Korean (and Japanese) morphology from a syntactic perspective

Hilda Koopman, UCLA July 2004

[email protected]

1. Introduction.

In Korean and Japanese Morphology from a lexical perspective, Sells (1995) argues against the

view that complex words in Japanese or Korean like (1b) are derived from the underlying

syntactic structure in (1c) by syntactic head movement (Sells 1995: 280)*:

(1) a. Mary -ga oyog- ana- katta- to Japanese

Mary-NOM swim-NEG- PAST-C

b. C c. CP
3 3
T C TP C
3 to 3 to
Neg T DP 3
3 katta Mary-ga NegP T
V Neg 3 katta
oyag ana VP Neg
4 ana
oyog

Sells establishes that the complex word in (1b) does not contain any pronounced arguments of V,

and argues against the syntactic view in (1), on the grounds that it leads to expectations that are

not met. In particular, if the inflectional morphemes are heads, they should behave as heads: they

should not be transparent for selection, and they should determine the category of the complex

word. Sells shows that certain inflectional morphemes are transparent for locality of selection,

and that the leftmost element in Korean and Japanese ‘words’ shows head-like behavior in that it

determines the categorical feature of the complex word. Sells concludes that the properties of

these words should not receive a syntactic treatment (as Sells states ‘it is the lack of structural

similarity between the morphology and the syntax that is striking’ (p. 320) ). Sells outlines a

1
lexicalist account for the data which takes the templatic nature of the morphology as basic (in

particular, quite different syntactic elements can occupy the same slot in a template). Strictly

morphological principles regulate the flow of information within the word, a task normally

performed by X-bar theory or the theory of Projection. Sells’ account remains sketchy, and I will

not address it here: rather I will concentrate on Sells’ arguments against the syntactic view that

words are build in the syntax, and develop a syntactic account which yields a parsimonious

account of the properties of “morphological units”: neither templatic morphology nor special

morphological principles are necessary. The properties of morphological words follow from

regular syntactic principles, (the theory of projection, agreement and movement), in conjunction

with (idiosyncratic) phonological properties of the affixes. I will agree with Sells that the

syntactic structure in (1c) does not underly the inflected words, and that the words are not derived

by head movenent: instead, I will pursue the idea that the complex words in question are derived

by phrasal movement from merged head complement structures (Kayne 1994:52-53).

This reply is organized as follows. Section 2 discusses the basic surface constituency of Korean

(and Japanese) sentences in very general terms. I will make the basic case for phrasal (remnant)

movement, based on the phrasal affix nature of the affixes (Yoon 1994, among others) and

ellipsis (Otani and Whitman 1991), and show how this view reconciles the two mutually

incompatible analyses found in the literature (the V movement approach (Otani and Whitman

1991, Koizumi, 1995, 2000) and the V-in-situ approach (Yoon, 1994, Fukui and Sakai 2003).

This analysis undermines the evidence for an underlying OV order. Section 3 spells out some of

the relevant syntactic assumptions in which this reply is couched, and lays out what we should

expect to find. Section 4 turns to the particular cases Sells discusses, motivates syntactic analyses

for them, going beyond Sells discussions. Section 5 extends the analysis to other problems of

Korean morphosyntax, including the relative ordering of the focus particle man ‘only’, and the

structural case markers, and two types of subject agreement in Korean, honorific agreement and

2
plural agreement. Section 6 turns to a discussion of the interaction between word structure and

scope (Lee 2004) and section 7 concludes.

2. Surface constituency.

Let us start with a brief discussion of the surface constituency of a Korean (and Japanese)

sentence, paying particular attention to the position of the verb and the behavior of inflectional

morphology. In a simple OV sentence in Korean, the verb does not form a surface constituent

with any of its overt arguments, as Sells shows with phonological arguments. This can also be

shown by the placement of short negation an, which is (minimally) adjoined to VP (Whitman

2003, Hagström 2003).

(2) Chelswu-ka ppang-ul an mek-ess-ta.

Chelswu-NOM bread-ACC not eat-PAST-DECL

‘Chelswu did not eat the bread.’

(3) wuncen an ha- ta

drive not do-DECL

‘I didn’t drive’

Korean and Japanese are agglutinative languages, with the order of morphemes reflecting the

syntactic hierarchy. These morphemes are phrasal affixes (Yoon 1994), which allow their

dependents to be coordinated1, as we will see throughout the paper.

(4) John-i pap-ul cis-ko kwuk-ul kkulhi-ess-ta. (Yoon 1994)

J-NOM rice-ACC cook-CONJ soup-ACC boil-PAST-DECL

'John cooked the rice and made the soup.'

None of the arguments given in Chomsky (1970) for a lexical treatment apply, nor is there any

empirical evidence for affix hopping, or Marantz’s PF operation of Merger (Marantz 1988, Halle

and Marantz 1993). Thus, inflectional morphemes spell out the corresponding syntactic head

positions.

3
Since the verb and the inflectional morphemes form a phonological constituent, the question

arises if this surface phonological constituent is formed pre-spell out or not. Some linguists argue

the V remains in-situ in the syntax, and forms a constituent after ‘rebracketing’ or Marantz’s

‘Merger’ in the phonology (Yoon 1994, Fukui and Saka 2003), others argue the V raises overtly

pre spell-out on the basis of ellipsis and coordination (Otani and Whitman 1991, Koizumi 2000).

Otani and Whitman (1991) argue that the null object construction in Japanese and Korean should

be analyzed as VP ellipsis, with the verb outside the elided constituent and therefore pronounced.

Although Hoji (1995) shows that there are problems with this analysis concerning the

interpretation of null objects, elliptical answers to yes-no questions (McCloskey 1991) provide

further independent evidence that V and affixes are outside the elided constituent.

(5) romia-ka ejeo culiet-kwa kicang-e kass-ss-ni? (Fromkin, 2000:317)

Romeo.NOM yesterday Juliet.with movie-LOC go-PAST-Q

‘Did Romeo go to the theater with Juliet yesterday?

(6) ing, ka-ss-ta-

yes, go-PAST-DECL(informal) (he go to the movies with Juliet yesterday)

“yes, he did ( go to the movies with Juliet yesterday)”

(7) ani, an ka-ss-

No, not go-PAST-DECL(informal) (he go to the movies with Juliet yesterday)

No, he didn’t (go to the movies with Juliet yesterday)

The same holds for Japanese (Fromkin 2000:317):

(8) Romia-wa tamanegi-o kay-ni kodomotachi-o maaketto-ni ik-ase-ta-ka

Romeo-TOP onion- ACC buy-to children -ACC market-to go-CAUS.PAST-Q

“Did Romeo send children to the market to buy onions?”

4
(9) Hai, ik-ase-ta

Yes go-caus-PAST (lit: yes, send)

Yes, he did [he send them to the market to buy onions]

Under the reasonable hypothesis that we are dealing with an elided constituent, it follows that the

verb must be outside the elided constituent when ellipsis applies, i.e. the surface position of the

verb must be the result of pre-spell-out movement, consistent with (1) and Sells arguments.

Furthermore, since the elided constituent can contain high adjuncts, the verb must have raised

high into the clausal structure, at least to the polarity head Sigma (Laka, 1991).

This raises the question how the verb gets there. A head movement account runs into the problem

that inflectional morphemes behave like phrasal affixes: this is unexpected, since adjoined heads

in general do not show phrasal characteristics ((*to [[red]and [white]]en, *[[use] and [destroy]]-

ed 2). We thus need an account that reconciles the phrasal affix nature with overt syntactic

movement, and that is what phrasal movement to Spec achieves, as we can see in English for

example, where phrasal affixes follow their dependents in Spec (John’s brother, I’ll go etc), and

allow their dependents to be coordinated: ([John and Mary]’s brothers, [John and Mary]’ll go.

The verb then must be within a phrasal constituent attracted to the Spec of the affix:

(10) [TP[VP .. V..] [T’[T…..

This analysis reconciles the two competing syntactic analyses for Korean and Japanese: the V is

in-situ (Yoon 1995, Fukui and Saka 2003), since it is in the VP. And the V appears to have

moved (Koizumi (1995, 2000), Otani and Whitman (1991)), because the (remnant) phrase that

contains it has undergone movement to Spec, ending up in a Spec position high up in the clause. 3

Note that (10) undermines the basic evidence that Japanese and Korean are head final languages

underlyingly, since this apparent head complement structure is in fact a surface Spec head

configuration (Kayne 1994). Furthermore, if the V is within a remnant VP in Spec, TP,

5
complements of V must have moved leftward, either individually, or as a remnant constituent

(Whitman 2003, Hagstrom, 2003) regardless of the issue whether the verb is underlying head

final or head initial. The difference between a traditional head final analysis and a Kaynian

antisymmetric one thus becomes minimal. The remaining discussion in this paper is cast in terms

of antisymmetry: as I will show the merged head complement order surfaces in certain instances,

even in these ‘strict’ head final languages.

3. Expectations under the syntactic view.

How can the phrasal movement approach explain the properties of complex words in Korean and

Japanese Sells that describes? Let us briefly examine the expectations and predictions of the

phrasal movement approach to complex words.

3.1. Specifiers

Specifiers are to the left of heads, and attract phrases. Phrases in Spec can be attracted to higher

Spec positions, as is the case with NP-movement, and successive cyclic wh-movement.

(11) a. Johni seems Johni to be Johni likely Johni to Johni become Johni an artist

b. whoi do you think whoi that Mary wrote an e-mail to whoi

Given the phrasal movement view, we expect to find similar cases in morphology. Thus, with H1

merged higher than H2 (H1>H2) , we should find cases that result in the linear order XP H1 H2

by extraction of XP stranding H2, where XP H1 and H2 form a phonological “word”.

