim-NEG-PAST-COMP

However, if we take each morphemeto represent a functional category, as it does in


the English sentence, and if we take the surface constituency of the morphemesto be
derived via movement (indicated by the arrows), we could give an analysis like that
shown in (5) for the verbal part of (4). In English each head precedes its complement;
in Japanesethe order is reversed.
3 Roughly speaking, CP (Comp Phrase)correspondsto the traditionalS' and TP to the traditionalS.
NegP, suggestedby Pollock (1989),is includedin the structuresfor comparison;in fact, it turnsout that there
is no evidencefor NegP in Japanese,andpositiveevidenceagainstit in Korean(argumentspresentedin section
5.1). There is also no independentevidence for an Agr(eement)Phrase in Japanese and Korean (also see
section5.1), so I have not includedthatin the structuresshown. Besides the functionalcategories,the inventory
of lexical (major)categoriesincludes N, V, and Adj (in Japanese).

280

SELLS

PETER

CP

(5)

TP
NP

T'
NegPT
VP

Neg

V
Mary-ga

oyog

-ana

-katta

-to

Whatis appealingabout (5) is that it displays a close, and potentiallyhighly interesting,


parallelof structurebetween the two languages:apartfrom the position of the subject
NP, the Japaneseexample is given a structurethat is the mirrorimage of the one given
for English. All of the labels are the same, and all of the hierarchicalrelationshipsare
the same. Thus, there is the potential for expressing and explaining deep similarities
between the two languages.
After head movement applies, successively, the structureunder the node C in (5)
would be that shown in (6).
C

(6)

T
T

Neg
V

Neg

oyog

ana

katta

to

KOREAN

AND JAPANESE

MORPHOLOGY

FROM A LEXICAL

PERSPECTIVE

281

I will refer to the set of theoretical assumptions that underlie an analysis like that in
(5)-(6) as the syntactic view (a syntactic view of the inflectional morphologyof these
languages).My purposein this articleis to arguethatthe syntacticview cannotbe correct
(for the languages under consideration)and that the syntax of (4) is more accurately
representedby the simple structurein (7). More articulatedproposals for the syntactic
and morphologicalstructuresare given in section 5.
S

(7)
NP

VP

Mary-ga

oyog-ana-katta-to

In other words, the inflectional suffixes are all attached in the lexicon, and they have
no syntactic status, other than whatever features they contribute to the overall word
containingthem.4I will referto this as the lexical view. As will become clear, the "syntax
vs. lexicon" part of the argumentationis in fact subordinateto a more importantclaim
that I wish to make: namely, that all of the relevant inflectional structures in these
languagesare not right-headed,no matterwhere in the grammarthey are actuallyformed.
My argumentsfor the particularlexical view fall into two broadclasses: those that argue
that(5) could not possibly be a well-formedunderlyingstructure(D-Structurerepresentation), and those that arguethat movementis not the appropriatemechanismfor morphological combination.
The organizationof this article is as follows. In section 2 I present some of the
evidence that the inflectionalsuffixes combine morphologicallywith their hosts. In section 3 I show that these morphologicalformations do not have the internal structures
that a syntactic derivationwould lead one to expect. In section 4 I discuss three kinds
of problemsthat arise as a result of the attemptto derive words by movement. In section
5 I outlinea lexical analysis of the inflectionalmorphology,factoringapartdifferentkinds
of informationthat are confoundedunder the syntactic analysis.
2 MorphologicalCombination
In this section I survey some of the evidence for the claimthatthe inflectionalmorphemes
must actually be combined with their hosts. The data in this section, all from Korean,
are selected from those presented in Cho and Sells 1994, though I use them to make a
4Such proposalshave been made for Koreanby Cho (1988), Lapointe(1991), and Park(1988).

282

PETER

SELLS

slightly differentpoint here. The data show that a structuresuch as (5) is itself not an
adequate S-Structurerepresentationunless the movement indicated takes place-that
is, they show that head movement is a necessary part of the syntactic view.
2.1 KoreanPalatalization
Korean has a rule that palatalizes a dental consonant when it precedes the high front
vowel i. However, the rule is restricted:it does not apply in underived, compound, or
phrasalenvironments.The examples in (8) show underivedwords; the consonantvoices
by a regular rule, but does not palatalize. Regular orthographicforms are shown in
parentheses.
(8) a.

[ati] (eti)

o> [di]

(*3ci)

'where'
b.

[canti] (canti)

[candi] (*canci)

'grass'
In cases where the environmentis formed by morphologicalsuffixation, the rule must
apply, as in (9).
(9) a.
b.
c.

[kath+ i] (kath+ i)
same + ADV
[path+ i] (path+ i)

> [kachi]
'together'
[paChi]

field-NOM

'field'

ta] (path+ i + ta)


field-coP-DEcL

ida]
[paCh
'be a field'

[path +i +

However, the palatalizationrule does not apply between parts of a compound, or between an object and a verb.
(10) a.

[path+ ilai] (path + ilang)

->[pad

irag] (*[pajirag]) (Compounding)

field ridge
'the ridge of a field'
b. [pathilkuko](path ilkwuko) -> [pad ilgugo] (*[pachilgugo]) (VP)
field till
'till the field'
In other words, the rule applies only in a restricted domain: between a root and any
inflectionalsuffixes, but not in an underivedword; between two roots in a compound;
or between two words in the syntax. The domain of the rule is then essentially those
complexes formed by supposing head movement from an underlyingstructure, where
each morphemeis separate, and where there is one lexical category and a (potential)
series of functionalcategories.'
s The discussionhere is not quite precise in the sense that derivationalsuffixes, which I do not consider
in this article, behave like inflectionalsuffixes and fall in the domainof the palatalizationrule.

KOREAN

AND JAPANESE

MORPHOLOGY

FROM A LEXICAL

PERSPECTIVE

283

2.2 Verbal Suffixation

As discussed in detail by Cho (1991),it is possible to attach suffixes such as -man 'only',
-(n)un(focus), and -to 'even' to tenseless verb forms. However, for many Koreanspeakers there is a phonologicalcondition that must be met in such cases: the host verb form
must be at least disyllabic.6
As shown in (1la), a tenseless verb is formed by suffixing onto the verb root a
morphemethat has the properties of a type of complementizer(glossed as COMP;see
section 4.2 for a fuller discussion of their properties). For certain verbs the suffixation
of the complementizerfails to add another syllable, as indicatedin (1ib) and (1ic), and
these forms resist furthersuffixation,for they remainmonosyllabic.
(11) a.

mek-e-man

cwu-sey-yo

eat-coMP-only give-HON-LEVEL

'give the favor (only) of eating'


b. *hay-man
(< ha-e-man;
cwu-sey-yo
do-coMP-only give-HON-LEVEL
orthographic /ay! =
'give the favor (only) of doing something'
c. *ka-man
(< ka-a-man)
cwu-sey-yo

[ae])

go-COMP-Only give-HON-LEVEL

'give the favor (only) of going'

Many putatively "incorporated"noun-verb constructions show similar behavior. Although both kongpwu hata 'study' and kwen-hata 'encourage' consist of a verbal noun

in close connectionwith the verb hata 'do', only the latterallows suffixationof a particle
such as -man.
(12) a. *kongpwu hay-man
(< ha-e-man)
cwu-sey-yo
study
do-coMP-only give-HON-LEVEL
'give the favor (only) of studying'
b. kwen-hay-man
(< ha-e-man)
cwu-sey-yo
encourage-do-coMP-only

give-HON-LEVEL

'give the favor (only) of encouraging'


The phonological evidence indicates that the predicate in (12a) is not formed lexically,
as argued on independent grounds by Sells and Cho (1991), and by Poser (1991) for the
parallel Japanese cases like benkyoo suru 'study'.
The fact that (12a) is ungrammatical is problematic for an incorporation analysis of
such verbal-noun-plus-verb structures: if incorporation-that is, head movement-produces structures that are then input to the lexical phonology, (12a) should be good on a
6 For some speakers, the conditions on particleattachmentare differentfrom
those reportedby Cho:
some have no restrictionseven with monosyllabichosts, yet find that certain longer forms do not accept
particles.For such speakers,the particularargumenthere cannot be made.

284

PETER

SELLS

par with (13b), for kongpwu-ha-is polysyllabic; alternatively,if head movement has no
input to lexical phonology, then (1la) is unexpectedly good-for -man is attachingjust
to the COMP-e.
It is easily demonstrablethat the COMPalone does not count as the host of the
suffixingparticle. The contrast seen in (13) shows that -man is attached to the result of
combiningthe COMP with the verbalroot; it is only when the result of combiningthe root
and the COMPis monosyllabic(as from the underlyingha-e-man) that furthersuffixation
is blocked. With a differentCOMP,-key, there is no unusualphonology and the surface
form ha-key can host -man. Independentlywe know that -key-manshould be locally ill
formed, since -key is only monosyllabic;thus, -man must attach to the surface form of
the verb+ COMP.
(13) a. *hay-man (< ha-e-man)
do-coMP

b. ha-key-man
do-coMP-only

Finally, and most importantly,these facts indicate that the word-internalphonology is


cyclic-if the ill-formednessof *hay-man is to be explained, the host of -man must
actually be hay-, not *ha-e-. This means that, having formed units via head movement
in the syntax, the syntacticview mustthen allow each terminalto be inputto phonological
rules that apply cyclically (see, e.g., Halle 1990). That is, the derived structureafter
head movement must be (14) (cf. (6)); the natureof the category X will be discussed in
section 4.3.

(14)
X
X

man
e

ha
Thus, in order to account for the phonologicalfacts presented here, it seems that the
syntactic view must assume that the root and its dependent suffixes are brought into
combinationvia head movement. This will become importantbelow.
3 Properties of the Morphological Structures

Thereare manycases where particlesintervenebetween two heads that show selectional


or other head-governedproperties.These, I will argue, are structureswhose head is not
on the right. To begin, I will present the problemin a slightly more general way.

KOREAN AND JAPANESE MORPHOLOGY FROM A LEXICAL PERSPECTIVE

285

3.1 General Considerations

Consider an example like (15). Here, it is a lexical property of the verb cwu-ta 'give'
that it determinesdative case on the goal argument,expressed by -hanthey. However,
a delimitingparticle-kkaciand the focus marker-nun intervene. If these are heads that
projectphrases, then cwu-ta is separatedfrom its dative argumentby two other phrasal
projections.
(15) Swuni-hanthey-kkaci-nun
cwu-ess-ta.
Sooni-DAT-even-FOC

give-PAST-DECL

'I gave it even to Sooni.'


Such situationsclearly violate the idea that such processes as selection and subcategorization are local, an idea that is generallywell supported,but quite systematicallyviolated
by many currentproposals to expand the numberof functionalprojections(as noted by
Ernst (1992), Grimshaw(1991)).
Noting that the same situationholds in English(as in the translationof (15)), Tonoike
(1991) proposes that words like even in English occupy a position that is the specifier
of the relevant phrase (PP in this case). This allows the locality of selection to be preserved, since even to Sooni is a PP headed by the P to. For Japaneseexamples similar
to (15), he proposes that each phrase has a specifier generated on the right (thus, for
him, Japanese is a true mirrorimage of English). Under his account, -hanthey in (15)
would be the head of its own functional projection, -kkaci would be its specifier, and
-nun would be the head of a projectionof the category D (Determiner).7
(16) [[[SwuniNP] -hantheyp-kkaci pp] -nunD

DP]

However, this analysis does not really help for the case in point: the intuitionis that
cwu-ta 'give' selects for -hanthey, and yet the sister of cwu-ta is a DP headed by the
determiner-nun. On the other hand, if the structureof (16) were not right-headedat
each level, no such violations of locality would arise; the relevantpropertiesof -hanthey
could be inheriteddirectly throughthe nonheads -kkaci and -nun.
Moreover, the view that some suffixes are specifiers creates a technical problem
with the syntax-phonologyinterface.If -kkaciin (16) is a specifier,then it cannotcombine
with the other morphemesby head movement, since movementis taken to be disallowed
from a head position to a specifier position (Chomsky 1986).8
Movingto anotherexample, both Japaneseand Koreanhave productiveformations
7 Actually, Tonoike proposes IP where I have PP.
8 Thereis a responsethat could be made to this argument.Any view that specifiersare uniformlyon the
rightin these languagesmust necessarilyentail that there are no XP specifiers, only X? ones, for XPs never
follow the head in these languages.Thus, it could be claimed that heads and specifierscan combineby X?movement,sincebothtypes arein fact X?s.It shouldalso be noted, however,thathavingexclusivelynonphrasal
specifiersis a rathernovel positionand wouldhave to be stipulatedas a point of divergencebetweenJapanese
or Korean, on the one hand, and a languagelike English, on the other. Moreover, the motivationfor the
distinctionbetween head and specifieris greatlyreducedif all are X?s.

286

PETER

SELLS

of the light verb 'do' with a verbal noun, as shown in (17) (and as mentionedin section
2.2).

benkyoo suru (J)


study do
'study'
b. hakken suru (J)
discovery do
'discover'
c. phantan hata
judgmentdo
'judge'
d. wuncen hata
drive do
'drive'

(17) a.

It is temptingto analyze these as involving incorporationin the sense of Baker (1988),


but this cannot be correct, as arguedat length by Poser (1991), and as shown furtherby
the data in (12). Instead of incorporation,these examples involve a syntactic formation
of N and V (see Sells and Cho 1991)in the phrase structure,and a sharingof argument
structure(Sells 1991);the details of that process are not importanthere. The relevance
of such examples to the argumenthere is that various particles may intervene between
the verbal noun and the verb, as shown in (18).
(18) a.

benkyoo-sae suru (J)


'even study'
b. hakken-wasuru (J)
'discover (focus)'
c. phantan-tohata
'even judge'
d. wuncen-manhata
'only drive'

The mere possibility of such particleswith semantic and pragmaticforce casts doubt on
the incorporationidea. If instead of being heads the particles are specifiers, then the
incorporatedunit in each case would have to be NP, not N?, but general principlesof
the theory do not allow this. This leaves no viable incorporationanalysis. On the other
hand, if the particles in (18) are simply nonhead inflectionalsuffixes, then the presence
or absence of such a suffix becomes essentially irrelevant,which is correct.
3.2 Arguments against Right-Headed Morphological Structures

Here, I will presentthree directargumentsagainstthe idea that the morphologyis formed


out of right-headedstructures,which augmentthe generalconsiderationsjust presented.

KOREAN AND JAPANESE MORPHOLOGY FROM A LEXICAL PERSPECTIVE

287

Korean has many complex predicate constructionsin which one verb selects for a
particularform of another verb. I refer to the suffixes that comprise these particular
forms (underlinedin (19)) as complementizers.9In each case the example is acceptable
only if the particularsuffix shown is present.
(19) a.

b.

ssu-ci-man-un

anh-ass-ta

write-coMP2-only-FoC

not-PAST-DECL

'does not only write (focus)'


po-ass-ta
ilk-e-man
read-coMPl-only

try-PAST-DECL

'tries only reading'

c.

ka-ko-man-un

siph-ta

desire-DECL
go-COMP2-only-FOc

'wants only to go (focus)'


In an example like (19a) it is unclear how this selection could work if either -man or
-un or both head independentfunctional projections.10Clearly, the informationabout
the particular form of the COMPmust be specified on the part ssu-ci; however this is
done, it is not obvious how that informationwould be transmittedto higherlevels from
the nonhead (by hypothesis, -man and -un would be heads).1'
Another pertinent argument here comes from Japanese gerunds. Certain predicates

in Japanese select for a preceding verb in the form of a gerund. In certain cases the
gerunditself may be verbal or adjectival;however, only the verbalones are selected by
following main verbs. (20) shows the two possible negative counterpartsof the gerund
tabete 'eating'.
(20) a.

tabe-te
eat-GER

b.

tabe-nai-de

(V)

eat-NEG-COP.GER

c.

tabe-naku-te (Adj)
eat-NEG-GER

As (21b) shows, only the verbal gerund is selected by the higher verb, oita in this case,
which literally means 'put'; in this construction it means to do something in preparation
for some future eventuality.
(21) a.

oi-ta.
Ziroo-wa zenbu-no tabemono-o tabe-nai-dev
eat-NEG-COP.GER 'put -PAST
Ziroo-ToP all-GEN food-ACC
'Ziroo made the provision of not eating all the food.'

I The classificationcoMPi or CoMP2


is that of Cho and Sells (1994)and reflects the morphological"slot"
where the suffix appearsin the verbal morphology(see section 4.2).
10 I treat the form anh- as a verb, for it patternswith the other uncontroversialverbs in (19) (see Sells
1991and Yoon 1993);also see section 5.1.
'" Choi (1991)presents a similarargumentfor left-headedX? structures.

288

PETER

SELLS

oi-ta.
b. *Ziroo-wa zenbu-notabemono-otabe-naku-teAdj
eat-NEG-GER put'-PAST
Ziroo-ToPall-GEN food-Acc
'Ziroo made the provision of not eating all the food.'
The attachmentof particles such as -wa or -mo 'also' does not alter these selectional
facts: in other words, the category is preserved throughadditionof a particle, such as
-wa or -mo.12
ok-ana-katta.
Ziroo-wa zenbu-notabemono-otabe-nai-dev-wa
eat-NEG-COP.GER-TOP 'put'-NEG-PAST
Ziroo-TOP
all-GEN food-Acc
'Ziroo didn't make the provision of not eating all the food.'
b. *Ziroo-wa zenbu-notabemono-otabe-naku-teAdi-wa
ok-ana-katta.
eat-NEG-GER-TOP 'put'-NEG-PAST
Ziroo-ToPall-GEN food-Acc
'Ziroo didn't make the provision of not eating all the food.'

(22) a.

Finally, similarargumentscan be made for the nonexistence of T (or I) as a separate


head. In Japanese, verbs and adjectives have present or past tense forms, as shown in
(23).
(23) a.
b.

tabe-ru

tabe-ta

eat-PRES

eat-PAST

aka-i

aka-katta

red-PRES

red-PAST

(V)
(Adj)

Under a right-headedaccount, all of these forms would be of category T, after the stem
moves to combine with the tense morpheme. However, this renders such an account
unable to make a necessary distinction:adjectives, but not verbs, take the suffix -desu
to indicateformalspeech level. For example, -desu can be added to the past form of an
adjective, to make the speech level formal, but this is not possible for a verb.
(24) a. *tabe-ta-desu (V) (cf. tabemasita)
eat-PAST-LEVEL

b.

aka-katta-desu (Adj)
red-PAST-LEVEL

Thus, it seems that tabe-ta is a verb because tabe is, and that aka-kattais an adjective
because aka is. If these structuresare right-headed,then the only way to account for
the contrastin the examples in (24) would be to require(correctly)that -desu selects for
a precedingAdj, but also it would be necessary to categorize -ta as a V and -katta as
an Adj, these being the supposed heads of the structures.In other words, neithercould
be T (or any other functionalcategory).'3
12

Such argumentscannot be made in Korean since there is no formalsyntactic distinctionbetween V


and Adj; there arejust nonstativeand stative predicates,respectively, all of categoryV.
13 This kind of analysis is essentially that presentedby Grimshaw
(1991), discussed below.

KOREAN

AND JAPANESE

MORPHOLOGY

FROM A LEXICAL

PERSPECTIVE

289

3.3 Summary
As we have seen, there are many cases where local processes are interruptedby suffixal
elements in Japaneseand Korean. If we view these suffixes as inflectionalelements that
do not affect the majorpropertiesof their host (the part to the left of the suffix), all of
these cases are unproblematic.However, if we view each suffix as a (functional)head
projectingan independentphrase, the problems are severe.
As observedby Ernst(1992),there does not seem to be any principlecurrentlytaken
to be part of the grammarthat forces it to be the case that each functionalcategory is
a head that projects a phrase, though certainly almost all currentresearch assumes this
to be true (especially in putatively uniformlyright-headedlanguageslike Japanese and
Korean);and of course, without such a principle, which ones project and which ones
do not would have to be stipulated,as would the X-bar-theoreticrole of the latter class,
for they would not be heads, complements, or specifiers.14
Even thoughthere are several very strongargumentsthat the structuresin question
are not right-headed,which might seem to be immediately at odds with most of the
publishedliteratureespousing the syntactic view of affixation, there is no fundamental
incompatibility.It mightbe possible to treateach affix in the syntax as a nonhead,thereby
preservingthe requiredpropertiesof the (original)head in the structure.Grimshaw(1991)
has formulatedsuch a proposal, which allows majorcategory projectionsto be inherited
throughfunctionalprojections(called "extended projections"), in response to problems
similarto those raised in this section. Adopted in the analysis of Japanese and Korean,
this amountsto saying that the relevant structuresare left-headed.AlthoughGrimshaw
does not address data of the kind presented here, the structureswould have to be leftheaded in order to allow for any kind of idiosyncraticinformation(such as the form of
the COMP in the examples in (19)) to be inheritedfrom the left daughter.On one level,
the lexical view and the extended projectionsview are close in spirit, but they are not
notationalvariants, since one is committed to the idea that the relevant combinations
occur in the lexicon, and the other, to the idea that they occur in the syntax."5
This said, it becomes clear that the argumentI wish to make in this article has two
parts: first, that there are structuresthat are not right-headed,and second, that those
structuresare formed in the lexicon. I have addressed the first part in this section, and
I will take up the second in the next. Even adoptingsomethinglike Grimshaw'sextended
projections, the syntactic view still requires that the various morphemesbe combined
by X?-movement.As noted above, the differencebetween this view and the lexical view
is then the nature of informationflow in the morphology-on the syntactic view, XO14 Choi (1991) presents an account of certain Korean cases that have this
property,and extends X-bar
theory to accountfor them.
15 A reviewernotes that a simplefeaturepercolationmechanism,such as that proposedby Lieber(1980),
Selkirk(1982),or Williams(1981),would allow nonconflictinginformationto be passed up from all daughters
to the mother.I take Grimshaw'sproposalto be a particularand articulatedinstantiationof this view and thus
have addressedthat.

290

PETER

SELLS

movementis the underlyingforce. In the following section I present the problemsfacing


an account that takes movement as the basis of morphologicalformation,and I contrast
those properties of a lexical account that render it immune from such problems. The
lexical account is presented in more detail in section 5.
4 FunctionalCategories,Movement,and Selection
The fundamentalproblemwith views of structurethat posit several functionalcategories
is that there is no notion of "size" available, a notion that will appear as a recurrent
theme in the rest of this section; for now, I show in (25) a sample structureof the verbal
partof a Koreansentence, on the assumptionthat there are variousfunctionalcategories
such as the ones shown. An illustrativeexample is given beneaththe tree. The treatment
of agreementand negationinvolves some extra complicationsthat go beyond the simple
syntactic/lexicaldistinction;this issue is taken up at the beginningof section 5.
(25)

CP
MoodP
TP

Mood

NegP
AgrP
VP

T
Neg

Agr

ilk
-usi
-ci
read
-HON
-CoMP2
'that (someone (hon.)) did not read'

anh

-ass

-ta

-ko

NEG

-PAST

-DECL

-CoMP4

This structureis typical of those proposedin the relevantliteraturearguingfor or presupposing the syntactic view (e.g., (for differentparts) Ahn and Yoon 1989,Jung 1991, Lee
1993,Whitman1989).The head Neg is responsiblefor the negativemorphemeanh, which
cooccurs with the precedingCOMP -ci.
As noted above, analyses that posit functional categories as heads corresponding
to each morphemeappeal to (head) movement in order to account for the selectional
propertiesof each higherhead. For example, the Korean verb po-ta selects the form of

KOREAN

AND JAPANESE

MORPHOLOGY

FROM A LEXICAL

PERSPECTIVE

291

the coMPl in the precedingverb, as in (26); without the underlinedparticle (CoMPl) -e,
the example is ungrammatical.
(26) mek-e-man
eat-coMPl-only

po-ass-ta
try-PAST-DECL

'tried (only) eating'


Here, it is assumed that mek moves to -e, and then that composite element moves to
-man, as in

(5).16

This is essentially how the syntactic view works: the sister of po-ta is not headed
by somethingcontainingthe element that it selects for, but aftermovementthis condition
is satisfied.

Additionally,the syntacticview necessitates a duplicationof selectionalinformation.


As noted by Speas (1991),the syntacticview of inflectionalaffixationmust be augmented
by a theory of morphologicalsubcategorizationin order to make any predictionsat all
about the relationshipsbetween syntax and morphology(for Korean, see Yoon 1991a).
This means that a functionalhead must have two subcategorizationframes-one for its
complementin the syntax (say, a VP), and one for its "host" in the morphology(say,
a Vroot)-and, accordingto Speas, they must be "consistent" with each other, meaning
that the directionof subcategorizationmust be the same in each case. I take this issue
up again in section 4.3. Under the lexical view, no such duplicationis necessary, for it
is not even possible.17
Withthis generalbackground,we can look at several specific problemsthat a syntactic view must face; these considerationsconstitute the bulk of my argumentsagainst a
syntactic account of the morphologyof Korean and Japanese.
4.1 Problems of Underlying Structure

The first class of problemsconcerns the underlyingstructuresthat must be presentunder


the syntactic view. These are, ultimately, problems of transformation,since they arise
underthe assumptionthat transformationaloperations such as movement and spell-out
of irregularforms constitute the mode of combinationof morphologicalinformation.
First, there is a problemof doubly markedinformation.The idea of projectingmorphologicalinformationas functionalheads seems to be motivatedin partby the intuition
that each piece of information(such as tense or mood) should receive a discrete and
16

If -man lowers, the resultis the same;in the referencescited, the assumption(in some cases, a necessary
assumption)is that movementis upward.
17
A reviewer suggests that undera "checkingtheory" approachto morphology,morphemesneed only
be specifiedas boundor free, withthe syntaxconstrainingtheirappearanceandrelativeorder,throughrequirementssuchas conditionson licensingandthe HeadMovementConstraint.Thus, the morphologicalsubcategorization is eliminatedin favor of syntactic conditions, and the duplicationI refer to in this section does not
arise. However,as I discuss in section4.3, it seems thatit is the syntacticconditionsthat shouldbe eliminated,
in favor of the (possibly idiosyncratic)morphologicalsubcategorizations.