(12) H1P
3
XP 3
H1 H2P
3
XP 3
H2

linear order: XP H1 H2

6
Importantly, the linear order is the same as the order of merger: H1 is merged higher than H2,

takes H2P as a righthand sister, and precedes H2 at spell-out. Notice that the linear order in this

derivation does not “mirror” the syntactic hierarchy, in the sense that the inner affix is not merged

closer to X than the outer affix. In this sense, the example in (12) violates Baker’s Mirror

Principle (Baker 1985). As I will show, the configuration in (12) surfaces in some cases in

Japanese and Korean, thus revealing the head initial merged structure, even though the

morphology of Korean and Japanese strongly mirrors the syntactic hierarchy in (H1>H2>XP is

mirrored as XP]H2]H1), which arises through pied-piping (see (14) below).

3.2. Heads.

Heads, regardless of whether they are functional or lexical, project and determine the category

and properties of their projection. Thus, the category of a tensed verb is T, not V. It is widely

assumed that heads select for their complements, and satisfy selection under first merge. Heads

impose restrictions on their Spec as well, and I refer to this as Spec selection: the EPP feature is a

particular instance of Spec selection. Phrasal affixes can combine an EPP feature with a further

phonological requirement: thus a phrasal affix can attract some particular phrase to its Spec,

where its EPP feature is satisfied. If that phrase undergoes further movement, it can be stranded,

provided the phrasal affix can lean on some element within its phonological phrase. This will

yield the restricted and very local types of movements of (12) above.

3.3. Pied-piping and Spec head agreement.

Attracted elements in Spec positions find themselves in the canonical pied-piping triggering

configuration (Webelhuth 1992): a wh-feature in an embedded Spec can ‘percolate’ up to the

containing DP node, enabling the containing DP to check the wh-feature.

(13) [ [[whose brother] ‘s friend]’s car] did you borrow

7
Koopman (1996) and Koopman and Szabolsci (2000) propose that ‘percolation’ results from

cyclic applications of Spec head agreement4. Agreement copies5 some feature from a Spec onto

the head. If a head H acquires a feature through agreement, the HP will carry that feature, by

projection: properties of heads determine the properties of the projection as a whole.

If complex words are formed by phrasal movement, similar instances of pied-piping and

agreement are expected: in particular some property of a phrase in an embedded Spec should be

‘visible’ on the containing category.

(14) H1P<x>
3
H2P<x> 3
3 H1<x>
XP 3
H2 <x>
linear order: 3 2 1

cyclic agreement for feature x results in x being visible on H1P

Pied-piping and agreement will account for Sells’ observation that the leftmost element displays

head like behavior in Japanese and Korean. Section 5 extends agreement and pied-piping to other

cases of agreement in Korean.

3.4. Locality of Selection: Sportiche

In recent work, Sportiche (1998, 2000 class lectures, 2001) has argued for a strict enforcement of

The Principle of Locality of Selection: selection must be satisfied in a strictly local relation.

Apparent violations of locality of selection abound as a result of movement, as the simple

example in (15) shows:

(15) who did Mary see who

The wh-phrase is selected by V, but is not in a local configuration with its selector at spell-out.

Local selection of course holds pre-movement, i.e. there is a stage in the derivation in which the

selector and selectee are in a local configuration. Note that the wh-C also “selects” for a +wh

8
phrase. This selection, commonly referred to as the EPP property, is locally satisfied after wh-

movement.

As Sportiche shows, strictly enforcing the principle of locality of selection has far reaching

implications for syntax. For example, if V selects for NP (and not for DP) as Sportiche argues, the

standard view that V merges with DP cannot be correct. Instead V must merge with NP first, in

accordance with locality of selection, and D attracts NP through movement, i.e. D’s selection for

NP is locally satisfied after movement:

(16) D [V NP]

Sportiche is led to the view in (16) on the basis of reconstruction (Sportiche 1999)6. The Principle

of Locality of selection provides the rationale for why that state of affairs must hold. This view

converges with Kayne’s proposals that many traditional constituents are not constituents, but are

remnants within a larger (remnant) constituent formed by attraction and movement (Kayne 2000,

2003). The Principle of Locality of Selection accounts for Sells’ cases of violations of locality of

selection (section 4). It forces particular analyses, and makes further predictions, that can be

empirically tested.

4. Korean and Japanese morphology: a syntactic view.

Sells presents two types of arguments against what he calls the syntactic view of morphology, i.e.

the view that “morphological structures and syntactic structures are governed by the same set of

principles and constraints p. 320”. The first type is based on violations of locality of selection.

Rightmost inflectional affixes in general do not yield violations of locality of selection, and

therefore do not behave as heads are expected to behave. The second type of argument shows that

the leftmost element must be relevant for selection in certain cases, and thus shows unexpected

head-like behavior. Somewhat unrelated to the above, Sells argues against universal hierarchies,

and defends a templatic view of the morphology of Korean and Japanese.

9
4.1. Selection before movement

Sells points out that there is a very general and systematic problem with locality of selection, and

illustrates this with the following general example, where a delimiting particle kkaci and the

focus marker nun intervene between the verb and the dative suffix selected by the verb:

(17) Swuni-hanthey -kkaci-nun cwu-ess-ta (Sells 1995 (15): 285).

Sooni- DAT -even- FOC give-PAST-DECL

‘I gave it even to Sooni’

As Sells points out, locality of selection is systematically violated in this type of example, a well-

known problem that arises with the incorporation of functional heads in syntactic structures

(Grimshaw 1991). Sells argues that this problem would not occur if the inflectional particles were

simply non-heads. Sportiche’s Principle of Locality of Selection provides a different answer to

this problem: these particles are heads indeed, but at the point in the derivation where selection is

locally satisfied, they have not been merged yet. They are merged at a later point in the

derivation, and attract the focused constituent to their Spec (Kayne 1998). A simplified derivation

for (17) illustrates how this type of analysis works. Other cases of selection before movement are

discussed in section 4.2. We start the derivation at the point where the arguments have been

merged. The structure is simplified, and only considers the indirect object, V, T, and Topic (nun).

(18) cwu ‘give’ swuni ‘Sooni’

a. merge Phanthey move DP → [swuni [hanthey [cwu
b. merge X move VP → [VPcwu x [PPswuni hanthey
c. merge kkaci move remnant PP→
[[PPswuni hanthey [kkaci [VPcwu
7
d. merge T move VP →
[ [VPcwu[iss [[PP swuni hanthey [kkaci
e. merge nun move PP
[[PP swuni hanthey [kkaci [ nun [ [VPcwu[iss

Arguably, selection of V by P is satisfied at the point of merger in (18a). The movements in (18a)

(18c) and (18d) are forced by locality of selection, with selection satisfied after movement

(‘remerge’). The step in (18b) creates the remnant PP necessary for (18e) and looks exactly like

10
Kayne’s (1998, 2000) VP-movement to WP8. Müller (2000) argues that this movement is

motivated by the Principle of Shape Conservation (Williams 2003) (V>O), and hence this step is

not expected to take place in surface OV languages. If the derivation above is on the right track,

however, even in a rigid OV languages this step must be crucially involved in deriving a remnant

constituent that contains PP, so at to allow the “remnant VP ” to raise high into the structure, and

the PP to combine with the Topic marker in the left periphery.

4.1.1 An aside about the relative order kkaci/man and P .

The delimiting particles kkaci ‘even, up to’ and man ‘only’ can either precede or follow the P

(John Whitman, personal communication). Nun must be final, pointing to higher merger of nun.

(19) a. Swuni- hantley- kkaci nun

Swnui to even TOP

b. Swuni- kkaci hantley nun

Swuni even to TOP

c. *Swuni- kkaci nun hantley/ * hantey kkaci nun

Swuni even TOP to

There are in principle two options for analyzing the variable order in the adopted syntactic

framework. Either the relative order of merger between P and man/kkaci is free, in which case the

two linear orders reflect two different hierarchies, in accordance with the Mirror Principle:

(20) Two merged orders:

a. kkaci> P> DP Æ [[DP P] kkaci]

b. P>kkaci>DP Æ [DP kkaci] P]

Or the order of merger is fixed man/kkaci> P, and the derivation is responsible for the different

surface orders, either by pied-piping, or extraction from Spec9.

(21) kkaci> P> DP

a. move DP to Spec, PP, merge kkaci: kkaci [ [PPDP P ….[

11
choose b. or c:

b. pied-pipe PP [[PPDP P .. [kkaci [DP P

c. Spec extraction: [[DPDP [kkaci[PPDP P…

Coordination supports (21) as the only option. All and only the expected patterns of coordination

are attested given the analysis in (21). The coordination pattern predicted by (20b) is excluded:

(22) * [DP kkaci] & [DP kkaci] P] nun]

*swuni kkaci wa mira kkaci hanthey nun

Sooni even and Mira even to FOC

This surface string is not derivable from (21), since the DP and kkaci do not form a constituent.

The pied-piping derivation in (20b) underlies the patterns in (23):

(23) a. [ [PP] & [PP ] kkaci nun

swuni hanthey wa mira hanthey kkaci nun

Sooni to and Mira to kkaci FOC

b. [DP&DP] P kkaci

swuni wa mira hanthey kkaci (nun)

Sooni and Mira to kkaci FOC

c. [[PP] kkaci] & [[PP] kkaci] (nun)

swuni hanthey kkaci wa mira hantey kkaci nun

And the stranding derivation predicts the patterns in (24):

(24) [DP & DP ] kkaci hanthey nun

[ swuni wa mira] kkaci hanthey nun

[DP kkaci hantley] & [DP kkaci hanthey] nun

swuni kkaci hantley wa mira kkaci hanthey nun

Therefore, the unique hierarchy (21) (kkaci>P) underlies the linear orders. This is in accordance

with general principles of clausal structure, which force modifiers of Ps, or Focus to be merged in

12
a higher region than dative P (hence the principled unavailability of the merged order P>focus).

The fact that P can follow kkaci follows from phrasal movement from the Spec position and

reveals the underlying head complement order, in support of Kayne (1994).

4.2. Complementizer selection.

Sells presents complementizer selection as an example where local selection in word structure is

interrupted. Korean has different types of non-tensed verb endings, and their distribution is

determined by selecting verbs. Sells, following Cho and Sells 1995, glosses these as Comp1,

Comp2, Comp3, etc. I will gloss them by their spell-out forms, and return to their distribution in

4.5. I follow Sells and analyze ahn as a negative auxiliary (Sells:305) selecting for a clausal

constituent headed by -ci.