292

PETER

SELLS

unique structuralrepresentation.Such a view would lead one to expect that each piece
of informationshould correspond to a morpheme in the structure, in the absence of
suppletive forms or zero morphemes(see, e.g., Speas 1991). But what is problematic
for this view is the fact that informationmay be doubly markedin certain cases.
In Japanese one productive way of makingthe subject honorific form of a verb is
to put the verb stem in the context o-V-ni naru, for example, o-tabe-ninaru 'he (honorable) eats'. A few verbs have irregularsubject honorificforms; one of these is suru 'do',
for which the expected *o-si-ninaruis impossible, and the correctform is instead nasaru
'he (honorable)does'. A syntactic view of honorificationwould presumablytake o-V-ni
naruto be the realizationof some node in the tree expressinghonorificationinformation
(possibly related to Agr or T, since it is subject honorification);an irregularform like
nasaru would be derived by a special rule from the ungrammaticalsource *o-si-ninaru.
Now, as observed by Sells and lida (1991), honorificationmay be doubly marked
with irregularforms: that is, o-nasari-ni naru is acceptable, if a little redundant.The
facts are summarizedin (27).
(27) a. Japaneseproductive honorific:o-V-ni naru
b. Irregularform: *o-si-ni naru -> nasaru 'honorablydo'
c. Double markingis possible: o-nasari-ninaru
Under the syntactic view, there is no source for the form in (27c), since by hypothesis
nasari comes from what would be the underlyingform, namely, o-si-ni naru, as in (27b).
Clearly, what is happeningin this case is that the root of the irregularform is listed in
the lexicon and can be put in the honorificformjust like any other root. However, this
is inconsistent with (the strongest version of) what drives the syntactic view, namely,
that each relevant piece of morphologicalinformationshould be present as a discrete
node in the underlyingstructure.18
Similarly,as observed by Han (1991)and Martin(1992), suppletivehonorificforms
in Koreancannot simplybe thoughtof as "abbreviations"for largerpieces of structure.
Taking capswusi- 'hon. eat' as an example, Han observes that this irregularhonorific
18 For Korean,Choe (1993)suggeststhat honorificationis a realizationof Agr and that all Korean
clauses,
even those with intransitiveverbs, allow both Agrs and Agro to be present (followingChomsky (1991)on
Englishand French). Further,Choe suggests that the two Agrs can be linked or coindexed, so that double
instancesof subjecthonorificationmay be generated;if appliedto Japanese,this would account for (27c).
This alternativeanalysisseems to me to be problematicin that it requiresa differentinterpretationof Agr
from that standardlyassumed. Althougheach Agr is not inherentlyrelatedto any NP in the structure,this
proposalwouldrequirethatAgrohave no "object-related"propertieseven witha transitiveverb-for example,
it wouldrequirethatobjectscould not be licensed and/orCase-markedin the specifierpositionof Agro, since,
by assumption,Agro is in an agreementrelationshipwith the subject. However, there are Koreanforms that
are transitiveand that are (consideredby Choe to be) doubly marked, such as capswu-si-ta '(hon.) eats';
similarly,nasaru in (27c) is transitive.In other words, the only apparentfunction of this Agro is to account
for the double honorificmarking.
It is perhapsworthnotingthat although"object" honorificationalso exists in these languages,it probably
should not be taken as a reflex of any Agr. The phenomenonis much more widespreadin Japanesethan in
Koreanand is not restrictedto objects at all; rather,it indicatesa humblingrelationshipdirectedtowardany
nonsubjectargumentor adjunctand is better characterizedas "nonsubjecthonorification"(see Kuno 1983,
Matsumoto1993).Therefore,it is doubtfulthat nonsubjecthonorificationin Japaneseis relatedto Agro at all.

KOREAN

AND JAPANESE

MORPHOLOGY

FROM A LEXICAL

PERSPECTIVE

293

form can be followed by the COMPl -el-a, as in (28b), a sequence that is not possible for
any regularhonorificform, such as ilk-usi- 'hon. read', as in (28c).
(28) a.

Sensayngnim-kkeyse capswusi-ess-ta.
teacher-HoN.SUBJ

b.

(eat-HoN-PAST-DECL)

'The teacher ate.'


Sensayngnim-kkeyse capswusi-e

po-si-ess-ta.

teacher-HoN.SUBJ

try-HON-PAST-DECL

eat.HON-COMPI

'The teacher tried eating.'


c. *Sensayngnim-kkeyse ilk-usi-e
teacher-HoN.SUBJ

d.

(*mek-usi-ess-ta)

eat.HON-PAST-DECL

read-HON-COMPI

'The teacher tried reading.'


sensayngnim-kkeyse ilk-usi-key
teacher-HoN.SUBJ

read-HON-COMP

po-si-ess-ta.
try-HON-PAST-DECL

hay-ss-ta
make-PAST-DECL

'. . . made the teacher read . . .'


The standardand uncontroversialaccount of the crucialexample (28c) is that coMPi and
the honorific suffix compete for the same slot, so only one of them is possible. (The
example in (28d) indicates that ilk-usi-does allow other suffixes.) In turn, the contrast
between (28b) and (28c) shows that capswusi- cannot be irregularly"spelled out" from
an underlying*mek-usi-'hon. eat', because there would be no source for the COMPl -e
in (28b) in that case (-usi would be in its "slot").`9
Thus, these problemsall point to the same conclusion: that irregularforms must be
listed amongthe items insertedinto the underlyingstructure,even if they are informationally complex. However, this severely weakens the syntactic view, since it is predicated
on the assumption that the underlying structures for similar examples should not be
differentdependingon the presence of any irregularforms.
In an approachthat generatesfully inflected words independentlyof the syntax and
then uses head movementfor feature "checking," as in Chomsky 1993, these problems
of multiplemarkingshould not arise. However, such an approachmust essentially adopt
all of the lexical analysis (independentprinciples for the structure of words and for
passing informationaround within those structures)and must thereby also relinquish
one of the motivatingfactorsof the syntacticaccount, thatword structurecan supposedly
be determinedby syntactic principles (for recent views on either side of this question,
see Lieber 1992and Anderson 1992).However, the idea that movementcan account for
even the informationalcontent of complex words is not without problems, as discussed
in the following sections.
I turnnow to anotherexamplewhere the predictionsof a syntacticallydrivenaccount
of morphologyfail. Korean has two ways of markingnominative case: the honorific
19 Two reviewershave noted that forms like (28c) and (31b)are perhapsacceptablein colloquialKorean
(thoughnone of my consultantsaccepts them). For such speakersin such registers, my particularargument
cannot be made.

294

PETER

SELLS

subject marker-kkeyse and the regularnominative marker-il-ka. Both appear in (29).


The bracketed noun root is itself internally complex, but that is not relevant here; it
simply makes the example more natural.20
(29) [Sensayng-nim-tul]-kkeyse-man-i kulen
il-ul
that.kindwork-Acc
[teacher-HON-PL]-HoN.suBJ-only-NoM
ha-si-pnita.
do-HoN-LEvEL.Formal.DECL

'Only teachers do such work.'


What is interestingabout -kkeyse is that it does not pattern morphologicallywith the
(structural)case markers;instead, it falls in the same morphologicalslot as the Postpositions, comingrightafter the noun root. This particularform is completely without explanation under the syntactic view of morphology. Except for its morphologicaldistribution, -kkeysehas no syntactic or semantic postpositionalproperties:it has no spatial or
temporalmeaning, and it can only appear on surface (nominative)subjects. For these
reasons, it could not be identified with a markerof "inherent" case (as suggested in
Yoon 1991a);there is no constant thematic content associated with it, and it can mark
even raised subjects. It is in fact even more of a "grammatical"markerthan the regular
nominativemarker-il-ka, which can mark nominativesubjects, objects, or adjuncts:in
contrast, -kkeysemarksonly (honorable)nominativesubjects, even thoughit appearsas
the closest in of all the possible nominal suffixes.
The syntactic view has no way of accounting for the acceptability of an example
like (29). If nominativecase is assigned to previouslygeneratedphrases, then it is totally
unexpectedthat it would appearinternalto the constituentin the position of -kkeyse, as
well as at the end in the form of the regularnominativemarker.If we assume that case
is base-generated,then it is not possible for -kkeyse,which is not the head of the subject
phrase,to specify the syntacticconfigurationin which it mustappear(i.e., as a subject).21
20

I assumethat the morphemes-nim and -tul are attachedinside the morphologicalconstituentI referto
as the Nroot; in principle,they can appearinside all of the nominalsuffixes discussed in this article.
21 The distributionof -kkeyse cannot be directly linked to the presence or position of nominativecase,
since Koreanallows nominativeobjects in certain circumstances,but -kkeyseis restrictedto subjects. I do
not thinkthat it is possible to enforce the fact that -kkeysemust be a subjectby somehow requiringit to have
an agreeingverb-that is, to tradeon the idea that honorificagreementis with a subject and on the fact that
-kkeyseis an honorificmarker.Thispresumed"agreement"is not syntacticallynecessaryin the samesense that
morefamiliaragreementis: exampleswith-kkeysesubjectsandnonhonorificverbformsmightbe pragmatically
marked,but they are not strictlyungrammatical
(forinstance,(29)withoutthe -si suffixon the verb). For more
discussion,see Han 1991andPark1991,wherethe natureof honorific"agreement"is explicitlycontrastedwith
true subject-verbagreementin Frenchand German.
Also, certainpossessors inside the subjectmay triggerhonorification:for instance, in Korean,examples
like (i) are acceptable(Hong 1991:12);for more extreme examples, see Yun 1991.
(i) Sensayng-nim-uyson-i
khu-si-ta.
teacher-HON-GEN

handS-NOM

big-HoN-DECL

'The teacher's hands are big (hon.).'


Japanese allows similar examples (Harada 1976). I am not aware of any true syntactic agreement process that
looks into nonhead constituents like this.

KOREAN

AND JAPANESE

MORPHOLOGY

FROM A LEXICAL

PERSPECTIVE

295

The upshot of this for the syntactic view is that -kkeyse looks like it heads a PP,
but the grammaticalinformationit conveys is of the type associated with NP (and not
with other PPs in the language).Even grantingthat such a categorialdistinctionexists,22
there is no way to reconcile these two conflicting categorial properties. As with the
Japanesehonorificverb forms above, the irregularform (-kkeyse)can cooccur with the
regularform of the nominativecase marker,since the informationthe two provide does
not conflict, and a well-formedstructurecan accommodateboth.23On a view in which
morphologicalstructuresare transformationallyderived, this once again would require
some kind of double markingin the underlyingstructure,with a supposed postposition
markinga nominativesubject.
All of the data in this subsection show that informationallyirregularforms must be
given a lexical treatment, for they would be anomalous in the syntax. These are the
clearest instances of evidence that the formationof inflectionallycomplex words is lexical. As noted, the Minimalistapproachseems to allow such lexical formationin tandem
with syntactic aggregationof the feature structuresthat words must have. Yet even this
approachstill faces two problems: the problems of selection presented in section 4.2,
and a paradox regardingthe point at which selection is checked, presented in section
4.3.
4.2 Problems of Selection

Anotherclass of problemsconcerns the natureof what higherelements in the structure


must select for, under the syntactic view.
Korean has a series of nontensed verb endings, which I refer to as COMPS. Some
relevant data are presented below; each successive COMP allows some, but not all, of
the verbal suffixes to precede it. The basic patternfor verbs, illustratingthe morpheme
order, is given in (30), with the rough class of each suffix indicated (for more details,
see section 5.4); an illustrativeexample is included, and the positions are numberedfor
future reference.
22 Most of the generativeliteraturehas presupposeda distinctionin
the syntaxof these languagesbetween
PP andNP, on some kindof analogywith English,buttruesyntacticdifferencesbetweenthe two are essentially
impossibleto find. For example, right in Englishcan be a specifierof PP but not of NP (Jackendoff1973);I
am unawareof any similarphenomenonin Japaneseor Korean. Rather,there arejust two kinds of nominal
case (for example,the I(nitial)-caseand S(urface)-caseof Gerdtsand Youn (1988)).Moreover,the "grammatical" cases in Korean such as nominativeand accusative do not correspondnecessarily with any particular
functions, appearingas they may on arguments,adjuncts,and nontensedverbs.
Urushibara(1991)claims that there is a (syntactic)distinctionbetween postpositionsand case markersin
Korean.Most of the phenomenareportedin her papershow only that semanticcase markersare attachedto
the noun stem closer in thangrammaticalcase markers(somethingthat is not true for -kkeyse,in any event).
The only remainingevidence is from QuantifierFloat, which is presentedas being blocked by semanticcase
markers:this wouldfollow if they were postpositionsandprojecteda PP, blockingc-commandof the contained
NP and the floated quantifier.However, the claim that only nominativeand accusativephrasesallow floated
quantifiersin Koreanis false (Hong 1990).
23 It is possible that the final -i on the subject in (29) is a focus marker,ratherthan a pure markerof
grammaticalnominativecase. If so, this meansthat (29)does not illustrateanotherinstanceof doublemarking,
thoughthe problemswith the distributionof -kkeysestill remain.

296

PETER

(30)

SELLS

Vroot - Honorific - Tense - Mood - Discourse

ilk
read

-usi

-ess

-ta

-ko

-HON

-PAST

-DECL

-LEVEL

'that (someone (hon.))


read (it)'

The coMPls (which fall in the "1" slot) allow none of the other verbal suffixes (these
are the traditional"continuative" suffixes).
(31) coMPJ
a. cap-a
hold-coMPl

po-ta
try-DECL

'try holding'
b. *cap-usi-e
hold-HON-COMPI

po-ta

(precludesHON)

try-DECL

'try holding' (honorificsubject)


c. *cap-ass-e
po-ass-ta
(precludestense)
hold-PAST-COMPI

try-PAST-DECL

'tried holding'
CoMP2s(in the "2" slot) allow only the honorific suffix, and no others.
(32) coMP2
a. cap-usi-ci
hold-HON-coMp2

anh-ta
NEG-DECL

'(someone (hon.)) does not hold'


b. *cap-ass-ko
iss-ess-ta
(precludestense)
hold-PAST-coMP2

be-PAST-DECL

'(someone) was holding'


CoMP3sallow the honorificand tense markers,but not mood or discourse suffixes.
(33) CoMP3

a.

Cap-usi-ess-eya

hay-ss-ta.

hold-HoN-PAST-coMP3

dO-PAST-DECL

'(Someone (hon.)) had to hold (it).'


b. *Cap-usi-ess-eya-ta
hay-ss-ta.
hold-HoN-PAST-coMp3-DEcL

'(Someone (hon.)) had to hold (it).'


c. *Cap-usi-ess-ta-eya
hay-ss-ta.
hold-HoN-PAST-DECL-CoMp3

(precludesDECL)

dO-PAST-DECL

(precludesDECL)

dO-PAST-DECL

'(Someone (hon.)) had to hold (it).'