(25) po ‘ try’ selects for C1 [- e]

anh, ‘neg V ’ selects for C2 [-ci ] .

siph, ‘desire’ selects for C3 [ -ko]

Inflectional elements can intervene between C and the selecting V, leading to apparent non-local

selection:

(26) ilk- e- man-un po-ass- ta (Sells, 19b)

read-e-only-FOC try-PAST-DECL

‘tried only reading’

The violation of locality is not an argument against the syntactic account under consideration.

Indeed, the Principle of Locality of Selection forces an account where selection is satisfied

locally before movement to higher merged heads takes place, i.e. local selection of the C by the

verb try must precede merger of man and movement to the specifier of man:

(27) hierarchy: man > try>-e>read.

“VP”(read) moves to -e which (spec-selects) for a small constituent that I will simply abbreviate

as “VP” (it is at least as big as VoiceP (see section 4.4. ). -e is locally selected by try. In order to

13
make this derivation precise, and examine further predictions, an additional piece of information

is important. As Sells (1998) shows, VP-e obligatorily forms a complex verb with try. This can

be shown by short negation an which must precede the verbal complex, and negates try (hence

Neg(an)> try: (dependents of ‘read’ must precede short negation an)

(28) an ilk-e po-ass-ta

not read-e try-past-decl

‘I didn’t try to read ’

Verbal complexes can become quite big, and resemble Hungarian or West Germanic verbal

complexes, with each verb carrying its own inflectional morphology (Koopman and Szabolcsi

2000):

(29) an ilk-e po-ko siph-ta

not read-e try-ko want-decl

'didn’t want to try to read'

Verbal nouns do not form a surface complex verb, even though these form a tight unit with the

light verb:

(30) a. wuncen an ha-ass-ta a. wuncen an ha-e po-ass-ta

drive not do-past-decl drive not do-e try-past-decl

‘I didn’t drive’ ‘didn’t try to drive’

Thus, V-e forms a surface constituent that does not contain the phonological form of its selectee,

i.e. it is either an incorporated head, or, rather, anticipating the fact that it can undergo further

phrasal movement, a phrasal remnant. Koopman and Szabolcsi (2000) propose that there is a

universal set of complex verb formators, which must always enter into a local relation with a

small clause predicate (a constituent slightly larger than VP, called VP+). Complex verb

formation is achieved in the following configuration, which transparently fits the Korean verbal

complexes, with VP+ further embedded in inflectional layers:

(31) [VP+ [[VP+ Vread ]- e] [Vtry

14
In this configuration ‘try’ checks what it needs for complex verb formation (VP+) , and what it

needs morphologically (-e) under first merge (try > -e) . VP(+) movement to –e is forced, because

of the nature of the phrasal affix –e which is very specific about the category that it needs to

attach to. VP+ will pied-pipe –e to the position where a complex verb with ‘try’ is formed,

yielding the following structure:

(32) VP+try
3
eP 3
3 try
VP+read t
ilk e

eP must be a remnant, i.e. arguments and modifiers of this constituent must precede the short

negation an which modifies try. (see Koopman and Szabolcsi for general discussion of

‘restructuring’ of this type).10

(33) swuni-ka yenge-lul cal an ilk-e po-ass-ta

suni-NOM English-ACC well not read-e try-PAST-DECL

‘Suni didn't try to read the English passage carefully’

We are now ready to return to Sells’ observation that man (‘only’) can intervene between

the –e constituent and the selecting verb (26). This follows from the hierarchy man>

VP+> try >-e>read, complex verb formation and phrasal movement of the focused

constituent to man. As the reader can check all selectional relations are locally satisfied

either before or after movement.

15
(34) manP
3
eP 3
3 man VP+try
VP+read t 3
ilk e eP VP
6 3
ilk-e V
try

4.3. Interactions between Focus (man), short negation (an) and verbal complexes.

Next, we consider further predictions given the hierarchies in (27) and (35b):

(35) a. man > try > -e>read (ilk-e-man ilk-e po ilk-e(-ass-ta)

b. an > vp+ > try>-e>read (an ilk-e po ilk-e (-ass-ta)

As a first step, we establish the relative order of merger between man and an. The following

example shows that man ‘only’ scopes over an (Neg) in simple clausal structures, establishing the

hierarchy of merger as Focus(man)>Negation (an).

(36) wuncen-man an ha-ta

drive -only not do-DECL

‘I did everything but driving’ ‘It is only driving that I didn’t do’ (only>neg)

‘* It not the case that I did only driving and nothing else’ (*neg>only )

Since the focused predicate ilk-e raises to Spec, man (37), it should precede an, and scope over

negation:

(37) man > an >ilk-e > po

This prediction is borne out as the following example shows11:

(38) a. *an ilk-e- man po-ass-ta

not read-e only try-PAST-DECL

b. ilk-e man an po-ass-ta

16
 read-e only not try-PAST-DECL

‘I tried everything but reading’ only>not

‘It is only reading that I didn’t try’

These data fall out from the underlying the hierarchy of merger, and phrasal movements.

Complex verb formation is obligatory. Certain inflectional morphemes are allowed within verbal

compounding, because they are locally selected by the complex verb formator, and their

selectional properties are satisfied locally before the complex verb is formed. Other inflectional

morphemes are not allowed within the verbal complex (-man, -nun) because these can only be

merged at a later point in the derivation accordance with universal principles that guide the make-

up of clausal structure. Movement is forced and ubiquitous, because of locality of selection.

Finally, scope is determined by the hierarchy of merger, which determines the location of the

scope bearing heads.

4.4. Selection after movement.

Sells second type of argument is based on cases where the leftmost element shows head-like

behavior, and is the element which an outside selector needs access to. Sells illustrates this with

Japanese gerunds, and with Japanese speech level particles.

4.4.1. Japanese gerunds

Japanese gerunds can be adjectival or verbal:

(39) a. tabe- te

eat- GER

b. tabe –nai- de (Verbal gerund)

eat- Neg.cop- GER

17
 c. tabe-naku-te (Adjectival gerund)

eat Neg-GER

Only the verbal gerunds (39a, b) can be selected by a V like iota ‘to put/to prepare for some

future eventuality’ (Sells 1995: 21a)

(40) ziroo-wa zenbu-no tabemono-o tabe-nai-de oit- ta /*tabe-naku-te poita

Ziroo-TOP all-GEN food- ACC eat-neg.cop-GER‘put’-PAST

‘Ziro made the provision of not eating all the food’

The category of the gerund must be determined by the V/A that it contains, which is merged

lower than the negative copula. This configuration parallels the well-known case of pied-piping

where a feature embedded in a specifier is able to satisfy an outside selector, i.e. this is a case

where selection is not satisfied at the point of merger, but after movement, through spec head

agreement in category. Agreement in category has a phonological reflex: the spell-out of Neg and

the copula covaries with the A/V features in its Spec, as does the form of the Formal level affix.

This leads minimally to the following structures and derivations, with agreement indicated:

(41) a. Verbal gerund

teP<+V>
3
Neg.copP<+V> t
3 de<+V>
VP 3
4 nai<+V>
tabe[+V]

18
 b. Adjectival gerund
teP<+A>
3
Neg.cop<+A> t
3 te<+A>
AP 3
2 naku<+A>
4 t
tabe A

Cyclic application of Spec head agreement for category and projection carry the categorical

feature up to the gerund, thus allowing the verb to satisfy its selection of the gerund locally.

Agreement in category has as effect that the leftmost category appears to determine the category

of the constituent.

4.4.2. Speech level.

Sells argues that selection has to look deep into a constituent for the form and ordering of the

inflectional morpheme that indicates speech level in Japanese. Interestingly, this selection is also

sensitive to category (V versus A).

The formal speech level desu occurs with A, but not with V:

(42) a. *tabe-ta- desu

eat- PAST- FORMAL

b. aka-katta-desu

red- PAST- FORMAL

The formal speech level mas- occurs with verbs, but not with As. Furthermore, mas- precedes,

while desu, follows, T:

(43) a. tabe-masi-ta V-masi-T

eat- FORMAL- PAST

b. *tabe-ta- masi

19
 eat- PAST-FORMAL

The dependency of the form of speech level on category (desu with AP, mas with VP), is

analyzable in two ways under the syntactic view: selection is satisfied under first merge, or

selection is satisfied after movement through Spec head agreement (44b): 12

(44) a. selection is satisfied under first merge (T>desu >AP T>mas>VP ),

b. selection is satisfied after movement (desu> T> AP , mas>T>V).

The hierarchical order of merger would surface because of Spec extraction, the mirror order

because of pied-piping (“roll-ups”):

(45) T > formal> AP/VP

a. [AP [desu AP to Spec,Formal

b. [AP [katta [AP [desu Spec- extraction

a’ [VP [mas VP to Spec Formal

b’ [VP masi] ta pied-piping

(46) Formal> T>VP/AP

a [VP[ ta.. VP to Spec TP

b [VP [masi [VP [ta.. Spec-extraction

a’ AP katta AP to Spec T

b’ [AP katta] desu pied-piping

It seems to me there is a strong bias for (46). Formal speech level is in complementary

distribution with Force/mood markers, and restricted to root contexts. This suggests Formal is

located at the root level, higher than T. If correct, it must be explained why VP must extract from

TP in (46a), but AP cannot do so. In Koopman and Szabolcsi 2000, and Koopman 2002 this is

achieved by means of so-called complexity filters, which care about the structure at spell-out.

Only phrases of a certain size, category and phonological shape (i.e. with V spelled out) “ fit” in

20
Spec, masi, where size is determined by the most deeply embedded spelled-out element. This is

an idiosyncratic part of the lexical entry of masi.

(47) Complexity filter on Spec, masi: overt material in V cannot be more deeply embedded
than:
3
3 3
3 masi
V

VP-ta is simply too big to fit into the Spec, masi, as the following structure shows, where TP

further embeds the VP:

(48) 3*
TP 3
3 masi
VP 3
3 ta
3
V

Hence the obligatory stranding of T in (46a). Spec-extraction is a way to keep the representations

that the phonology accesses small. Note that this presupposes that phonological spell out must

directly access the hierarchical syntactic structure. The form of Formal speech level covaries

with the category in Spec: this is an overt reflection of Spec head agreement (in category), as

argued in 3.2.1.13 If this analysis is correct, we have identified another case of Spec extraction

within the Japanese “word”, where the merged order surfaces because of Spec extraction.

4.5. Arguments against universal hierarchies?

Sells argues against the syntactic view in part because it presupposes universal hierarchies, or

principles that determine the underlying order of merger, a view he assumes to be problematic.

The verbal (and nominal) morphology of Korean is usually presented as a template, in which

there are a number of morphological slots available after the verb or the noun. Different non-

21
finite verb endings (his C1, C2, C3, C4)) appear to occupy these slots, even though they have

nothing in common with the other elements that can occupy these slots:

(49) Vroot –Honorific- tense-Mood-Discourse

ilk- usi- ess- ta- ko
read
V 1 2 3 4
[ C1] (e/a)
[ C2] (ci/key/ko)
[ C3] (eya/aya and na )
[ C4] (ko )

Thus, C1 appears in the same slot as Honorific, and can only follow a verbal root, C2 appears in

the same slot as Tense, and can only be preceded by a Honorific suffix, C3 by Honorific and

Tense, and C4 by Honorific, Tense and Mood, etc. This, Sells argues, is exactly what one expects

under a theoretical approach that includes morphological templates, but not under a view that

includes syntactic hierarchies, since this would not allow a unique location of C. Under a

templatic view, all slots should be able to be filled: elements that fill a particular slot do not have

to form a natural class: there is no reason to find a unique position for C in Korean, nor is there

any a-priori reason to assume C and honorific cannot occupy the same slot.

The syntactic view in fact does not lead one to expect a unique syntactic position for a non-finite

verb ending or a subordinator (i.e. C), as much as we don’t expect English, to, -ing, for, that, etc

to occupy exactly the same syntactic position (i.e. to be merged at the same height), or English –

ing to be mergeable in a unique position in the hierarchy: as is well-known –ing can be merged at

different heights in the hierarchy, leading to different types of –ing clauses, with different

distributional properties. Even if the different Cs were all merged in the same position, the

patterns in (49) would still be precisely what one expects to find under a syntactic view:

individual heads can select for different pieces of the hierarchy, thus C1 selects for VoiceP/VP as

a EPP property, C2 for a constituent that includes Agrhon, C3 for TP etc. Since these Cs are final,

selection must minimally be satisfied after movement to their Spec. This is the only way in which

22
the syntactic view can derive the appearance that elements that do not form natural classes seem

to occupy the same slot. Sells further takes the ordering facts discussed in the previous section as

an argument against universal hierarchies.

“ However, to work correctly, this part of the syntactic view presupposes

that there is a consistent hierarchy of functional categories, 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 universally, or 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……” Sells, 1995, p.297.
“ For the syntactic view, the facts are even more puzzling, since even within the same
language there may be no consistent hierarchy” Sells 1995 p 299.

That different linear orders can be derived from a single underlying hierarchy, is not problematic,

but expected under the syntactic view, as we have shown in the previous sections. Finally,

evidence for underlying hierarchies, or principles that underly clausal hierarchies seems even

stronger now than at the time when Sells published his article (Cinque 1999, Rizzi 1997, 2000,

Pollock and Poletto 2003, Hallman 1998, Koopman and Szabolcsi 2000, Julien 2002, Williams

2003, among others).

5. Case and agreement

In order to show that the independently needed syntactic mechanisms introduced so far account

for further properties of Korean morphology. In 5.1, I argue that the surprising position of

structural case markers in Korean is due to very local movement stranding the case markers. In

5.2. and 5.3, I argue that honorific agreement and plural agreement should be analyzed as cases

of agreement triggered under pied-piping, and stranding, in the case of plural agreement..

5. 1. The position of structural Case markers.

The structural case markers, i/ka (nom), ul (acc), and uy (genitive), follow man ‘only’, as well as

a number of other particles, (kkaci ‘even’, mace ‘even’, cocha ‘even’, pakkey ‘only’), which are

23
in complementary distribution). Case markers occupy a position quite distinct from Ps, which

precede man (Focus), as the following template shows:

(50) NP-HON-PL-(P)- man(only)- CASE

The question arises how a structural Case marker can follow DP externally merged material like

focus. The type of analysis that can be provided is clear: either the merged order is

Focusman>Nom or it is Nom>Focusman, each forcing specific syntactic derivations to satisfy

locality of selection. Focus >Nom is expected, given the general layering conception of

hierarchies: nominative belongs to the TP level, and Focus to the CP level. In similar vain, we

expect Focus> Acc>DP14 and Focus> Gen in the DP domain. We illustrate the derivation based

on Focus> Casenom:

(51) Focusman ‘only’ >Nom (i) > DP

a. DP [i DP to Spec, nom
b. [DP [ man DP [ i merge man, DP extracts
from NomP

A DP is attracted to Spec, Nom satisfying local selection by the Case head, and further extracts

to Spec, Focus, if focused. The resulting phase is send off to phonology, and the nominative head

is incorporated into the phonological phrase and spelled out as i or ka depending on the

phonological properties of the preceding segment. Under this analysis, focus and Case surface in

the merged head initial order, not in the mirrored order. Coordination provides independent

support for Focus>Case. Neither man nor i can be repeated under coordination, but DPs can.

This is consistent with man> i > DP &DP15:

(52) a. Swuni hako Chelswu man- i “A” -lul pat- ass- ta

Swuni and Chelswu only-NOM A -ACC receive-PAST-DECL

‘Only Swuni and Chelswu received an A’

b. *Swuni man hako/kwa Chelswu man i ………..

Swuni only and Chelsu only nom

24
 c. *Swuni man- i hako/kwa Cheslwu man- i ………..

Swuni only- nom and Chelswu only-nom

The hierarchy Nom> Focus (man) > DP also allows a syntactic derivation, provided we find some

way to turn DP-man into a remnant constituent before it moves on to Focus. However, it wrongly

predicts coordination under nominative to be available (52c), and it goes against the idea that

crosslinguistically Focus heads seem to be merged higher than Nom or Acc.

5.2. Honorific agreement.

The honorific suffix follows the verbal ‘root’ and precedes Tense. It cooccurs with a structural

subject that is marked [+honorific] and is thus a particular instance of subject agreement.

Coordination shows Agrhon is lower than T (Choi 2001), hence T>Agrhon>VoiceP:

(53) [AgrhonP ] ko [AgrhonP ]T

Kim sacang- nim-un ilccik chulkun- ha- si- ] ko

Kim president-hon-top early arrive.office-do-hon-conj

[illcik toykun- ha-si- ]ess-ta

early leave.office- do-hon-]past-decl

‘President Kim arrived at the office early and left early’

Under standard approaches to agreement, the constituent triggering honorific agreement is either

in a Spec head relation with the relevant Agr head at some point, or locally c-commanded by it

(under Agree). The discussion in section 4.3. offers another possibility: agreement could be

triggered under pied-piping (as argued for independently in Koopman 2003). This may find

further support by the ability of an honorific possessors to trigger honorific agreement, without

evidence for overt possessor raising. (Sells 1995, fnt. 21, citing Hon 1991:12):

25
(54) [Sensayng-nim-uy son-i ] khu-si-ta

teacher- Hon-Gen hand-Nom big-Hon-decl

‘The teacher’s hands are big (hon.)’

The possessor in Spec, DP agrees with D and hence the entire DP carries the feature [+hon],

allowing Agrhon to agree with the possessor. Under an agreement under pied-piping approach, an

honorific NP in Spec, vP triggers honorific agreement on v; vP therefore inherits the +hon

feature, and triggers further agreement under pied-piping. The honorific agreement is checked off

in Spec Agrhon and will therefore not trigger agreement on higher heads. The honorific head

must be merged low, both within the DP, where it precedes plural, and within the clause, where it

precedes tense (and plural agreement as well). It will pied-pipe with the vP to higher merged

heads, hence its linear position.

(55) AgrhonP
3
vP<+hon> r
3 Agrhon
NP 3 si<+hon>
[+hon] v
<+hon>

5.2.1. Some honorific mysteries.

There are some mysteries with honorifics which Sells takes as additional strikes against the

syntactic view. The purpose of this section is to show that the syntactic view can in fact make the

relevant distinctions using the available mechanisms, and hence that these data do not form any

particular obstacle to the syntactic view.

The first observation concerns Japanese irregular honorific verb forms. Though a bit redundant,

irregular honorific verb forms may appear in the syntactic honorific construction (Sells and Ida

1991), but the regular verb stem cannot appear in this environment.

(56) a. Japanese productive honorific: o-V-ni naru

26
 b. Irregular form: *o-si-ni-naru Æ nasaru ‘honorably do’

c. Double marking is possible: o-nasari-ni naru

Sells points out that double marking seems problematic for a syntactic account since

honorification should have a unique syntactic expression. In light of the preceding section, we can

understand the honorific construction as follows: whenever a V agrees with a honorific subject, it

inherits the feature +honorific. In accordance with the elsewhere principle, the regular verb stem

is blocked in (56b) by the listed honorific form. The listed form can appear by itself, since it

carries the feature +honorific. The listed form can also occur in the syntactic honorific

construction, just like double agreement is allowed.

Sells’ second observation concerns irregular honorific forms in Korean, which ‘fit’ into different

syntactic contexts than regular ones. As shown in (49), C1 cannot cooccur with an overt honorific

suffix. However, C1 can cooccur with an irregular honorific form: (capswusi-e versus *ilk-usi-e)

(Sells:293 (28)):

(57) a. Sensayngim-kkeyse capswusi-e po-si- ess-ta (*mekk-usi-ess-ta)

teacher- Hon.Subj eat.hon C1 try-Hon-past-decl

b. * Sensayngim-kkeyse ilk-usi-e po-si-ess-ta

teacher- Hon.Subj read-Hon-C1 try-Hon-past-decl

The irregular honorific form thus has the distribution of a regular verbal root, yet at the same

time, it must have an honorific subject.