As these examples show, there is no coherent syntactic category of "COMP" in Korean:
with regardto (25), the coMPl appearsin the Agr position, the CoMP2in the T position,

KOREAN

AND JAPANESE

MORPHOLOGY

FROM A LEXICAL

PERSPECTIVE

297

and so on. Only one, -ko, which is basically a markerof someone's words or thoughts,
appearsin the expected head of CP position, as in (34).
(34) coMP4

Cap-usi-ess-ta-ko

ha-n-ta.

hold-HoN-PAST-DECL-CoMP4

say-PRES-DECL

'(Someone) says that (someone (hon.)) held it.'


It might be objected that the coMP1-3 forms are not true complementizers.However,
there is no basis for this: all of these COMPS have an embeddingfunction of some sort,
and their form is specifically selected by the higherpredicate. These are the traditional
hallmarksof a complementizer.As far as I am aware, there is no theoreticalreason why
such an element must appearperipherally.True syntactic complementizersmust in fact
appearperipherally,usually to respect principlesof syntax (such as feature percolation
to the dominatingclausal node). However, lexically attached affixal complementizers
need not be so constrained, and this is precisely what we find. Under the lexical view
of Koreancomplementizersdeveloped here, one that distinguisheswhat morphemesdo
from where they appear, these facts are not unexpected.
The next question we mightask is, What does each COMP in turn select for? Unlike
in English, COMP in Korean does not select for IP alone; under the syntactic view, the
coMPl, which precludes other verbal suffixes, should select for VP; the CoMP2should
select for AgrP or NegP, dependingon the analysis; the CoMP3 -eya should select for
TP; and the CoMP4should select for MoodP. To illustrate,given that alongside cap-usiess-eya in (33a), cap-eya and cap-usi-eya are also possible, selection for TP would work
if null morphemesfor Agr/Neg and T were allowed (i.e., cap-eya would have two null

morphemesin it), which is not implausible(see Cho and Sells 1994).In the lexical view,
the analysis would be essentially isomorphic,with each COMP morphemefallingin some
word-internalposition and thereby allowing positions to its left to be filled (see section
5.4).
However, to work correctly, this part of the syntactic view presupposes that there
is a consistent hierarchyof functionalcategories, such that, for example, the existence
of CP always entails MoodP, MoodP entails TP, and so on. However, it is clear that
there is no such hierarchy,either universallyor even within a given language:if we just
look at the expression of Speech Level and Tense, we find that they are reversed in
Japanese and Korean (by "Level" I refer to the relative formality and/or politeness
towardthe addressee that the form encodes). Althoughthe Korean examples in (35) are
not perfect minimalvariantsof each other, they clearly show that Level follows Tense;
this means that Level is projected structurallyhigher than Tense.
(35) a.

mek-ess-ta
eat-PAST-DECL

'ate'

298

PETER

b.

c.

SELLS

mek-ess-eyo
eat-PAST-LEVEL.Plain
'ate'
mek-ess-supnita
eat-PAST-LEVEL.Formal.DECL

'ate'
The opposite is true of Japanese, as shown in (36). The Level morpheme -mas(i)- is
followed by the regularTense suffixes (-ta for past, -(r)u for present). Even from such
simple examples, it is clear that there is no universal hierarchy of the projection of
functionalcategories.
(36) a.

tabe-ta
eat-PAST

'ate'

b.

tabe-masi-ta
eat-LEVEL.Formal-PAST

'ate'

For the syntactic view, the facts are even more puzzling, since even within the same
languagethere may be no consistent hierarchy. In Japanese, for example, the relative
position of these two categories is switched. Polite forms for adjectives have Level outside Tense, as in (37b): the suffix -katta encodes past Tense, and the copular -desu is
the Level marker.The order of these two elements is the opposite in verbs, as in (37c).
(37) a.

ooki-katta (Adj)
big-PAST(Plain Level)

'was big'
b. ooki-katta-desu (Adj)
big-PAST-LEVEL.Formal

c.

'was big'
tabe-masi-ta (V)
eat-LEVEL.Formal-PAST

'ate'

In (37b) -desu is identicalto the present form of the copula-but the overall form has a
past tense meaning, indicating that the only function of -desu is to mark Level. This
leads to what is essentially a paradox for the syntactic view: there are no consistent
underlyingstructuresthat can account for both (37b) and (37c).
Under a lexical view, these facts can be accounted for quite simply. Again, for
purposesof illustration,let us assume that there arejust two word-internalaffix positions
in the Japanese examples in (37). Then, for verbs, the Level slot precedes the Tense
slot; for adjectives, the opposite happens to be the case, with Tense precedingLevel.
The crucial difference between the lexical and syntactic views is that the lexical

KOREAN AND JAPANESE MORPHOLOGY FROM A LEXICAL PERSPECTIVE

299

view separateswhat informationmorphemesprovide from where they appear, without


havingto assign such morphemesto a particularcategory in a hierarchicalstructure.On
the other hand, the syntactic view must assign categories, since each morphememust
be a head that combines with other materialby selecting for it. As the data in (37) show,
there is no possible consistent assignment of categories in Japanese and Korean that
correctly accounts for morphologicalcombinations.
4.3 Problems with Movement

Finally, for the syntactic view, there is one insurmountableproblem that arises from
the nature of the movement operation itself, concerning the point at which selection
applies-namely, the evidence regardingthe point at which selection applies is paradoxical. The data in (38) can be explainedon the movementaccount only if selection follows
movement:in each case the tensed final verb selects for the form of the precedingverb,
indicatedby the underlinedmorpheme.However, some other affix intervenes between
the two verbs.
(38) Selection follows movement

a.

mek-e-man

po-ass-ta

eat-coMP-only

try-PAST-DECL

'tried only eating'


b. tabe-te-mo sima-u (J)
eat-coMP-even

end.Up-PRES

'end up even eating'


For an example such as (38a), the movement account would posit movement of mek to
-e, and then of mek-e to -man, at which point the fact that the first word contains -e
would be visible to the selecting tensed verb. Tonoike (1991) discusses examples like
(38b). Under his proposal,-te would be of category I, and -mo of category C. After head
movement, the structurewould be that shown in (39).
C

(39)
I

tabe

te

mo

However, the argumentabove concerningexamples (21)-(22) indicates that verbs like


simau 'end up' and oku 'do in advance' must subcategorizefor a sister of category V,
in order for the distinctionbetween verbal and adjectivalgerundiveforms to be main-

300

PETER SELLS

tained. Thus, under Tonoike's view, the only possible analysis of (38b) would be that
tabe- moves to -te, and then that whole complex moves to -mo, at which point the
category of the word tabe-te-mo is V. The correct structurefor the first word in (38b)
must then be (40), not (39).
V

(40)
V

tabe

te

mo

Formally,this structurecan be generatedin two ways: one would be to lower the suffixes
down onto the head (Chomsky 1986, Pollock 1989); the other would be to maintain
upwardmovementand to requirethat each functionalcategoryis not in fact a true head,
so that when, say, V adjoins to I, the result is V (as in Grimshaw 1991). It does not
matterwhich alternativeis chosen, for the problemlies in the natureof morphology-viamovement itself.
We have seen that selection can follow movement, since in the underlyingstructure
the particularitem that must be selected for is shielded by other categories; movement
allows this shield to be erased. Consequently,the movementaccount predictsthat selection for a particularaffix will not be affected or disturbedby other affixes that appear
outside the targetedone. However, there are some cases that are clear counterexamples
to this prediction. Forms precedingthe copula must lack certain suffixes, as shown in
(4la-b); in particular,membersof the outermostset of nominalsuffixes are not allowed
(such as the topic marker,and the nominative and accusative case markers),and this
suggests that selection must precede movement.24
(41) Selection precedes movement

a.

b.

sensee-(*wa)-da (J)
teacher-(*TOP)-COP. PRES
'is a teacher'
sensayng-(*un)-i-ta (K)
teacher-(*TOP)-COP-DECL
'is a teacher'

24 In Korean none of the Z-LIM


categories listed in (57) may precede the copula, which itself may be
phonologicallyabsent (in (41b-c)). This could be analyzedeither as a phonologicaldeletion process or as a
null copula.

KOREAN

AND JAPANESE

MORPHOLOGY

FROM A LEXICAL

PERSPECTIVE

301

c.

Ilen
iyaki-nunSwuni-hanthey-man-i-ta.
this.kind story-TOP Sooni-DAT-only-CoP-DECL
'This story is only for Sooni.'
d. *. .. Swuni-hanthey-man-un-i-ta.
. ..

Sooni-DAT-only-ToP-CoP-DECL

'...

is only for Sooni.'

Let us again consider Tonoike's (1991) account for the Japanese example (41a). His
proposalis that sensee is an NP and that sensee-wa is a DP, since wa is of category D.
Thus, the facts would seem to be that the copula selects for NP and not DP. However,
if the copulais generatedas sister to DP, underthe movementaccount the head N sensee
will join up with the D -wa via movement, and given what we saw above for the verbs,
the result will itself be an NP. Hence, the fact that (41a) does not allow -wa remainsa
mystery.The alternativewould be that selection precedes movement,but this is precisely
the opposite of the conclusion from the verbal examples in (38).
The movementaccount mighttry to avoid this problemby enforcing,in some way,
the idea that the copula selects for a bare, or underived, noun. However, this strategy
would not work, as the Korean examples in (41c-d) show: the copula can host a noun
that is suffixed (Swuni is suffixed with -hanthey and -man), so long as no membersof
the last series of suffixes (such as -un) are present.
Movement is a way of allowing a head to communicateacross another head, by
adjoiningto it. The data in (38) and (41) show that sometimesthis works, and sometimes
it fails, which is truly a paradox for the movement approach. The crucial difference
between the cases with verbs and the cases with nouns is that in the former instance
selection takes place in the syntax and involves category. That is, as is standardlyassumed, certainauxiliaryverbs select for other verbs, and they will not allow adjectives.
As furtherinflectional suffixationdoes not change the category of the resultingword,
this selection will be unaffectedby interveningaffixes (see (21)-(22), (38)). On the other
hand, the copula is a lexical suffix (see Cho and Sells 1994), and so its selection is
morphological-essentially, it selects for a nominalconstituentof a certain "size," since
it falls in a certain position in the structureof the word. Consequently, this must be
sensitive to the presence of other affixes. However, the syntactic view of morphology
(by design)conflates the notion of "size" into the notion of category, for each additional
suffix is a head of a separatecategory. In fact, there is really only one "size," namely,
the maximalprojection.Having confoundeda necessary distinction, the syntactic view
finds itself with a paradoxsuch as the one presented here.
The apparentsolution to this paradox is that, in general, the syntactic view will
require that bound affixes have two types of subcategorization(see the beginning of
section 4), syntactic and morphological(see Speas 1991);Yoon (1991a)suggests that the
Korean copula is a phrasal affix that subcategorizes for XP in the syntax and for a
sublexical category in the morphology, a constituent that excludes the final nominal

302

PETER

SELLS

suffixes, as shown below. Yoon refersto this as X- l, using a word-internalX-bartheory


(see Selkirk 1982).25Thus, its morphologicalsubcategorizationaccounts for the data in
(41).
Let us look more closely at how this might work. The upper part of the structure
in (42) shows the relevant part of (41c), the complex nominal structureand the copula
root. For convenience, I have labeled the delimiter-man 'only' as the head of FP. However, following Grimshaw(1991) and Yoon (1991a), I take PP and FP to be extended
projectionsof N.
VP

(42)
FP
PP

Cop
F

NP

hanthey

man

Swuni
Swuni N-3]

hanthey N-2]

man N-1]

i V-3]

The lower part of (42) shows the morphologicalstructure,derived by successive head


movement:adoptinga word-internalX-bar theory, in this structurethe copula subcategorizes for N-'; the delimiter -man for N 2; and the postposition -hanthey for N-.
The copula itself is listed as a V-3 under this scheme, since it can take a full range of
verbal suffixes. By assumption, the N -I lexical constituent will exclude the outermost
nominalsuffixes, and so the unacceptabilityof (41d)is accountedfor by the morphological subcategorizationof the copula.
In itself, dual subcategorizationis not problematic,but once morphologicalsubcategorizationis admitted,it takes away much of the appeal that the syntactic view is supposed to have. As we have seen, given the morphologicalrequirementsof the copula,
the syntactic part of the subcategorizationbecomes wholly redundant;if the syntax
provideda constituentof the wrong type, such as a phrase ending with a topic marker,
the morphologicalcombinationwould fail. Similarly,if the FP were generatedwithinthe
PP in (42) (as *. . . -man-hanthey ... .), the morphological subcategorization of -hanthey
25 Note that this is anotherpiece of evidence that the copula is
a true affix, not a clitic, for it interacts
with whateverconstraintsdeterminewell-formednesswithinthe morphologicalstructures.