Suppose that irregular honorific verb forms and regular honorific verb forms have the following

syntactic representations when handed over to the phonology. A zero head cooccurs with a

phonologically specified list of stems. Regular honorifics represent the elsewhere case, with the

honorific affix merged as the head of the projection:

27
(58) a. irregular honorifics regular honorific
AgrhonP b. AgrhonP
3 3
VP 3 VP 3
/+list/ [e]hon <+hon> [u)si]hon
<+hon>

These structures express their common distribution, but differ in the distribution of overt/covert

material. The difference in basic syntactic constituency is not without syntactic consequences. VP

extraction in (58a) would strand a silent head but an overt head in (58b). Furthermore, pied-

piping in the latter case would yield the ungrammatical string (59c).

(59) a. [[VP capswusi ] [C-e[AgrhonP[VP capswusi [Agrhon]]

eat.hon C
b. * [[VP ilk- [C -e-[ AgrhonP [VP ilk- [Agrhon usi ]..
read- -e si
c. *[[ AgrP [VP ilk- usi]- [C -e]
read- sihon- e

(59a) shows that an irregular honorific verb has the same distribution as a root V, i.e. it can be in

the Spec, CP of this particular C satisfying the selectional requirement of this C. (59b) probably

violates a phonological requirement on the spell-out of the overt honorific–si (i.e. must have a VP

with overt V in its Spec at spell-out); and (59c) probably violates a size requirement on overt

material in Spec –e (the overt V may not be more deeply embedded than vP). The silent honorific

(59a) “escapes” phonological conditions by virtue of being silent, and therefore only (59a) yields

a well-formed surface form. This allows a CP headed by –e to contain an honorific subject, and

an honorific verb form, as long as the honorific head itself is silent.

5.3. Plural agreement.

Korean has a second kind of subject agreement, resembling complementizer agreement. Plural

subjects optionally trigger plural agreement tul, and do so lavishly. Sells did not include tul in his

article: “The plural marker tul does not appear in the charts above because it shows considerable

28
freedom to where it may attach within the word”. (Sells: 316, footnote 40). In this section, I will

show that the quite challenging distribution of tul can be quite nicely integrated in the syntactic

account . More specifically, I will argue that plural agreement is triggered under pied-piping and

stranding of Agrplural. The differences with honorific agreement follow from the high location of

plural Agreement, which seems to correspond more closely to a very high AgrS.

5.3.1. Distribution of tul.

The data in this section are mostly drawn from Park and Sohn 1993, and references cited

therein.16 Tul is a plural suffix on nouns, and a plural agreement marker optionally co-occurring

with a plural subject17, (optionality is indicated in the examples below by italics); tul can follow

any of the root Cs, provided the CP contains a plural subject18.

(60) a. ai-tul-i cywusi-lul masi-ess- ta- tul Decl-tul

child-PL-NOM juice-ACC drink-PAST-DECL-PL

‘ The children drank juice’

b. ai-tul-i cywusi-lul masi-ess- ni- tul Q-tul

child-PL-NOM juice-ACC drink-past-Q-PL

‘Did the children drink juice?’

c. ttenass –eyo -tul …Level-tul

left- INFORMAL-PL

‘ They left’

Tul may also follow any of the embedded Cs (C1 through C3, section 4.2.), if that CP contains a

plural structural subject.19

(61) a. ilk- e- tul po-ass- ta- tul

read-e- PL try-PAST-DECL-PL

‘They tried to read it’.

b. ilk-ko- tul siph- keyss- ta- tul

29
 read-ko –PL want-MODAL-DECL-PL

‘They might want to read it’

c ilk- ci-tul ahn- ass- ta- tul

read-ci PL neg.aux- PAST-DECL-PL

‘they didn’t read it’

d ilk -usi - ess - eya-tul hay-ss- ta

read-HON -PAST -eya-PL do- PAST-DECL

‘they(hon) had to read it.

Tul may not follow any of the mood/force markers however, if the latter are embedded under the

general subordinator ko (Park and Sohn: 1993, p. 201), which Sells (p. 297) describes as

“basically a marker of someone’s words or thoughts”:

(62) a. John-i [chinkwu-tul-i ttenas-ta (*tul)-ko(*tul)] malha-ess-ta

John-NOM friend-PL-i left-DECL-(*PL)-ko(*PL)said-PAST-DECL

‘John said that his friends left’

b.John-i [chinkwu-tul-i ttenas-nya (*tul)-ko(*tul)] mwul-ess-ta

John-NOM friend-PL-NOM left-Q ko ask-PAST-DECL

‘John asked if his friends left’

But tul may follow the subordinator ko, if ko itself can be said to have a plural subject (Park and

Sohn 1993: 197 (13):

(63) chinkwu-tul-i [John-i ttenas-ta –ko-tul] malha-ess-ta-tul

friend- PL-i John-i left- DECL-ko-PL said- PAST-DECL-PL

‘His friends said that John left’

Park and Sohn (1993) further show that tul cannot occur in relative clauses nor on noun-

complement structures (we discuss examples in 5.3.4.)

Plural tul cannot appear anywhere between Mood/Force and the verb:20

30
(64) ilk- (*tul) usi- (*tul) ess – (*tul) ta – (*tul) ko

V Agrhon T Mood/Force “C”

Finally, tul can optionally follow adverbs and PPs as long as these are “controlled” by the plural

subject, and occur on the projection line between V and the surface position of the plural subject

DP. 21 (65b is drawn from Yim 2001).

(65) a. hakko-eyse-tul wass-ni-tul

school-from-PL came- Q-PL

‘ Did they come from school’

b. ai- tul-i kwaja-lul masitkey-tul mokotta

child-PL- i cookie-ACC taste-C -PL ate-DECL

‘ The children ate the cookies with gusto’

5.3.2 Tul as stranded AgrS.

Plural agreement is distinct from honorific agreement: it occurs in a different position, and it is

neither in complementary distribution with honorific, nor fused with it (61e). Plural agreement is

spelled out in a somewhat unusual position, after C, which makes it look very similar to

complementizer agreement in the West Germanic SOV languages. Like complementizer

agreement, plural agreement is optional. The linear position suggests it spells out a much higher

Agr projection than honorific agreement, say a very high Agr projection just below the C region,

i.e. C > AgrS, or just above C, i.e. AgrS>C. If plural spells out high AgrS, as in (67a), the only

possibility to derive this order from the underlying hierarchy is by obligatorily stranding AgrS.

Since TP precedes C, it is natural to think of the extracted XP that contains the agreement

triggering category as TP, and that agreement is triggered under pied-piping:22

(66) a. C> AgrS >

b. [TPi [C [ TP<+plural> AgrS <+plural>

31
This derivation accounts for the linear position of plural agreement: it must follow C, and cannot

occur anywhere in the verbal complex between V and C, because AgrS is merged higher than T,

and stranded by TP movement.23

(67) *(pl)T-*(pl)-“C”

Instead of C>Agrpl, it could be argued that Agr is merged higher than C, as Shlonsky (1994)

proposes for similar cases of complementizer agreement in the West Germanic OV languages. If

so, the linear order would obey the Mirror Principle:

(68) Agrpl> C> T

Park and Sohn (1993) adopt a variant of (68), in which the presence of a higher Agr projection is

only available in root Cs, thus explaining why tul may not appear in non–root contexts. Their

basic generalization seems to be incorrect, however. Plural agreement does occur in non-root

contexts, though not in all non-root contexts. Non-root clausal complements allow for optional

plural agreement when they contain a plural subject with C1, C2 and C3 endings (61).24

Embedding these predicates under a verb like say, does not affect the plural marking on the

embedded C , but only affects tul’s possible co-occurrence before the subordinator ko (64),

showing that this must be related to some property of ko. (section 5.3.3.)

(69) b. John-i ilk-ko-(tul) siph- keyss-ta- (*tul) ko malha-ess-ta

John-nom read-ko –PL want-PAST- DECL- * PL- ko say- PAST-DECL

John said they wanted to read

The underlined tul is licensed by a plural DP internal to the embedded complement, i.e. PRO, the

external argument of read. Moreover, tul may appear in embedded contexts, as long as the

subordinator ko is absent25:

(70) na-nun kutul-i tolawa-ss- nunci-tul(-ul) mul-ess-ta

I- NUN they-NOM return-PAST-N- TUL(-ACC) ask-past-decl

'I asked if they have returned.'

32
Since (68) fails to account for tul’s possible appearance in embedded environments, I will

continue to assume that tul spells out AgrS (C>Agr).26 Thus, Korean has a crosslinguistically

quite common high AgrSpl, as well as a much lower AgrS (Agrhon).

5.3.3. Problems: ko-subordination

Why cannot tul precede the subordinator ko? Consider a representative example (63), repeated

here as (72):

(71) John-i [chinkwu-tul-i ttenas-nya (*tul)-ko(*tul)] mwul-ess-ta

John-NOM friend-PL-NOM left- Q ko ask- PAST-DECL

‘John asked if his friends left’

A clausal complement with a plural subject cannot trigger plural agreement, neither after the

question particle, nor following the subordinator ko. Plural agreement can follow ko though , if ko

can be said to have a PRO controlled by the plural DP of the verbs of thinking or speech:

(72) chinkwu-tul-i [John-i ttenas-ta –ko-tul] malha-ess-ta-tul

friend- PL-i John-i left- DECL-ko-PL said- PAST-DECL

‘His friends said that John left’

In our paper on logophoricity (Koopman and Sportiche 1989)), certain types of complementizers

(“say” type complementizers) are analyzed as Vs projecting a PRO subject, dominated by a CP

node, and taking a CP complement. This proposal directly captures the distribution of plural

agreement following ko, as can be illustrated with the underlying hierarchy, abstracting away

from leftward movements.

(73) C1 > AgrS1>T>DPpl say> C2 > AgrS2 >T> PROpl> ko> [C3> AgrS3> T > DP3pl

A plural subject in C1 will yield plural agreement following C1, a plural subject in C2, plural

agreement following ko, a plural in C3 plural following C3. In (71), PRO is singular, since its

33
controller is singular; hence ko cannot be followed by tul. (* ko tul). In (72), PRO is plural, since

its controller is plural, whence ko-tul.

This leaves the question why tul cannot be triggered on AgrS3 when embedded under ko, yielding

. *C-tul-ko. The derivation underlying this order would contain the following substructure:

(74) 3
CP 3
3 ko
TPpl 3
5 C 3
TP<+pl>3
tul

This configuration contains overt phonological material on a right branch lower than the head C,

and is notorious for yielding ungrammaticality (*a proud of John mother) . I will assume that

whatever explains the ill-formedness of such configurations (see Koopman 2002 for discussion),

will rule out (74) as well. Simply not agreeing, an option that is always available in Korean, will

yield convergence, and removes the phonological problem. Alternatively, making merger of

AgrSpl optional would yield the same results.

5.3.4. More problems: Relative clauses and noun complement structures.

The plural marker can neither follow nor precede the complementizer in relative clauses or noun-

complement structures (Park and Sohn 1993).

(75) salam-tul-un [ kaysenha-(*tul)-un-(*tul) ] cangpyeng-tul-ul] hwanyengha-ess-ta

man-PL-TOP return C soldiers- PL-ACC welcome-PAST-DECL

‘People welcomed the soldiers who returned’

Relative clauses precede the head of the relative clause, as do clausal complements to the noun.

Kayne (1994) proposes that UG makes a D0 CP structure available for relativization, but prohibits

the mirror image CP D0. Therefore, head final relative clauses must be analyzed as deriving from

34
head initial structures, and involve raising of the head noun to Spec, CP, followed by IP

preposing to Spec DP (Kayne 1994: 94).

(76) a. D0 [[ NP ]i Co [IP …………[NPe]i

b. IPj D0 [[ NP ] Co [e]j

Filling in the details in Korean, assuming nun equals D and C is silent yields the following

intermediate structure:

(77) D0 [[ NP ] Co [ Agrpl

nun soldiers

Given this derivation there are two options. either AgrP pied-pipes to the left of D, or Agrpl is

stranded. There is no possible derivation that would result in tul immediately following D*(n)un

tul NP. The prediction of this analysis then is that this form should be invariably excluded across

dialects of Korean27. The analysis would yield the following forms:

(78) a. * [[ TP] tul] nun soldiers

b. * [TP ] un soldiers [TP [tul]

(78a) poses a similar problem as (77), with tul on a right branch lower than T, preceding nun. If

TP movement in Korean strands plural AgrS, AgrS should be stranded after the head noun, but

never before or after the D. (Note that this results in the same linear order as the plural nominal

affix tul ).

(79) [TP(pl) ] [D0 un [NP [ Co [AgrS TP<pl> [tul]<+pl>

Thus, plural agreement should be triggerable by a plural subject within the relative clause or

complement clause. At this point, this prediction is not borne out (though plural subjects seem to

facilitate the plural following the head noun for many speakers), and further research is necessary,

taking into account the semantics of plural tul.

As I have shown in this section, the syntactic account adopted in this paper yields new insights

into the distribution of tul. The idea that agreement can be triggered under pied-piping plays a

35
crucial role (Koopman 2003). This possibility is expected to exist given what we know about the

properties of syntactic pied-piping in general. Phonological properties of individual lexical items

play a further filtering role: sometimes derivations are filtered out because a head would be too

deeply embedded at spell-out, sometimes individual heads filter out certain otherwise licit

derivations because there is pied-piped overt phonological material on a right branch.

6. On the interaction between scope and word structure.

In this section, we further examine the interaction between scope and word structure. If word

structure is derived in the syntax, and scope is determined by the merger of the heads that

determine scope, word structure should interact with available scope possibilities as a

consequence of the syntactic derivation that underlies the linear order.

Lee (2004, in prep) makes exactly this point. She uses word structure to account for scope

interactions in Korean. In a nutshell, she shows that an accusative object that precedes a universal

subject cannot scope over a subject QP, but a preceding PP can either scope over or under the

quantified subject.

(80) John man-ul motun-salam-i salangha-ta (every>only *only>every)

John FOCUS-ACC every person-NOM love-DECL

‘Everyone loves John (and no one else)’

(81) John hako-man motun-salam-i akswuhay-ss-ta (every>only and only>every)

John with only every-person-NOM shook.hands-PAST.DECL

‘Everyone shook hands with only John’

‘John is the only one with whom everyone shook hands’

Lee proposes that man is an agreement morpheme, that triggers movement of the focus

marked constituent to a silent Focus head ONLY, where semantic interpretation is

determined. The silent focus head can be merged at several points in the hierarchy. If

36
ONLY is merged low , universal> focus results, if ONLY is merged higher than the

universal, in the left periphery, the reading is focus>universal. DP-man-acc cannot scope

over universal subjects because the linear order DP-man-acc signals the silent Focus head

must be merged below accusative hence Universal>only is the only available reading.

Lee appeals to the Mirror Principle (“check inner affixes before you check outer

affixes”), to force the low position of the silent Focus head with accusative case. A PP

(81) can scope either over universals, or below universals, because the linear order is

consistent with both high or low merger of ONLY. Lee thus presents an extremely strong

case for the syntactic relevance of affix ordering.

As far as I can tell, these scope facts fall out in very similar ways in the proposal made in

this paper, where man is merged directly as a focus head, higher than accusative,

provided of course that we accept head complement order. The linear string man-acc

shows that man (only) must be merged immediately above accusative, in conjunction

with the universal hierarchy Focus>Case, since this is the only way to form this particular

surface constituent28 Hence DP-man-acc will scope under universal (and preposing to the

left of the universal must be achieved by scrambling). There is no other derivation that

Korean speakers can converge on given the primary data:

(82) DP man [ DP[ul [ …

Consider, for example, what outputs would result from merger of man higher than the subject

QP. This derivation would necessarily yield the reading only>every.

(83) Merge man (focus):

a. [Focus [ man [QP every DP ]nom .. [DP+focus[ ulacc

Move DP +focus: move [DP-ul] (assuming a remnant can be formed)

37
 b. *[ DP- ul acc -man [QP every DP-nom [DP acc [

or: extract DP[+focus], and strand accusative

c. *[Focus DP+focus [man [QP every DP-nom [DP [ ul acc

Neither string corresponds to the input strings that native speakers are asked to judge in (81), nor

are they well-formed. It must be explained how these strings are ruled out, and the reason here

does not seem to be syntactic, but phonological. The derivation in (83b) yields overt phonological

material on a right branch, and I will tentatively assume that this is the reason (83b) is out.29 (83c)

results in a stranded accusative –ul. It is likely that this can be excluded by a phonological

condition on ul: it needs to lean on overt phonological material in a specific phonological domain,

and though it can be stranded, it cannot be the sole remnant in that domain. Interestingly, not

pronouncing ul is a way to remove the phonological restriction, and as expected, both high scope

(=high merger) and low scope (=low merger) of man is possible in this case. (Lee, 2004, ex.

(26))30

(84) a. John-man1 [motun-salam-i e1 salanghanta].

John-only every-person-NOM love

‘Only John, everyone loves e.’

(i) Everyone loves John and no one else. (very > only)

(ii) John is the only one whom everyone loves. (only > every)

Finally, as in Lee’s account, Case markers must be part of the “narrow” syntax, since they

interact with the derivations in specific ways forcing decisions about the location of scope bearing

heads, and hence making very specific predictions about scope interactions.

7. Conclusion.

In this paper, I have argued in favor of a syntactic view of Korean (and Japanese) morphology,

which derives the surface constituency of “words” from an underling Spec H complement order

by means of local phrasal movements, providing strong support for Kayne (1994). This accounts

38
for the fact that the affixes are phrasal affixes. This view uses underlying hierarchies, movement

from Spec stranding heads, pied-piping, and generalized Spec head agreement under pied-piping.

My primary concern has been to show that syntactic analyses can be motivated, contra Sells

1995, and that the vocabulary necessary to describe word structure is the usual syntactic

vocabulary. Inflected words are derived by phrasal movement, with many, but not all, words

mirroring the syntactic hierarchy of merger. Underlying head complement order surfaces in

particular instances, directly supporting antisymmetry. Furthermore, I argued that certain strings

that the syntax should generate are filtered out by “phonological” properties of individual lexical

items (complexity filters) when phonological material is inserted. Sometimes not inserting

phonological material can salvage particular derivations.

References

Baker, M. 1985 “The Mirror Principle and Morphosyntactic Explanation, “ Linguistic

Inquiry 16: 373-415.

Bayer, Josef: 1984 “ COMP in Bavarian syntax,” The Linguistic Review 3, 209-274.

Beghelli, F. and T. Stowell, 1997, Distributivity and negation: the syntax of each and every.

In A. Szabolcsi, ed. Ways of Scope Taking, 71-109. Dordrecht, Kluwer.

Belletti, Ad (2002), Aspects of the low IP area. in L. Rizzi (ed) The Structure of IP and CP.

The Cartography of Syntactic Structures" vol.2. Oxford University Press.

Bennis, H. and L. Haegeman. 1984. On the status of agreement and relative clauses in West

Flemish. In W. de Geest and Y. Putseys, eds. Sentential Complementation. Dordrecht:

Foris. 33-53.

Carstens, V. 2003 ‘ Rethinking Complementizer Agreement: Agree with a Case-Checked

Goal’ Linguistic Inquiry, 34.3. 393-413.

Cecchetto, Carlo. 1999. A comparative analysis of left and right dislocation in Romance.

Studia Linguistica 53, 40-67

39
Cho, Young-mee Yu, and Peter Sells. 1995. A lexical account of inflectional suffixes in

Korean. Journal of East Asian Linguistics 4, 119–174.

Chomsky, N. 1998 “Minimalist Inquiries: The Framework.” MIT Occasional Papers in

Linguistics, 15, MITWPL, Cambridge, Mass.