KOREAN

AND JAPANESE

MORPHOLOGY

FROM A LEXICAL

PERSPECTIVE

303

for an N -3 would not be satisfied, since -man makes an N- 1, by assumption.Pursuing


this line of reasoning,it would seem that the syntactic subcategorizationshould be eliminated, since it provides no useful information;the same is true for the syntactic category
information.Note, for example, that there is nothing about classifying -hanthey as a
postpositionthat does any work underthe dual subcategorizationview; no matterwhat
its syntax, the well-formednessof the overall structureis determinedonly by its morphological properties.This is in fact as it should be, as evidenced by the behaviorof -kkeyse
in (29), which falls in the "postposition" morphologicalslot but has no syntactic or
semanticpostpositionalproperties.
At this point, though, much of the motivationfor the syntactic view of morphology
falls away-it is saddledwith providinga syntacticcategoryand subcategorizationwhere
none is needed and, as -kkeyseshows, where none shouldbe given. If the grammarstates
the minimalamount of idiosyncraticinformation,then there is no commonalityacross
languagesafterall: the Englishcopulawould subcategorizefor NP, yet the Koreancopula
would subcategorizefor N- l. This will be true for all of the suffixes; there is no need,
for example, to requirethat a tense marker,as the head of a TP, should subcategorize
for a VP, since its morphologicalspecificationwill be for a verbal sublexicalconstituent
that could only have come from (the head of) a VP.
On the other hand, the lexical view offers a straightforwardaccount of the copula:
it is lexically attachedand, as a result, cannothave a syntacticcategoryor subcategorization. It attaches in the final slot of the noun, thereby precludingany other suffix from
appearingin that slot, and derives a verb.26Whatevercommonalitiesthere are between
the English and Korean copulas should emerge at some other level, be it semantic or
functional;any attemptto show commonalitiesin the phrasestructureis at best redundant
and at worst paradoxical.
4.4 Summary

There are, then, various argumentsthat can be made that challenge every aspect of the
idea of doing morphologyin the syntax: difficulties in providingconsistent underlying
structures,missed generalizationsin categorization,and a paradoxof selection that can
only be solved at the expense of much of the motivationfor the whole approach.All of
this evidence illustrateshow the morphologyof Japanese and Korean is in fact formed
by differentmechanismsand principlesfrom the syntax.
5 Lexical Account

Havingestablishedthe necessity of a lexical account for the data presentedabove, I will


concentratein the remainderof this article on the primaryfeatures that such an account
26 A Minimalistapproachto the data involvingthe copula appearsto be forced to a fully lexical analysis:
a form like that in (42), which is supposed to involve head movementfrom N to V, cannot be derived by
movement.If the morphologyproducesfully derivedand inflectedforms, this one is of categoryV and cannot
be insertedunderan N node. Therefore,this approachgeneratesfully inflected words in (or, at least, close
to) their surfaceposition, which appearsto be a variantof the lexical analysis.

304

PETER SELLS

should have. The materialin this section is based in part on the analysis presented in
Cho and Sells 1994and is only intendedto be the broadoutline of what a fully articulated
lexical view must aspire to.27
5.1 From Syntax to Morphology

The tree in (25), repeatedhere, showed the complex clausal structuretypically assigned
to Korean verbal forms.
(25)

CP
MoodP
TP

Mood

NegP
AgrP
VP

T
Neg

Agr

ilk
-usi
-ci
read
-HON
-CoMP2
'that (someone (hon.)) did not read'

anh

-ass

-ta

-ko

NEG

-PAST

-DECL

-coMP4

Thereis some variationamongresearchersas to which functionalprojectionsthey adopt,


but the upperones in (25) enjoy particularlywidespreadacceptance.
Withregardto AgrP, Yoon (1990)rejects the idea that Koreanhas Agr in the syntax
(he does, however, adopt the other functionalprojectionsshown in (25)) (see also Yoon
1993).Many other researcherstake subject honorificationto be a reflex of Agr, though
I believe that this is mistaken, as discussed in footnote 21. And at a purely theoretical
level, the idea that there is Agr conflicts directly with Kuroda's (1988) proposal that
Japanese (and Korean) differ parametricallyfrom English in that they have no forced
agreement.28

27 In Cho and Sells 1994the details of the analysis are presentedwithinthe


frameworkof Lexical-Functional Grammar(LFG) (see Bresnan 1982).Here, I present a slightlydifferentversion that does not depend
on the details of LFG.
28 A reviewerobserves that certaineffects in the blockingof long-distanceanaphorahave been attributed
to Agr, in languagessuch as Korean(see Cole, Hermon, and Sung 1990).Specifically,the interventionof a
firstor secondpersonsubjectblocks long-distanceanaphorato a thirdpersonsubject.As an alternativestarting
point in an account that did not posit Agr, I would take it that such blockingeffects are to be explainedin

KOREAN

AND JAPANESE

MORPHOLOGY

FROM A LEXICAL

PERSPECTIVE

305

Negation can surface either in the "short-form"preverbalexpression in (43a) or in


the "long-form"postverbal expression in (43b), with the complementizer-ci.
(43) a.

an mek-ta
NEG eat-DECL

b.

'does not eat'


mek-ci anh-ta
eat-COMP NEG-DECL

'does not eat'


Most proponentsof the syntactic view of morphologytake the short-formnegation to
be the reflex of a functional head; others take it to be lexically prefixed (Yoon 1990).
Followingthe argumentsin Sells and Cho 1991,I take it here that the short-formnegative
is in fact an adverb, and thereforeit does not participatein the (suffixal)morphologyat
all. Treatingit as an adverbexplains its position and the fact that it is invisible to lexical
phonologicalprocesses (see Sells and Cho 1991).
In an unfortunateanalogy with English, the long-formnegative is often treated as
a head that cooccurs with or triggersinsertion of a dummy verb ha-ta 'do' to support
the tense marker(for recent proposals, see Ahn 1991 and Cho 1992);here I treat anhta as an independentauxiliaryverb, which patternswith other uncontroversialauxiliary
verbs (see Sells 1991, Yoon 1993). Crucialevidence that anh-ta is not derived through
insertionof a verb into T in (25) comes from the fact that honorificationcan both precede
and follow anh-ta-the form ilk-usi-cianh-usi-ess-ta-kois an acceptablevariantof (25).
This fact follows automaticallyfrom the view that takes anh-ta to be an independent
verb, but is totally at odds with the idea of insertion of a dummy verb: if the dummy
verb is introducedto supportthe tense marker,which is the usual analysis, then there
is no explanationfor the honorific markeron it (the second -usi), assumingthe clausal
structurein (25). On the other hand, if the hierarchyof functional categories assumed
is different,with the dummyverb introducedunder the honorific markeritself (i.e., an
underlying order V - Neg - T - Agr), then honorification should not be present on ilk-

'read'.
Additionally,if the form to which the honorificationattaches is the dummyha- (or
in fact, any verb with the form ha-), then the expected output for the negative part of
the long-formnegation would be *an-ha-si-ess-ta-ko,which is impossible (the correct
formis anh-usi-ess-ta-ko).This shows that the host for honorificationis consonant-final,
ratherthanvowel-final,and it counts directlyagainstthe dummyverb idea.29(My thanks
to Steven Lapointefor helpful discussion of this point.)
termsof a sensitivityto some kindof discourseperspective,the domainof which would be influencedby first
and second personpronouns,which inherentlyhave their own direct perspective.To my knowledge,no one
has claimedthat the presence or absence of subject honorificationaffects anaphoricbindingpossibilitiesin
Korean;yet, as my discussion in this article suggests, honorificationis usually taken to be the most likely
true instantiationof Agr.
29 In the case of JapaneseI know of no evidence for NegP-in fact, negationseems to be a derivational
suffix that derives an adjectivefrom a verb.

306

PETER

SELLS

In contrastto the syntactic view that leads to the structurein (25), the lexical view
holds that there is some kind of generalword-internalscheme for languageslike Korean
andJapanese,illustratedin the structuresin (44).3? For this particularexample, the longformnegationinvolves a sequence of two V's (formedin the syntax, the second selecting
for the first), each one of which is morphologicallycomplex.
V0

(44)

V0
V

V0
Af

Vroot

Vroot

ilk
-usi
-ci
anh
read
-HON
-CoMP2 NEG
'that (someone (hon.)) did not read'

Af

Af

Af

-ass

-ta

-ko

-PAST

-DECL

-CoMP4

I use V? to represent words in the syntax; V and Vrootrepresent constituents in the


morphology.
As shown in (44), the morphologyitself is hierarchical,with a root position at the
left, as shown above for the verbs in (30), repeated here as (45). The differentclasses
of verbal suffixes are presented in more detail in section 5.4.
(45)

Vroot - Honorific - Tense - Mood - Discourse

Alongside the verbal forms, there are also complex nominalstructures;a Korean word
like sensayngnim-hanthey-nun'teacher-DAT-Foc'
will have the structureshown in (46).
(46)

NO
N

Nroot

sensayngnim
30
COMP'.

Af

Af

-nun

-hanthey

The (or rather, one) Japanese equivalent of this example is yom-are-na-katta-to

'read-HON-NEG-PAST-

KOREAN

AND JAPANESE

MORPHOLOGY

FROM A LEXICAL

PERSPECTIVE

307

This nominal structureis actually slightly more complex in that it is licensed through
two morphologicalpatterns.The first is formed on Nroots and licenses the Postpositions
such as -hanthey,as well as the Conjunctivesuffixes; see section 5.3. The second licenses
the Delimiters such as -nun and is a cross-categorialstructure, in the sense that the
Delimiters can attach to verbal forms as well (details regardingDelimiters, which fall
into the two classes known as X-LIM and Z-LIM, are given in section 5.5).
(47)

Nroot-

Postposition - Conjunctive

(48) X - Delimiter

(X-LIM)

Delimiter

(Z-LIM)

I assume that morphologicalcombinations are freely generated as binary structures,


with these patternslicensingthe acceptableoutputs. Thus, sensayngnim-hantheyis well
formed because it has the order Nroot- Postposition, respecting (47); this form counts
as an X, which is followed by the Z-LIM delimiter-nun, respecting (48).
For the phrasalsyntax, I assume that it consists primarilyof base-generatedadjunction at the X' level, following the ideas of Fukui (1986). Clause-internalscramblingis
just the free generationof phrases in any order (see Sells and Cho 1991). For example,
the nonhead (N') phrases in (49) may scramblewith respect to each other, and this is
accounted for by allowing them to be generatedattached at the V' level, in any order.
(To keep the example simple, each X' in (50) dominatesonly an XO.)
(49) Swuni-ka yenge-lul mikwuk-eysepaywu-ess-ta.
Sooni-NOM
English-AcCAmerica-in learn-PAsT-DECL
'Sooni learned English in America.'
VI

(50)

N'

Swuni-ka

V'

VI

N'
N'

VI

mikwuk-eyse

paywu-ess-ta

yenge-lul

In these structures,grammaticalfunctions such as subject and object are defined via the
informationin the case-markingand argumentstructure(see Hong 1991),ratherthan via
some designatedstructuralposition.
In Koreanat least there is also evidence for certainrestrictedtypes of V0 adjunction
(see Choi 1991,Sells and Cho 1991,Cho and Sells 1994)illustratedin (51) and (52); only
the X' phrases may scramble, but within the domain of V's, the structureis fixed.

308

PETER

SELLS

po-ass-ta.
(51) Swuni-ka yenge-lul cal ilk-e
Sooni-NoMEnglish-Accwell read-coMptry-PAST-DECL
'Sooni tried readingEnglish well.'
V

(52)

V'

N'

V'

Swuni-ka N'
yenge-lul Adv
cal

V?
V0

V0

ilk-e

po-ass-ta

5.2 Combinatoric Type

The view of the syntaxjust presented is the traditionalone, in that it is uniformlyrightheaded. And althoughthe inflectionallexical structuresin (44)-(46) are essentially leftheaded, in that the suffixes are invisible to categorial selectional processes, there is in
fact importantinformationthat is always determinedby the rightmostelement, called
combinatoricinformationin Cho and Sells 1994. For example, suffixing the genitive
marker-uy to a noun licenses that noun to appearin the syntax as sister to anothernoun
or nominalprojection, whereas suffixing the nominative-ka would license it to appear
as sister to a verb or verbalprojection.This combinatoricpropertyis always determined
by the outermost(= rightmost)suffix, which I refer to as TYPE. The flow of information
within a word is shown schematicallyin (53).
(53) Information flow in inflectional structures

Xo
combinatoric
information

categorial
information
(what heads

Root .

. Suffix

(what can be

select for)
X's right sister)
Informationconcerning semantics, case, and so on, is inheritedfrom all
morphemes.

KOREAN

AND JAPANESE

MORPHOLOGY

FROM A LEXICAL

PERSPECTIVE

309

Specifically, althoughthevwordconsists simply of a root followed by some suffixes, in


a flat structure,the initial memberdeterminesthe category of the whole word, and the
final member(whateverit is) determinesthe TYPE. As will become clear below, syntactic
category and combinatorictype are independentparametersof specification. The categorial informationis what was discussed in the firstpartof this article-anything thatcan
be selected for by some other head, in particularsyntactic category. The combinatoric
informationaugments the very general rules for phrasal syntax to predict the correct
distributionof phrases, described in more detail below. The rest of the informationis
primarilysemanticin natureand is directly inheritedfrom either side of a morphological
combination.For instance, with a verb, this informationmight concern the meaningof
the verb itself, tense, aspect, mood, and speech-level information.3"
Formally, the combinatoricproperties are expressed by assuming that each node
has an attributeTYPE, which can have the following values: V-SIS, N-SIS, and ROOT. In
the syntax, v-sis means that the element can be sister to a verbalprojection,N-SIS means
that it can be sister to a nominal projection, and ROOTmeans that it can have no right
sister.32An example of a suffix that carriesthe ROOTvalue is the Koreanplain-stylelevel
marker-yo (as in ka-yo 'I'm going'); phrases whose specification is ROOTmust fall at
the end of a whole sentence, or more precisely, at the end of an utterance.
Varioussuffixes change the TYPEspecificationof the word that they are partof, but
in the syntax TYPEremainsinvariant,always inheritedfrom the right. The three values
are illustratedin (55), the syntactic structureof (54) (p. 310). The TYPE contributionof
each relevant suffix is indicatedby a subscripton it. The genitive marker-uy specifies
that the word that it ends is [TYPE:N-SIS]. This specificationis inheritedthroughthe X'
projection in the syntax, and thus the leftmost N' in (55) takes a nominal projection as
its right sister. That right sister is itself specified for a TYPEvalue of v-sis, and thus the

whole subject takes a verbal projectionas its right sister. The verbal projectionitself is
[TYPE:ROOT], meaningthat it can have no right sister.
I should point out that TYPEis not merely some descriptive device that happens to
coincide with obvious propertiesof syntactic combination,necessitated by what some
would consider a rather naive view of the syntax of Korean. In the remainderof this
article I will demonstratehow the notion of TYPE plays a crucial and irreduciblerole in
constrainingand predictingthe possible morphologicalcombinations. In the next two
subsections I look at the particularsof nominaland verbal suffixalinflections and at the
role of TYPE in regulatingthe morphologicalcombinations.

3' The fact that much informationcarriedby morphemesis directly passed up to the whole from any
daughteralso seems to suggestthat these combinationsare not syntacticin the true sense. On the other hand,
Selkirk(1982:76ff.)notes the necessity of this "summingup" of informationon the mother in inflectional
word-formationprocesses.
32 In Japanesethe primaryopposition is between N and V/Adj, the membersof the latter group both
allowingtense inflection.Thus, the values for TYPE mightbe more perspicuouslyexpressed as somethinglike
"inflecting"and "noninflecting."

PETER SELLS

310

(54) Kim-uy chinkwu-kawus-ess-ta.


Kim-GEN

smile-PAST-DECL

friend-NoM

'Kim's friend smiled.'


V'

(55)

[TYPE:ROOT]

N'

VP

[TYPE:V-SIS]

[TYPE:ROOT]

N'

N'

V0

[TYPE:N-SIS]

[TYPE:V-SIS]

[TYPE:ROOT]

N0

!TN?

wus-ess-taROOT

[TYPE V-SIS]

[TYPE:N-SIS]

chinkwu-kav-sis

Kim-UYN-SIS

5.3 Nouns
The nominal suffixes of Korean fall into the classes shown in (56) and (57). The names

for each columnare chosen purelyfor convenience; they have no theoreticalstatus. The
terms X-LIM and Z-LIM are taken from Yang 1972.
(56) Postpositions
eykey(se),
hanthey(se)
ey, eyse
ey, (u)lo
(u)lo
kkaci
hako, (k)wa
kkey
kkeyse

dative
locative
directive
instrumental
goal
comitative
dative (hon.)
hon. subj.

Conjunctives
conjunctor
hako,
(k)wa
comparator
pota
disjunctor/
(i)na
'something like'
'from'
pwuthe
'like'
chelem

(57) Delimiters
'X-LIM

Z-LIM

man

'only'

(n)un

TOPIC/FOCUS

kkaci
mace
cocha
pakkey

'even'
'even'
'even'
'only'

to
(i)lato
i/ka
(l)ul

'also'
'even'
ACC

uy

GENN-SIS

NOM

KOREAN

AND JAPANESE

PERSPECTIVE

FROM A LEXICAL

MORPHOLOGY

311

These suffixes participatein the word-internalpatterns as indicated at the top of each


column, and most of them specify TYPE as v-sis. The conjunctive suffixes require a
special treatmentthat I will not describe here (see Cho and Sells 1994).The genitive -uy
specifies [TYPE:N-SIS], as indicatedby the subscriptN-SIS. The partitioningshown is not
arbitrary-the X-LIM and Z-LIM categories (even the case markers)may also attach to
verbalformsin some instances, whereasnone of the Postpositionsor Conjunctivesmay.33
Consequently, the pattern that licenses their appearance ((48), repeated here) is not
restrictedto nominalcategories. The purely nominalsuffixes are licensed by the pattern
in (47).
(47) Nroot - Postposition - Conjunctive
(48) X - Delimiter (X-LIM)

Delimiter (Z-LIM)

As an example of the combinatoricsof the morphology,the word Swuni-hanthey-manun 'only to Sooni (focus)' has the structurein (58) (the first two morphemesare licensed
by (47), the last two by (48)). The TYPE value is v-SIS since the rightmostelement -un is
of that TYPE. The relevant semantic contributionof each morphemeis inheritedby the
whole word.
N

(58)

'Sooni'

PRED

PCASE
X-LIM

DAT

'only'

DISC =

FOC

TYPE:XV-SIS

Z-LIM
[DISC

[TYPE:V-SIS]

[TYPE:V-SIS

FOCi

-un

X-LIM

[TYPE:V-SIS]

rX-LIM

[TYPE:v-sIs]

'only l

TYPE: V-SIS

-man
N
[PRED=

POST

'Sooni'1

-TYPE

Swuni

[PCASE

= DATi

TYPE: V-SIS

-hanthey

3 The suffixes -cocha, -mace, and -uy in (57) can only appearwithin nouns. Also, in apparentviolation
of the orderimposedby (56)-(57), the sequence -manulois attested. However, there is evidence that this is
actually a lexicalized postpositionitself; for example, like other postpositions, it precedes the conjunctive
marker-hako, whereas-man by itself does not. See Cho and Sells 1994.

312

PETER

SELLS

Noun (and verb) roots themselves have no value for TYPE on the reasonableassumption
that they can never appear"bare," without any suffixes. In (58) the root Swuni-has no
value for TYPE, indicatedby "-TYPE."34
Withouta specificationfor TYPE, they cannot
be insertedinto the syntax. The words Swuni-hantheyand Swuni-hanthey-manwill also
be [TYPE:V-SIS] since the last suffix in each case specifies this. Thus, any of these, along
with (58), is correctly predicted to be able to appear as sister to a verbal projectionin
the syntax.
I should reemphasizethat each "slot" in (56)-(57) does not correspondto a unified
(syntactic)category;as mentionedabove, -kkeysefor exampleappearsin the Postposition
slot, but shares no other postpositionalpropertieswith the other morphemesthat appear
in that slot. It marksan honorific nominativesubject (the example is repeated in (59)),
assigning(surface)CASE ratherthanthe semanticor "inherent"PCASE that postpositional
morphemesassign.
(59) [Sensayng-nim-tul]-kkeyse-man-i kulen il-ul
[teacher-HoN-PL]-HoN.suBJ-only-NoM
that.kindwork-Acc
ha-si-pnita.
dO-HON-LEVEL.Formal.DECL

'Only teachers do such work.'


This example provides crucial evidence against any view of Korean case whereby postpositionalcase is supposed to representunderlyinglevels of structure,and grammatical
case (nominativeand accusative) is supposed to represent surface levels (such as the
derivationalview of the orderingof case marking,as in the Satellite Principleof Gerdts
and Youn (1988), or its reinterpretationby Yoon and Yoon (1991)).35 Rather, -kkeyse
must specify that the case is nominative and, perhaps unusually, that the word that
contains it must head a (surface) subject phrase in the syntax. I use "head a phrase"
here literally, since a -kkeyse-markednoun cannot even appear coordinated inside a
subject.36

Another example of the heterogeneity of each morphemeslot can be found in the


delimitersin (57); these were not presented as nominal suffixes, for most of them can
appearafter nontensed verbs and are actually cross-categorial.Thus, althoughit might
be plausibleto think of the topic marker-(n)un as being of syntactic category D(eterminer), this becomes less attractive when we find such morphemeson verbs (which are
not nominalizedfirst, as arguedin section 5.5). Similarly,the nominativeand accusative

3 Cho and Sells (1994)posit a zero morpheme(specifyingTYPE) for nouns that appearwithoutany (overt)
case marker.
3 The Satellite Principlestates, "If an element A is licensed in an earlierstratumthan element B, then
A appearsinside B."
36 In the LFG notation of Dalrymple 1993 (used there for stating anaphoricconstraints),the relevant
informationassociated with -kkeyseis ((SUBJ t ) SUBJ) = T . This requiresthat the word containing-kkeyse
functionas the subject. Outsideof LFG, I do not even know how to state this restriction.

KOREAN

AND JAPANESE

MORPHOLOGY

FROM A LEXICAL

313

PERSPECTIVE

case markerscan appearon verbs suffixed by the (negative)coMP2 -ci. It is implausible


to think of such suffixes as representingany (nominal)functionalcategory.37
5.4 Verbs

The verbal suffixes fall into the classes shown in (60) and (61). Those forms that specify
TYPE as either v-sis or N-SIS are notated with the relevant subscript. The classes are
categorizedby their morphologicalpositions-plausible heuristicnames for some of the
classes are hard to find (cf. the many terms used in Martin 1954).38
V2

(60) VI
(U)Si

HON

ess, ass

past tense/aspect

e/a

COMPlvwsis

keyss

future tense/aspect

nun, n

(pres.) processive

ci, key, ko

coMP2v-sis

nun, (u)n
(u)l
ten

pres. prenom. mod.N SIS


fut. prenom. mod.NSIS
past habit. prenom. mod.N

SIS

37 Some speakers also allow the accusative case marker-lul on verbs with the cOMPl suffix -el-a (Cho
1988,Cho and Sells 1994).
38 Of course, not all combinationsof suffixes are possible; see Cho and Sells 1994for a discussion of
some of the restrictions.The V2 suffixes allow some limitedrecursionthat I do not account for here; -essess is possible in some circumstances,possibly as a kind of remote past (Martin1992),and the regulartense
markerscan apparentlycombinewith some nominalmodifierforms, as in (i) and (ii).
(i) mek-ess-ten
pap
eat-PAST-MoD.PAST.HABIT

rice

'the rice (someone)ate up'


(ii) mek-ess-ul
ttay
eat-PAST-MOD
time
'when (someone) had eaten it'
It is also necessary to posit zero morphemes for nonhonorific forms, and for present tense, in the V2 slot, as
in (iii), which compares with (iv).
(iii) coh-0-0-ta
good-NONHON-PRES-DECL

(iv)

'is good'
coh-usi-ess-ta
good-HON-PAST-DECL

'was good (hon.)'


Such zero morphemes can be eliminated in favor of a kind of default inheritance if a "structured lexicon" is
adopted (see Han 1991, Pollard and Sag 1994), though I do not explore that here.
A reviewer points out that forms like (v) and (vi) are possible, seemingly having two V3 suffixes.
(v) mek-ess-ta-ciman
eat-PAST-?-though

'though(someone)ate'
(vi)

ha-n-ta-myen
do-PRoC-?-if

'if (someone) does'


However, as far as I can ascertain, the only V3 suffix that can precede -ciman and -(u)myen is -ta, so I would
assume that -ta-ciman and -ta-myen are themselves (complex)V3 suffixes.

PETER SELLS

314

V4 ("Discourse")
yo
semiformallevel
familiar
ney

(61) V3 ("Mood")
declarative
ta
interrogative
kka, nya

ko
man
nun

conjunctive
coMP3vsis

ko
eya, aya
na

coMP3v-sis

(su)pnita
(su)pnikka
ciman
(u)myen
e/a
ela/ala
ca
ci

decl. formal
intern.formal
disjunctive
conditional
plain
imperative
propositive
tag

coMP4v-sis
polite
prenom. mod.NSIS

An example like ilk-usi-keyss-ta'(someone (hon.)) will read' will have the structure
shown in (62); the verbal patternin (45) is repeated here.
(45)

Vroot -

Honorific (VI) - Tense (V2) - Mood (V3) - Discourse (V4)

(62)

= 'read'

PRED

HON
TNS
MOOD

+
FUT
DECL

TYPE:ROOT

V3
FMOOD = DECLi

[-TYPE]

[TYPE:

ROOT |

-ta
V

V2

[-TYPE]

rTNS = FUTi

- TYPEJ

-keyss
VI

V
PRED =
[

-TYPE

ilk

read'1

[HON =

- -TYPE

-usi

+
J

KOREAN

AND JAPANESE

MORPHOLOGY

FROM A LEXICAL

PERSPECTIVE

315

Onlythe last of the morphemesin (62) has a positive specificationfor TYPE. By definition,
these [TYPE:ROOT] forms have no rightsister and thereforeshow little about the morphology/syntax interface. The next two subsections discuss cases more interesting in the
present context: verbs whose sister is verbal, and verbs whose sister is nominal.
5.5 Cross-Categorial Suffixes

Cho (1988)claims that the delimitersuffixes in (57), such as -man 'only', attach only to
N. Under her account, in which the morphologyis fully right-headed,this entails taking
the locative particle-eyse in (63b) to be of category N, as well as the COMPS in (63d) and
(63e).
(63) a.

kkoch-man
flowers-only
'only flowers'
b. samwusil-eyse-man
office-Loc-only

'only in the office'


ecey-man
yesterday-only
'only yesterday'
d. coh-key-man

c.

good-coMP3-only

e.

'only well'
ilk-e-man
read-coMPl-only

'only read'
Given Cho's assumptions, the form ilk-e in (63e) is of category N. However, as noted
in section 2.2, -manis phonologicallysensitive to the categoryit attachesto: if it attaches
to V, the host must be disyllabic. If the hosts of -man in (63d)and (63e) were of category
N, the phonologicalasymmetrybetween N and V would be lost. Thus, the correct account of -man must be that it attaches to various categories, as proposed here, and the
examples in (63d) and (63e) are verbs.
Thus, there are certaincross-categorialsuffixes that may follow the COMPS, as exemplified once more in (64). This shows that these COMPS do not simply "complete" the
wordthat they are partof and that they are really nothingmore thaninflectionalsufflxes.
Moreover, it indicates that many of the COMPS (such as coMPl) are in no way related to
such putative categories as CP, since they do not appearfinally.
(64) Ilk-e-man-un

po-ass-ta.

read-coMPl-only-FOC try-PAST-DECL

'tried only reading'

PETER

316

SELLS

At this point it is appropriateto ask what the COMPSare for. The answer is quite simple:
the COMPSare all and only those morphemesthat license their host verb to appear as
sister to another verb; that is, they are the only verbal suffixes that specify [TYPE:
v-sis].39 They may not all have a true embedding function, but they all allow one verb

to be in constructionwith another verb, a notion that is representedhere by TYPE. In


Englishthe COMPShave an embeddingfunction, but no restrictionin terms of TYPE: the
clauses they head may equally well appear in constructionwith verbs, adjectives, and
nouns.
Above, I have made several argumentsbased on the fact that the apparently"nominal" delimitingsuffixes attach to verbs. If we now ask what governs their distribution
on verbs, we find that once again combinatorictype is the relevant factor. All of the
post-coMPsuffixes shown in (65) are intuitively "nominal" suffixes."4
(65) a.

phyenci-lulssu-ci-lul
letter-ACc

anh-ass-ta

write-coMP2-ACc not-PAST-DECL

'did not write a letter'


b. ilk-e-to
po-ass-ta
read-coMPl-also

c.

try-PAST-DECL

'tried also reading'


ha-n-ta
ilk-eya-man
read-coMP3-only

have.to-PRES-DECL

'must only read'


d. ka-ko-tul-un
siph-keyss-ta
go-CoMP2-PL-FOC desire-FUT-DECL

e.