Chomsky, N. 2001 'Derivation by Phase' in Michael Kenstowicz; Ken Hale: a Life in

Language. 1-52. . MIT Press. Cambridge

Chomsy, N. 1970. “Remarks on Nominalizations.” In: R. Jacobs, and P. Rosenbaum (eds)

Readings in English Transformational Grammar.

Chung-hye Han, Jeffrey Lidz, Julien Musolino (2003) Verb-raising and Grammar

Competition in Korean:Evidence from Negation and Quantifier Scope

Cinque, G. 1999. Adverbs and functional projections: A cross-linguistic perspective. New

York: Oxford University Press.

Cinque, G. 2002 ‘A note on mood, modality, tense and aspect affixes in Turkish’ in Eser

Taulan (ed) The verb in Turkish. John Benjamins, Amsterdam/Philadephia.

Embick R and R. Noyer, 2001 “Movement operations after syntax” Linguistic Inquiry,

Volume 32, Number 4, Fall 555–595

Fukui, N and H. Sakai 2003. The Visibility Guideline for Functional Categories:Verb Raising

in Japanese and Related Issues . Lingua, 113. 321-375.

Fromkin, V. 2000. Linguistics: an Introduction to Linguistic Theory. Blackwell Publications.

Grimshaw, J. 1991 Extended Projections, ms. Brandeis.

Halle, Morris and Alec Marantz. 1993. 'Distributed Morphology and the Pieces of

Inflection.' In The View from Building 20, ed. Kenneth Hale and S. Jay Keyser. MIT Press,

Cambridge, 111-176.

Hallman, P. 1997. “Reiterative Syntax”. In Clitics, pronouns, and movement, J. Black and V.

Motapanyane (eds).87-131, Amsterdam: John Benjamins

40
Julien, M. (2002) Syntactic heads and Word Formation, Oxford University Press.

New York

Kayne, R.S.1994. The antisymmetry of syntax. Cambridge, Mass: MIT Press.

Kayne, R. S. 1998. “Overt vs. covert movement.” Syntax 1: 128-191.

Kayne, R. S. 2003 “Antisymmetry and Japanese”, English Linguistics, 20, 1-40.

Kayne, R.S. 2000. Parameters and Universals. New York.: Oxford University Press.

Koizumi, M. 1995. Phrase structure in minimalist syntax. Doctoral dissertation, MIT.

Koizumi, M., 2000. String vacuous overt verb raising. Journal of East Asian

Linguistics 9, 227–285. d in 1987]

Koopman, H. 1984 The syntax of Verbs: from Verb movement rules in the Kru languages to

Universal Grammar, Foris Publications

Koopman, H. 1994. Licensing Heads. in N Hornstein and David Lightfoot (ed) Verb

Movement. Cambridge: Cambridge University Press.

Koopman, H. 2000. The syntax of specifiers and heads. Routledge, London

Koopman, H. 2001 “Topics in Imperatives” Imperatives in Generative Grammar. W. van der

Wurff (ed). Benjamins Publications (in press)

Koopman, H. 2002. ‘Derivations and Complexity Filters’ in Dimensions of

Movement: from features to remnants. Alexiadiou A, Anagnostopoulou E, Barbiers

S, Gaertner H (eds). Amsterdam: John Benjamins.

Koopman, H. 2003. Inside the Noun in Syntax at Sunset : Head Movement and Syntactic

Theory. A. Mahajan (ed). pp. 77-116. UCLA/Potzdam Working Papers

Koopman, H. and D. Sportiche. 1989 "Pronouns, Logical Variables and Logophoricity in

Abe", Linguistic Inquiry 20 (p555- 589)

Koopman, H. and Szabolcsi A. 2000, Verbal Complexes Current Studies in Linguistics 34,

Cambridge, Mass: MIT Press.

41
Kornfilt, J. (2002). Agreement and its Placement in Turkic Non-Subject Relative Clauses, ms.

Syracuse University.

Kural, M. 1994 ‘Inf and Comp in Turkish subordinate clauses’, WCCFL 12, p. 339-354.

Kuroda, S,-Y. 1986.

Kuroda, S,-Y.2003 Complex predicates and predicate raising, Lingua, 113, 447-480.

Laka, I. 1990. Negation in Syntax: On the Nature of Functional Categories and Projections.

Ph.D. Dissertation, MIT.

Lee, Y. 2004 ‘Only, scope and the Mirror Principle’ to appear in Natural Language

Semantics.

Marantz, A., 1988. Clitics, morphological merger, and the mapping to phonological structure.

In:M. Hammond, M. Noonan (eds.), Theoretical morphology: Approaches in modern

linguistics, 253-70. San Diego: Academic Press.

Muller, Gereon. 2000. Shape conservation and remnant movement Proceedings of NELS 30

Nkemnji 1996, Heavy Pied-Piping. PhD thesis, UCLA.

Otani, K. and J. Whitman (1991). V-raising and VP-ellipsis. Linguistic Inquiry 22, 345-358.

Park Myung-Kwan and Keun-Won Sohn (1993) A Minimalist Approach to Plural

Marker Licensing in Korean, in MITWPl, 20, p. 193-209.

Pearson, M. 2000. Two types of VO languages. In P. Svenonius (ed) The

Derivation of VO and OV Languages. Amsterdam: John Benjamins Publishing

Company.

Poletto C. and Pollock J-Y. 2003. “ On the Left Periphery of some Romance Wh-

questions.” In The Structure of IP and CP. The Cartography of Syntactic Structures

2" G. Cinque (ed.) New York: Oxford University Press.

42
Rizzi, L. 1997 .“On the fine structure of the left-periphery” in Elements of Grammar.

Haegeman L. (ed.) 281-337. Dordrecht: Kluwer Academic Publishers. [Also in Rizzi, L.

2000.Comparative Syntax and Language Acquisition, London: Routledge].

Rizzi, L. 1999. On the position “Int”(errogative)” in the Left Periphery of the Clause,

msUniversità di Siena.

Schütze, C (2001) 'On Korean "Case stacking": The varied functions of the particles ka and

lul.' The Linguistic Review 18(3), 193-232.

Sells, P 1995 Korean and Japanese Morphology from a lexical perspective, Linguistic

Inquiry, 26. 237-325.

Sells, P 1998. Three aspects of negation. Journal of Linguistic Studies 6, 1-15.

Sells, P. 1998. “Structural relationships within complex predicates” In Byung-Soo Park and

James Yoon (eds.) Proceedings of ICKL 11. Seoul, International Circle of Korean

Linguistics, 115-147.

Sells. P. and M. Ida 1991. Subject and Object Honorification in Japanese. In Proceedings of

the Seventh Annual Meeting of the Berkeley Linguistic Society, 312-323. Berkeley Linguistic

Society, University of Berkeley, California.

Shlonsky, U. 1994. “Agreement in Comp”. The Linguistic Review 11:351-375.

Sportiche, D. (1998) Atoms and Partitions of Clause Structure, Routledge, London

Sportiche, D. 1997 “Reconstruction and constituent structure.”Class lectures and lectures

presented at MIT and various other places. .

Sportiche, D. 1997 “Reconstruction and constituent structure”, GLOW abstract.

Sportiche, D. 2001 Structural Bonds and Strict Locality, Tools in Linguistic Theory

#1, Rijksuniversiteit Utrecht.

Webelhuth,G. 1992. Principles and parameters of syntactic saturation, New York: Oxford

University Press.

43
Whitman, J. (2001) “Kayne 1994: p. 143, fn. 3,” in G. Alexandrova & O. Artunova (eds.),

The MinimalistParameter, John Benjamins, Amsterdam, 77-100.

Whitman, J. (2003) “Preverbal Elements in Korean and Japanese,” in G. Cinque and R.

Kayne (ed s.) Handbook of Comparative Syntax, Oxford University Press, New York.

Williams, E. (2003) Representation Theory. MIT Press, Cambridge Mass.

Yim, Changguk, 2001 “Subject Agreement in Korean: Move F, Attract F, or Agree?

“to appear in proceedings of The role of agreement, University of Texas, Austin

Texas.

Yoon, J. (1994) "Korean Inflection and Checking Theory," MIT Working Papers in

Linguistics, C. Philip & H. Harley eds., Department of Linguistics, MIT, pp. 251-270

Zwart, C. J.-W 2001. Syntactic and phonological verb movement. Syntax 4.1

Zwart, C. J.-W. 1993. Dutch syntax: a minimalist approach. University of Groningen PhD

thesis.

44
*
I would like to thank Sun-Ah Jun, Seungho Nam, Soyoung Park, Genhee Kim, Chistina Kim,

and Chunmin Lee for their generous help with Korean data. Many thanks to John Whitman for

extensive feedback and comments on an earlier version, as well as to Dominique Sportiche,

Richard Kayne, Gugluelmo Cinque, Christina Kim, David Schüller, Greg Kobele, the participants

of my UCLA seminar (spring 2003), and to two anonymous LI reviewers. Romanization of

Korean examples needs to be made consistent.

1
See also Kuroda (2000) who shows that disjunction can be under the scope of causative, clearly

an indication of the syntactic nature: (Kuroda 2000: 445 (16)) (also Kuroda 1986a: (16); 1993a:

(15))

(16) Hanako ga Masao ni uti o soozisuru ka heya-dai o haraw-aseru koto ni sita

Hanako NOM Masao DAT house ACC clean or room-rent ACC pay cause that to do

‘Hanako decided to make Masao clean the house or pay room rent’
2
In Koopman and Szabolcsi (2000), we propose all inflectional morphology is derived through

phrasal movement, in which case the different behavior of the different types of affixes need to be

independently explained. The point in the text simply states the fact that Japanese and Korean

affixes do not behave like –ed, but rather like ‘s.

3
This analysis leave open the possibility that material to the left pied-pipes with the phrasal

constituent to higher functional projections, as indeed must be the case. In Korean, short negation

an precedes the verb, and a subconstituent an [V-X] is formed (Whitman 2002). However, (non-

verbal) dependents of the verb occur to the left of short negation, and some of the elements

(idiomatic parts, small clause predicates, and low adverbs) must remain after ellipsis, showing

they are outside the elided constituent, i.e. they have undergone movement as well. Higher

material (definite DPs, quantified DPS, localitive and temporal adverbials) are part of the elided

constituent(in fact, this looks suspiciously like the Mittlefield in West-Germanic).