'will want to go (pl.)'


ka-key-to
hay-ss-ta
go-coMP2-even

do-PAST-DECL

'made (someone) even go'


In terms of the concepts presented here, the majorityof the nominal suffixes specify
[TYPE:V-SIS];
and, as I have noted, so do the COMPs. What emerges from these observations is the correctgeneralization:nominalsuffixes thatattachto verbalformsare suffixes
of [TYPE:V-SIS], and the verbal forms that they attach to end in COMPs,which are also of
In otherwords, the nominalsuffixes thathave extendedto the verbaldomain
[TYPE:V-SIS].
can do so, since attachingthem makes no difference to the combinatoricsof the unsuffixed form. Their attachmentis not sensitive to the category of their host, only to its
39 In Japanese the relevant COMPS seem to be -to for tensed-clause embeddingand -te for nontensed
constituents.
40 The plural marker-tul in (65d) does not appear in the charts above because it shows considerable
freedomwithrespectto whereit mayattachwithinthe word. However,this does not affectthe characterization
of the TYPE (v-sis) of its host, which is the relevantconsiderationhere.

KOREAN

AND JAPANESE

MORPHOLOGY

FROM A LEXICAL

PERSPECTIVE

317

A significantconsequence of this is that case markerscan appear on verbs and


adverbs in Korean, which remainsa puzzle under purely category-basedaccounts.4
Considerfor example the contrast between the examples in (66).

TYPE.

(66) a. *Cap -usi -ess -ta -kkaci-nun.


catch -HON -PAST -DECL -even -TOP
'(Someone (hon.)) even caught (it).'
b. Cap -usi -ess -ta -ko
-kkaci -nun (malha-ci anh-ass-ta).
catch -HON -PAST -DECL -CoMP4 -even -TOP (say-CoMP2 not-PAST-DECL)
'(Someone didn't say) that (someone (hon.)) even caught (it).'
In (66a), the Vroot cap- has honorific, tense, and mood suffixes. The declarativeending
-ta is of [TYPE:ROOT], and suffixation that changes this TYPE specification is blocked;
hence, the delimiter-kkaci and (or) the topic marker-nun cannot attach to this form.
But if we add the COMP4 -ko, as in (66b), the example is grammatical.The reason for
this is that -ko specifies TYPE as v-sis, and then suffixationof -kkaciand -nun preserves
this specification.42
The specificationof TYPE, then, is not just an artifactof this analysis but is in fact
an irreduciblecomponentof the morphologyand syntax. To illustratethe interactionof
the verbal and delimiterpatterns ((45) and (48)), the structureof the first word of (64)
is shown in (67). Some indication of the semantic contributionof each node, which
41 A reviewersuggests that the delimitingparticlesdiscussed here may attach to VP. This would appear
to requirethat each COMP
be classified as a type of V. However, if the delimitersattachto VP, it is not clear
whatwouldrule out theirattachingdirectlyto the bareV (i.e., the examplesin (65) withoutthe COMPs).Under
my proposal,such examples are ruled out because the TYPE of Vroots is not v-sis.
42 A reviewer notes that -ko can be droppedin certain contexts (similarto that-"deletion"in English),
in examples such as (i) or (ii).
(i) Ku-ka olh-ta(-ko)
sayngkakha-n-ta.
he-NOM right-DECL(-coMP4) think-PRoC-DECL

'(I) thinkthat he is right.'


(ii) Ku-ka hakkyo-eyka-nunya(-ko)mwul-ess-ta.
he-NOMschool-to

go-Q(-coMP4) ask-PAsT-DEcL

'(I) asked if he went to school.'


This would seem to requirethe postulationof a null coMP4,a phoneticallynull variantof -ko. As far as I can
tell, this nullcoMP4is only possiblewhenadjacentto the governinghigherverb, whichcouldbe accomplishedby
some selectionalfeatureon the higherverb (muchlike otherverbs select for adjacentcomplementscontaining
particular coMps); see Cho and Sells 1994.

The reviewer notes that it is also possible to postpose certain phrases, as in (iii) (see Choe 1987aand
Simon 1989for discussionsof postposingin Koreanand Japanese,respectively).
(iii) 0-ass-ta,
ku-ka.
come-PAST-DECL

he-NOM

'He came.'
As Choe and Simonshow, postposinghas the propertiesof a rightwardmovementprocess, and as such would
requirea particularconstructionalrule, under my account. Althoughthe lexical specificationof TYPE will
governthe well-formednessof the majorityof the structuresin a language,it is to be expected that there will
be particular"movement"constructionsthat producenoncanonicalconfigurationsand that requirespecific
rules.

318

PETER SELLS

inheritsupwardfrom each daughter,is graphicallydisplayedon that node and on the V0


node.
(67)

V
PRED

'read'

FORM
X-LIM

COMPI

='only'

DISC = FOC

TYPE:V-SIS

Z-LIM

[TYPE: V-SIS]

DISC =FOC_

~~~~TYPE:

-SIS

-un
V

X-LIM

[TYPE:V-SIS]

rX-LIM

\TYPE:
/TYPE:v-sIJ

'only'l
V-SIS

-man

V
PRED =
L

VI

'read'1
J

-TYPE

|FORM = COMPI

TYPE:V-SIS

ilk

-e

5.6 Prenominal Modifiers

Most of the prenominalmodifier suffixes are classified as falling in the V2 slot, since
they allow only a preceding honorific suffix. These modifier suffixes carry their own
informationabout tense; compare the examples in (68).
(68) a.

Pap-ul mek-ess-ta.
rice-AcC

b.

eat-PAST-DECL

'(Someone) ate the rice.'


mek-un
pap
eat-MoD.PASTrice

'the rice which (someone) ate'


Unlike other V2 suffixes, these precludeany and allfollowing suffixes. This follows from
the need to specify TYPE; any possible suffix that could be attached, such as the X-LIM
suffix -man, specifies [TYPE:V-SIS], but prenominalmodifiers must be [TYPE:N-SIS], because they modify nouns. This is true of both the verbalprenominalmodifierforms and
the genitive -uy, as shown in (69a) and (69b-c), respectively.

KOREAN

AND JAPANESE

(69) a. *mek-un-man

MORPHOLOGY

pap

FROM A LEXICAL

([TYPE:v-sis]

PERSPECTIVE

319

cannot modify N)

eat-MOD.PAST-only rice

'the rice which (someone) only ate'


na-man-uy ton ([TYPE:N-SIS] modifies N)
money
me-only-GEN
'money of only me'
c. *na(-uy)-man ton ([TYPE:V-SIS] cannot modify N)
money
me(-GEN)-only

b.

We thus have a unified explanationfor all of the data in this section, in terms of TYPE.
WithoutTYPE, the data are completely puzzling-there is no obvious or even remotely
plausiblecategorialsimilaritybetween a verbalform like mek-unin (69a)and the nominal
form na-uy in (69c), yet both show the same behaviorwith regardto furthersuffixation.
The other prenominalsuffix (pointed out to me by Jong-Bok Kim and a reviewer)
appears in noun complement constructions, rather than relative clauses (as in (68b)).
This falls in the V4 slot and is illustratedin (70).
(70) ka-ss-ta-nun

sosik

news
gO-PAST-DECL-MOD

'the news that (someone) went'


This suffix seems to be a formreducedfrom V-ta-koha-nunN 'N that says that(someone)
does V', though synchronicallyit functions as an independentmorpheme.43
Unlike the other prenominalmodifierforms, this one is present tense only, and it
marksonly noun complements. Thus, (71a) is well formed, but (71b) is not, since there
is no past form, and for many speakers (71c) is unacceptableunless it has an (unlikely)
relative clause interpretation.
(71) a.

sasil

ka-ss-ta-nun

gO-PAST-DECL-MOD fact

'the fact that (someone) went'


b. *ka-ss-ta-n
sasil
gO-PAST-DECL-MOD.PAST fact

c. *ka-n

sasil

gO-MOD.PAST fact

The relative clause interpretationarises in (72a), and (72b) is excluded because there is
no plausiblenoun complementinterpretation.
(72) a.

ka-n

cangmyen (V2-n)

gO-MOD.PAST scene

'the scene (in which) (someone) went'


43 Japanese

has a similar form: -to iu (sometimes

rendered

-toyuu).

320

PETER

b. *ka-ss-ta-nun

cangmyen

SELLS

(V4-nun)

scene
gO-PAST-DECL-MOD

'the scene (that says) that (someone) went'


Finally, this prenominalmodifierform can follow other V3 suffixes besides -ta, such as
the interrogativeand propositive suffixes.
phathi-eyka-nya-nun cilmwun
party-to gO-INTERR-MOD
question
'the question whether (someone) (should) go to the party'
b. phathi-eyka-ca-nun
ceyan
party-to gO-PROPOS-MOD proposal
'the proposal to go to the party'

(73) a.

5.7 Summary

In this section I have outlined a lexical approachto Korean morphology, showing the
key contributionthatthe notionof TYPEplays in constraining the syntax and in accounting
for morphologicalgeneralizations.It is this aspect that is a common threadin the structuralformationsin the morphologyand the syntax, ratherthan the configurationsand
categories in those formations.
6 Conclusion
The conclusionI wish to drawin this articleis quite simple. The hypothesis that morphologicaland syntacticstructuresare governedby the same set of principlesandconstraints
is an interestingone, and perhapsone that yields insightinto the analysis of a greatmany
languages.However, it is not necessary that this be a universalfeature of languages,as
I have arguedon the basis of data from Japanese and Korean. In these languagesit is
the lack of structuralsimilaritybetween the morphologyand the syntax that is striking
and that must be accountedfor. Structuralanalyses developed on the basis of typologically quite differentlanguages simply do not apply to Japanese and Korean, as shown
in section 4; rather, what emerges is a series of inconsistencies and paradoxes.
I should emphasize that the argumentsin this article primarilyinvolve the correct
analysisof morphology,and the interactionof that system with the phrasalsyntax. I take
this to be a separateissue fromthe issue of whetherfunctionalcategories-particularly, I
and C-exist in the syntax to define certain positions among which lexical heads may
alternate(such as the VP-internal/clause-secondpositional alternationof tensed verbs
in the Germaniclanguages).This aspect of the syntactic descriptionis perfectly compatible with a lexicalist approachto morphology-see Kroeger 1993(on Tagalog),and especially King 1993(on Russian)for illustrations.
Such positionalalternationsare very difficultto find in Japaneseand Korean,outside
of those forced only by theory-internalconsiderations.In terms of actual distributional

KOREAN

AND JAPANESE

MORPHOLOGY

FROM A LEXICAL

PERSPECTIVE

321

evidence, Whitman(1989) argues that certain kinds of Mood in Korean license topics,
and he develops the idea that topics appearin the specifierof MoodP. However, we can
accept this conclusion without having to accept the particularstructuralproposal: for
example, in the more detailed LFG account of Korean given in Cho and Sells 1994, it
would be straightforwardto constrain certain kinds of topics to cooccur with certain
kindsof Mood. As for head movementitself, Lee (1993)suggests that the verb in Korean
moves from its base position to Agr, across a certain kind of adverb; however, under
his proposal, there are no occurrences of VP that are not dominatedby AgrP, so the
movementalways happens, and no surface alternationis evident.
In the finalpartof the articleI have outlinedan analysisthattries to exploit particular
featuresof the morphologyof these languagesand that avoids the problemsfacing other
kinds of accounts. (Furtherdetails of an account specific to Korean are given in Cho
and Sells 1994.)As for the deeper similaritiesthat these languagesmighthave with other
languagesof the world, I do not take my argumentshere as denyingthat such similarities
mightexist; rather,what the present article shows is that such similaritiescannot exist
at the structural/categoriallevel and should instead be found at some more abstractlevel
of linguisticrepresentation.
References
Ahn, Hee-Don. 1988.Preliminaryremarkson KoreanNP. In Papersfrom the Sixth International
Conferenceon Korean Linguistics, 1-15. InternationalCircle of Korean Linguistics and
Departmentof East Asian Studies, University of Toronto, Toronto, Ont.
Ahn, Hee-Don. 1991. Light verbs, VP-movement,negation, and clausal architecturein Korean
and English. Doctoral dissertation,University of Wisconsin, Madison.
Ahn, Hee-Don, and Hang-JinYoon. 1989. Functionalcategories in Korean. In Harvardstudies
in Koreanlinguistics3, 79-88. Departmentof Linguistics,HarvardUniversity, Cambridge,
Mass.
Anderson, Stephen R. 1992.A-morphousmorphology.Cambridge:CambridgeUniversity Press.
Baker, Mark. 1988.Incorporation:A theoryof grammaticalfunctionchanging. Chicago:University of ChicagoPress.
Bresnan,Joan, ed. 1982. Themental representationof grammaticalrelations. Cambridge,Mass.:
MIT Press.
Cho, Dong-In. 1992.Functionalprojectionsand verb movement.In Paperspresentedat the Eighth
InternationalConferenceon KoreanLinguistics.To appear,StanfordLinguisticsAssociation, StanfordUniversity, Stanford,Calif.
Cho, Jae Ohk. 1988. Suffixed verb forms and compoundverb constructions.In Papersfrom the
SixthInternationalConferenceon KoreanLinguistics, 77-106. InternationalCircle of Korean Linguisticsand Departmentof East Asian Studies, University of Toronto, Toronto,
Ont.
Cho, Young-meeYu. 1991. A phonologicalconstrainton the attachmentof particles in Korean.
In Harvard studies in Korean linguistics 4, 37-46. Departmentof Linguistics, Harvard
University, Cambridge,Mass.
Cho, Young-meeYu, and Peter Sells. 1994. A lexical account of inflectionalsuffixes in Korean.
Ms., StanfordUniversity, Stanford,Calif. To appearin Journal of East Asian Linguistics.
Choe, Hyon Sook. 1987a.Successive-cyclic rightwardmovementin Korean. In Harvardstudies

322

PETER

SELLS

in Koreanlinguistics2, 40-56. Departmentof Linguistics,HarvardUniversity, Cambridge,


Mass.
Choe, Hyon Sook. 1987b. Syntactic adjunction,A-chain and the ECP: Multiple identical Case
constructionin Korean.In Proceedingsof NELS 17, 100-120. GLSA, Universityof Massachusetts, Amherst.
Choe, Hyon Sook. 1993. An iterative I-projectionanalysis of inflection. Ms., YeungnamUniversity.
Doctoraldissertation,University
Choi, Kiyong. 1991.A theory of syntacticXO-subcategorization.
of Washington,Seattle.
Chomsky, Noam. 1986.Barriers.Cambridge,Mass.: MIT Press.
Chomsky,Noam. 1991. Some notes on economy of derivationand representation.In Principles
and parameters in comparative grammar, ed. Robert Freidin, 417-454. Cambridge, Mass.:

MIT Press.
Chomsky,Noam. 1993.A minimalistprogramfor linguistictheory. In The viewfrom Building20:
Essays in linguistics in honor of Sylvain Bromberger, ed. Kenneth Hale and Samual Jay

Keyser, 1-52. Cambridge,Mass.: MIT Press.


Cole, Peter, GabriellaHermon,and Li-MaySung. 1990.Principlesandparametersof long-distance
reflexives. Linguistic Inquiry 21:1-22.

Dalrymple,Mary. 1993. The syntax of anaphoric binding. Stanford, Calif.: CSLI Publications.
Distributedby University of ChicagoPress.
Ernst, Thomas. 1992.The phrasestructureof Englishnegation. TheLinguisticReview 9:109-144.
Fukui, Naoki. 1986. A theory of category projectionand its applications.Doctoral dissertation,
MIT, Cambridge,Mass.
Gerdts, Donna, and Cheong Youn. 1988. Korean psych constructions:Advancementor retreat?
In CLS 24. Part 1: The GeneralSession, 155-175. ChicagoLinguisticSociety, University
of Chicago, Chicago, Ill.
Grimshaw,Jane. 1991. Extended projections. Ms., BrandeisUniversity, Waltham,Mass.
Gunji, Takao. 1987. Japanese phrase structure grammar. Dordrecht: Reidel.

Halle, Morris. 1990. An approachto morphology.In Proceedings of NELS 20, 150-184. GLSA,
Universityof Massachusetts,Amherst.
Han, Eunjoo. 1991.Honorificationin Korean. Ms., StanfordUniversity, Stanford,Calif.
Harada, S.-I. 1976. Honorifics. In Japanese generative grammar (Syntax and semantics 5), ed.

MasayoshiShibatani,499-561. New York: Academic Press.


Hong, Ki-Sun. 1990. Quantifier Float in Korean. In Proceedings of the Sixteenth Annual Meeting
of the Berkeley Linguistics Society, 175-186. Berkeley Linguistics Society, University of

California,Berkeley.
Hong, Ki-Sun. 1991. Argumentstructureand Case markingin Korean. Doctoral dissertation,
StanfordUniversity, Stanford,Calif.
Ishii, Yasuo. 1988.Head-internalrelativeclauses in Japanese.In Proceedings of the Fifth Eastern
States Conferenceon Linguistics, 234-245. Departmentof Linguistics,Ohio State University, Columbus.
Jackendoff,Ray. 1973.The base rules for prepositionalphrases. In A festschriftfor MorrisHalle,
ed. Stephen R. Anderson and Paul Kiparsky, 345-356. New York: Holt, Rinehartand
Winston.
Jung, Yeun-Jin. 1991. X-bar theory, SPECs, and directionality. In Proceedings of NELS 21,
187-201. GLSA, University of Massachusetts,Amherst.
King, Tracy H. 1993. Configuringtopic and focus in Russian. Doctoral dissertation, Stanford
University, Stanford,Calif.

KOREAN

AND JAPANESE

MORPHOLOGY

FROM A LEXICAL

PERSPECTIVE

323

Kroeger, Paul. 1993. Phrase structure and grammatical relations in Tagalog. Stanford, Calif.:

CSLI Publications.Distributedby University of ChicagoPress.


Kuno, Susumu. 1973. The structure of the Japanese language. Cambridge, Mass.: MIT Press.
Kuno, Susumu. 1983. Sin nihon-bunpoo kenkyuu (New studies in Japanese grammar). Tokyo:

Taisyukan.
Kuroda,S.-Y. 1965.Generativegrammaticalstudies in the Japaneselanguage.Doctoral dissertation, MIT, Cambridge,Mass.
Kuroda,S.-Y. 1988.Whetherwe agree or not: A comparativesyntax of Englishand Japanese. In
Papers from the Second International Workshop on Japanese Syntax, 103-143. Stanford,

Calif.: CSLI Publications.Distributedby University of Chicago Press.


Lapointe,Steven. 1991.Koreanverb markersandautolexicaltheory. To appearin the Proceedings
of FLSM 2 (University of Michigan Working Papers in Linguistics).

Lee, Jae Hong. 1993. Postverbal adverbs and verb movement in Korean. In Japanese/Korean
linguistics,vol. 2, ed. PatriciaClancy,429-446. Stanford,Calif.:CSLI Publications.Distributed by University of ChicagoPress.
Lee, Young-Suk.1990.Is INFL universal?A case study of Korean.In Proceedingsof the Seventh
Eastern States Conference on Linguistics, 204-214. Department of Linguistics, Ohio State

University, Columbus.
Lieber,Rochelle. 1980.On the organizationof the lexicon. Doctoraldissertation,MIT,Cambridge,
Mass.
Lieber, Rochelle. 1992. Deconstructing morphology: Wordformation in syntactic theory. Chicago:

University of ChicagoPress.
Martin,Samuel. 1954.Koreanmorphophonemics.Baltimore,Md.: LinguisticSociety of America.
Martin,Samuel. 1992.A referencegrammarof Korean. Rutland,Vt.: CharlesE. Tuttle.
Matsumoto,Yoshiko. 1993.Objecthonorificsin Japanese.Paperpresentedat the LSA 67thAnnual
Meeting, Los Angeles, Calif.
Park,Byung-Soo. 1988.How to handleKoreanverbalsuffixes:A UnificationGrammarapproach.
In Papers from the Sixth International Conference on Korean Linguistics, 593-600. Interna-

tional Circle of Korean Linguistics and Departmentof East Asian Studies, University of
Toronto, Toronto, Ont.
Park,Byung-Soo. 1991.The semanticand pragmaticnatureof honorificagreementin Korean:An
information-basedapproach.Ms., Kyung Hee University and CSLI, StanfordUniversity,
Stanford,Calif.
Park, Kabyong. 1990. Negation, verb movement and light verbs in Korean. In Papers from the
Seventh International Conference on Korean Linguistics, 387-398. International Circle of

Korean Linguisticsand Osaka University of Economics and Law.


Pollard,Carl,and Ivan Sag. 1994.Head-DrivenPhrase StructureGrammar.Stanford,Calif.:CSLI
Publications.Distributedby University of ChicagoPress.
Pollock,Jean-Yves. 1989.Verbmovement,UniversalGrammar,and the structureof IP. Linguistic
Inquiry 20:365-424.

Poser, William.1991.Japaneseperiphrasticverbs and noun incorporation.Ms., StanfordUniversity, Stanford,Calif.


Selkirk, Elisabeth. 1982. The syntax of words. Cambridge,Mass.: MIT Press.
Sells, Peter. 1991.Complexverbsandargumentstructuresin Korean.In Harvardstudies in Korean
linguistics4, 395-406. Departmentof Linguistics, HarvardUniversity, Cambridge,Mass.
Sells, Peter, and Young-meeYu Cho. 1991. On the distributionof XOelements in Korean. Ms.,
StanfordUniversity, Stanford,Calif.To appearin the Proceedingsof the SantaCruzWorkshop on Korean Syntax and Semantics.

324

PETER

SELLS

Sells, Peter, and Masayolida. 1991.Subjectand object honorificationin Japanese.In Proceedings


of the Seventeenth Annual Meeting of the Berkeley Linguistics Society, 312-323. Berkeley

LinguisticsSociety, University of California,Berkeley.


Simon,MutsukoEndo. 1989.An analysisof postposingin Japanese.Doctoraldissertation,University of Michigan,Ann Arbor.
Speas, Margaret.1991. Functionalheads and the MirrorPrinciple.Lingua 84:181-214.
Suzuki, Tatsuya. 1989. A syntactic analysis of an honorificconstructiono . . . -ni naru underthe
DP hypothesis:Towarda unifiedtheoryof honorification.In Proceedingsof the EighthWest
Coast Conference on Formal Linguistics, 373-383. Stanford, Calif.: CSLI Publications.

Distributedby University of ChicagoPress.


Tateishi,Koichi. 1988.On the universalityof X-bartheory:The case of Japanese.In Proceedings
of the Seventh West Coast Conference on Formal Linguistics, 331-345. Stanford, Calif.:

CSLI Publications.Distributedby University of ChicagoPress.


Tateishi,Koichi. 1989.Subjects,SPEC, andDP in Japanese.In Proceedingsof NELS 19, 405-418.
GLSA, University of Massachusetts,Amherst.
Tonoike, Shigeo. 1991.The comparativesyntax of English and Japanese:Relatingunrelatedlanguages. In Current English linguistics in Japan, ed. Heizo Nakajima, 455-506. Berlin: Mou-

ton de Gruyter.
Toribio, AlmeidaJacqueline. 1991. Specifier-headagreementin Japanese. In Proceedings of the
Ninth West Coast Conference on Formal Linguistics, 535-548. Stanford, Calif.: CSLI Publi-

cations. Distributedby University of Chicago Press.


Urushibara,Saeko. 1991. EylEykey: A postposition or a Case marker?In Harvard studies in
Koreanlinguistics4, 421-431. Departmentof Linguistics,HarvardUniversity, Cambridge,
Mass.
Whitman,John. 1989.Topic, modality,and IP structure.In Harvardstudies in Koreanlinguistics
3, 341-356. Departmentof Linguistics, HarvardUniversity, Cambridge,Mass.
Whitman,John, and S. Hahn. 1988. Korean morphologicalpassive/causatives. In Papers from
the Sixth International Conference on Korean Linguistics, 714-728. International Circle of

KoreanLinguisticsandDepartmentof East Asian Studies, Universityof Toronto,Toronto,


Ont.
Whitman,John, KWee-OckLee, and BarbaraLust. 1991.Continuityof the principlesof Universal
Grammarin first languageacquisition:The issue of functionalcategories. In Proceedings
of NELS 21, 383-397. GLSA, University of Massachusetts,Amherst.
Williams, Edwin. 1981. On the notions "lexically related" and "head of a word." Linguistic
Inquiry 12:245-274.
Yang, In-Seok. 1972. Korean syntax: Case marking, delimiters, complementation and relativiza-

tion. Seoul: Paek Hap Sa.


Yoon, James Hye Suk. 1991a.Inflectionalstructuresin Korean and headedness. Studies in the
Linguistic Sciences 21:179-192.

Yoon, James Hye Suk. 1991b. Korean nominalizationsand morphosyntacticinteraction. Ms.,


Universityof Illinois, Urbana-Champaign.
Yoon, JamesHye Suk. 1993.Tense, coordination,andthe clausalstructureof Englishand Korean.
Ms., Universityof Illinois, Urbana-Champaign.
Presentedat the 5th HarvardInternational
Symposiumon Korean Linguistics, HarvardUniversity, Cambridge,Mass.
Yoon, James Hye Suk, and JeongmeYoon. 1990.Morphosyntacticmismatchesand the functioncontent distinction. In CLS 26. Vol. 1: The main session, 453-467. Chicago Linguistic
Society, University of Chicago, Chicago, Ill.
Yoon, James Hye Suk, and Jeongme Yoon. 1991. Chain Condition, ambiguityof government,

KOREAN

AND JAPANESE

MORPHOLOGY

FROM A LEXICAL

PERSPECTIVE

325

and derivationalgrammars.In Proceedings of NELS 21, 415-429. GLSA, University of


Massachusetts,Amherst.
Yoon, Jeongme. 1990. Verb movement and the structureof IP in Korean. Language Research
26:343-371.
Yun, Sung-Kyu.1991.Topic-orientedhonorificsi in Korean.In Harvardstudies in Koreanlinguistics 4, 563-575. Departmentof Linguistics, HarvardUniversity, Cambridge,Mass.
Department of Linguistics
Stanford University
Stanford, California 94305
sells@csli.stanford.edu

Anda mungkin juga menyukai