45
(i) Mary-ka motun panana-lul eoje jengwon-e [cal [an mek]-ess]-ni

M-nom every banana-acc yesterday garden-in well not eat-past-Q

(ii) *(cal) an mek-ess ta

well not eat-past-decl (she every banana yesterday in the garden)

‘She didn’t’

Coordination yields parallel results showing that the low adverb is within the ATB moved

constituent (to C): (from Han, Lidz, and Musolino 2003)

(ii) Mary-ka [[FP motun sakwa-lul ti] kuliko [FP motun panana-lul ti]] cal mek-ess-ta.

Mary-NOM every apple-ACC and every banana-ACC well eat- PAST-DECL

‘Mary ate every apple and every banana well.’

Many questions remain about the finer constituent structure of the inflected verb phrase.

However, these go beyond the basic point in this paper and should be examined in the context of

Cinque 1999.
4
For PP pied-piping, see Koopman and Szabolci 2000:p42.
5
I phrase agreement in terms of copying, instead of checking, since I think this better describes

the basic agreement mechanism. Nothing hinges on this particular choice however.
6
For expository reasons, the picture is simplified here. If followed through, all functional

categories are merged outside of V NP. (See Sportiche 2002).

i have lost footnote 8,. and seem to be unable to delete it, or renumber the following notes.
8
This step could perhaps also motivated by the Principle of Locality of Selection. If P selects VP

and merges with VP, there would be no motivation for VP movement. However, suppose that the

verb and the P are actually not in a local relation at the point of merger, but that P merges like a C

(Kayne, 2000). If that is correct, locality of selection must be met after movement, hence both DP

movement and VP movement are obligatory to establish the necessary local configurations.

46
9
These derivations do not show the VP movements, which have been added in (i) below, and

which mimic Kayne’s (1998) derivations

a. move VP VP w [PP DP P ]

b. Merge kkaci,

b.1. piedpipe PP- kkaci: [PP kkaci [ [ VP

c1. move VP VP x [PP kkaci..]

or: b2. move DP: [[DPDP [kkaci [VP[PP P…

b3. move VP VP x [ DP kkaci ..P]

10
Note that it is not reasonable to analyze an here as merged lower than adverbs and AgrO. Short

negation an does not modify the lower predicate but the higher one, and still part of the

complement clause must restructure (as we can see by the fact cal modifies the embedded verb,

yet precedes the high negation). It rather looks like an is a very high clausal head, which attracts

the complex verb to its right, with subsequent movement to its Spec.

47
11
Not all speakers accept the short form negation an with man focus in this context. For those

speakers that do accept this combination, the scope is focus>negation (Many thanks to Chungmin

Lee for extensive multispeaker feedback on this issue). For some speakers, focus can also scope

under negation, but only when an accentual phrase follows the focused constituent. This points to

a hierarchy where man is merged lower than negation, within the clausal complement of try.

Interestingly, this reading is unavailable when DP man an V2 V1--forms a single intonational

phrase, suggesting that the structure underlying this phrasing can only be man>an. (Thanks to

Christina Kim for the pointing out to me that the morphological structure interacts with available

scope readings, and to Sun-Ah Jun and Chungmin Lee for extensive feedback on this issue). It

seems, as so often with verbal complexes and restructuring, that there is considerable speaker

variation in this area (see Koopman and Szabolcsi (2000)). At this point, the variability can be

summed up as follows:

(i) some speakers do not accept: ilk-e man an po-ass-ta

read-e only an try-past-decl

Of those that do:

(ii) some speakers only accept Focus>negation, regardless of phrasing.

(iii) some speakers allow Negation>focus, but only if man is followed by an intonational

break, or the contrastive topic marker -(n)un).

I take the variation in (ii), (iii) to reflect whether individual speakers exceptionally allow focus to

be merged within the –e complement of ‘try’ or not. If they do (iii), low merger must be

consistent with the phonological phrasing/word structure, suggesting this is another instance

where word structure and syntax determine scopal possibilities (Lee 2004 and section 6. As a

final note, all speakers seem to allow focus to be merged within the fully clausal –ci complement

of long negation, yielding Neg>only.

12
I will not consider a mixed account where desu and masi are merged in different locations:

48
(i) selection satisfied after movement though Spec head agreement for desu:

(desu> T> AP)

(ii) selection satisfied under first merge for masi (T>mas>VP)

13
In order to account for the fact AP does not Spec-extract, it must be assumed that katta cannot

strand. This can be achieved if katta must have overt material in its Spec at spell-out. In

‘substandard’ Japanese –desu can also precede ta, but this requires a different morphology on A

(present tense i) (John Whitman, personal communication)

(i) aka- i-desi- ta

red- i- FORMAL - PAST

This suggests that Spec extraction might also be available for APs in substandard Japanese,

though the difference in morphology is unexpected, and in need of an explanation.

(ii) [[aka- i] desi [[aka-i] ta

14
I assume Focus can be merged in the accusative region (see also section 6). For recursion of

the left periphery in lower regions of the sentence, see Hallman 1996, Sportiche 1996, Cecchetto

1998, Koopman (2001), and Belletti (2003) among others.

15
As an anonymous reviewer points out, there are contexts where X-man-i can be coordinated by

hokun ‘or’.

(i) chwulsek-ul modu ha-n haksayn-man(-i), hokun swukcey-lul ta nay-n haksayn-man(-i)

A-lul pat-ul iss-ta

“Only students who are present everyday or only students who submit all the reports

could receive an A.”

Notice that this example is considerably more complex than the example in the text. In particular,

the DPs contain relative clauses. It may be the case that the more complex derivation here creates

the possibility for ATB movement.

16
I would like to thank Sun Ah Jun and Seungo Nam for their help with these data.

49
17
Tul on non-subjects is claimed to force a distributive reading. (see Park and Sohn (1993) and

H.-K Lee (1991) for some discussion). I will have to ignore questions of interpretation.
18
Some Korean speakers find the sequence ta-tul acceptable, but slightly uncomfortable.
19
man can further attach to the CP (ilk-e-tul man), suggesting (remnant) CP extraction to Focus

[ikl-e-tul] man [ilk-e tul ] V.

(i) ilk- e-tul- man po-ass- ta- tul

read-e-pl- only try-past- decl-pl

‘They tried only reading it’

As an anonymous reviewer points out, the order ilk-e man-tul is also acceptable, though generally

less preferred. This might suggest the presence of an additional Agr position just below man, and

CP recursion, with VP-e moving through AgrS and triggering plural agreement under pied-

piping:

(i) VP-e man [CP [ VP-e [AgrStul .. [ … [VP+[ CP VP [e[CP] [try

This analysis predicts that man must have scope over an(negation) given this particular linear

string, though not in the other order. (Note that this only holds for speakers who allow man to be

merged within the complement clause to start with) (see footnote 11) At this point, judgments

here are very difficult and not very secure., though they seem to go in the right direction.
20
The contrast between (61) and (64) is quite problematic for a templatic view of morphology. If

–tul can attach to C1 etc, and if C1 occurs in slot 1,why then cannot it attach to other elements in

slot 1?
21
As expected, ECM accusatives behave like subjects:

sonsaehngim-i [hakseng-tul-ul kyosil-e- tul itta-ko midotta

teacher-hon-nom student-pl –acc classroom-Loc-pl exist-ko believe.decl

‘ the teacher believed the students to be in the classroom’

50
22
Kornfilt (2002) shows that in certain Turkic languages a post-nominal agreement morpheme

agrees with the subject of a pre-nominal relative, and analyzes these in terms of AgrS stranding.

Turkish data presented by Kural (1994) initially inspired me to pursue this type of analysis.

Subject agreement follows what Kural analyzes as a complementizer.

(i) Ahmet benim uyu- du-¢g-um-u sanýyor

Ahmet my sleep-past-COMP-1SG-ACC think

‘Ahmet thinks I am sleeping’

(ii) Ahmet senin uyu- du- ¢g- un- u sanýyor

Ahmet your sleep-PAST-comp-2SG-ACC think

‘Ahmet thinks you are sleeping’.

23
This analysis forces a CP structure for restructuring predicates. Tul can appear after PPs and

adverbs that occur on the main projection line between V and C. It is tempting to assume that in

all these cases, the same configuration holds, i.e. each time tul shows the presence of C>Agr,

through which a plural XP has transitted.

24
Recall that the regular honorific suffix is excluded from this context (58), while irregular

honorifics can occur (see (59). In section 4.2.1. I took this to show that these complements can in

fact contain Agrhon.. (61a) shows that these complements can also contain a high AgrS, further

confirming Koopman & Szabolcsi’s (2000) idea that complements of restructuring predicates are

always dominated by full CPs , which are made to look small when part of the CP, which is

selected by the restructuring predicate, extracts.

25
The fact that tul can be followed by Case is expected, as the entire CP can move on to Spec,

Case.
26
This suggests that all cases of complementizer agreement should be analyzed as being triggered

under TP movement stranding agreement. It is suggestive here that only the SOV Germanic

languages have complementizer agreement, and that complementizer agreement is only possible

51
in clauses that have an OV order: embedded verb second is incompatible with complementizer

agreement ( cf. Shlonsky 1994, Kayne 1994: p.56, Zwart 1997, Bennis and Haegeman, 1983,

Bayer 1983, see in particular Carstens 2003).

27
On the other hand, if tul simply spells out features on the C node, this form should be found in

dialects of Korean.
28
The surface constituent: DP-man-ul is formed by further VP movement to the left of man .
29
The low interpretation shows that [DP-ul-][man has to be excluded independently.
30
For other cases where not inserting phonological material removes the phonological restriction,

see (4.3.2, 5.2.1, 5.3.3.). This account should extend to other cases where topic and focus markers

are incompatible (such as Japanese*ga-wa, *wa/ga, cf. Kayne 1994, p. 143.)

52