Structural Relation Within Complex Predicates

32
 

description

Complex predicates

Transcript of Structural Relation Within Complex Predicates

  • Structural Relationships Within Complex Predicates

    Peter Sells

    Stanford University

    Introduction

    Complex predicates are a highly salient feature of Korean sentences, and consequently have received

    a great deal of attention in the linguistic literature. As there have been many excellent papers

    written on aspects of the syntactic, semantic, and pragmatic relationships between the parts of

    many dierent kinds of complex predicates, it is quite a daunting task to take up the general topic

    of complex predicates without running the risk of merely recapitulating earlier scholarship. For

    this paper, I have tried to focus on certain aspects of the syntax of complex predicates which

    have not received much previous attention, but which have very important consequences for our

    understanding of what combinatory mechanisms are available at least in Korean grammar, and

    possibly universally.

    The usual understanding of the term `complex predicate' is a semantic one: a complex predicate

    consists in the argument structures of two separate predicates being brought together somehow or

    other; and further, typically, the argument structure of one of those predicates in isolation is taken

    to be incomplete, `light', or `bleached'. Within this basically semantic idea of a complex predicate,

    we can nd dierent structural manifestations. Complex predicates can be single words, as in the

    lexical causatives and passives of Korean. Looking to slightly larger units, there are the forms that

    we might consider to be `compounding' or perhaps `serializing', such as tol-a ka-ta (`return'), sal-a

    na-ta (`revive') (from Sohn (1994, 365.)). Going to the other end of the spectrum from these very

    tight combinations, we nd rather complex complex predicates such as the one expressing ability,

    shown in (1).

    (1) chelswu-nun ku chayk-ul/*-i ilk-ul swu iss-e-yo

    Chelsoo-TOP that book-ACC/*-NOM read be.able-POL

    `Chelsoo can read that book.'

    Such complex predicates seem to involve true phrasal subordination, and perhaps always involve

    an internal nominalized constituent. They may appear to be perhaps `too complex' to be worth

    investigating. Although I will not discuss this type, I think that there is in fact a lot to be learnt

    from them. One interesting point is that Japanese has a complex periphrastic expressing ability,

    somewhat like (1), yet Japanese diers from Korean in allowing a case alternation on the object of

    the most embedded verb.

    (2) taroo-wa tyuugokugo-o/-ga hanasu koto-ga deki-ru

    Taroo-TOP Chinese-ACC/-NOM speak.PRES fact-NOM be.able-PRES

    `Taroo can speak Chinese.'

    0

    This paper will be presented at the 1998 meeting of the International Circle of Korean Linguists, Hawaii, July

    6{9. I am grateful to Ivan Sag for very helpful discussion of the more technical parts of this paper, and to Hanjung

    Lee for her assistance with several of the Korean examples. All errors and misinterpretations are my responsibility.

    1

  • Now without implying any particular signicance for the categories given, we can identify at least

    the various types of complex predicate shown in (3). The rst two types appear to be formed

    lexically, while the rest are syntactic.

    (3) Syntactic Varieties of Complex Predicate

    a. Axal (lexical causative and passive)

    b. Compounding/Serializing (tol-a ka-ta)

    c. Verbal Complex (see (4){(5))

    d. Verbal Noun plus Verb (kongpwu ha-ta)

    e. Phrasally Periphrastic (see (1))

    For my purposes here, I will concentrate only on the verbal complex type, taking it that the relevant

    kind of complex predicate essentially has the form V

    1

    { V

    2

    , where in general V

    1

    is the main content

    verb and V

    2

    is the bleached or light predicate, as shown in (4).

    (4) Syntactic Verbal Complex: V

    0

    V

    0

    1

    - V

    0

    2

    -TNS-MOOD-etc.

    This is a `small' construction, formed in the syntax; I use `X

    0

    ' to indicate any item drawn from the

    lexicon and inserted in the syntax. In the verbal complex, V

    1

    always bears a linking or complemen-

    tizing morpheme which appears to be specic to the particular complex predicate in question;

    usually we think of this as a selectional property coming from V

    2

    . Some examples of this kind of

    complex predicate are given in (5).

    (5) a. ilk-e po-ass-e-yo

    read-COMP try-PAST-POL

    `tried to read'

    b. anc-ko iss-e-yo

    sit-COMP PROG-POL

    `is sitting'

    c. mek-e cwu-sey-yo

    eat-COMP give-HON-POL

    `give the favor of eating'

    d. ssu-na po-a-yo

    write-COMP seem-POL

    `seems to write'

    e. ilk-e-ya hay-yo

    read-COMP must-POL

    `must read'

    f. ilk-ci anh-a-yo

    read-COMP NEG-POL

    `does not read'

    Concentrating on complex predicates of this form, in the rst section of this paper I will look

    at dierent relationships that the two parts of a complex predicate may have with respect to

    each other, and propose a fundamental two-way distinction in the way the two parts enter into

    the whole structure. I will refer to these two types as `lexically governed' and `freely composed'

    complex predicates, using just the terms `governed' and `free' for ease of reference. Roughly, a

    governed complex predicate has properties which are highly specic to V

    2

    in its particular usage in

    that construction, while a free complex predicate is much less restrictive about the nature of V

    2

    .

    2

  • Correlating with this, governed complex predicates show tighter syntactic connections between the

    two verbs than do free complex predicates.

    In section 2, I will consider dierent types of governed complex predicate, and discuss some basic

    semantic dierences between them, to show the range of interpretations that a complex predicate

    may have. Then in section 3, I will show that, in spite of their dierent semantics, all of the

    governed complex predicates form a homogenous class with respect to an important diagnostic

    property, namely the placement of short form negation. This property distinguishes them from

    the free complex predicates. I will present an account of short form negation which explains this

    placement.

    In contrast to the governed complex predicates, the freely composed ones all mark V

    1

    with the

    marker -key . In section 4, I will take up the question of what a free complex predicate is, and how

    it is formed. Related to this, I will take up the issue of the identity of -key , if I am to substantiate

    my claim that it is distinct from all the other elements which can appear on V

    1

    in combinations

    like (4).

    In section 5, I briey consider the similarities of verbs marked with -key and verbal nouns, to further

    illustrate why such verbs are less like verbs with other COMPs. Finally, and rather sketchily, I take

    up the issue of correcly predicting linear order for the structures considered here.

    1. Types of Complex Predicate

    One of the main points I would like to establish is that complex predicates formed with -key-

    marking on V

    1

    are dierent in character from all other complex predicates. First of all, I begin

    with a summary of the range of properties that -key displays, and the constructions in which it

    typically appears.

    1.1. The Status of -key

    The -key-form of a verb is often considered to be an adverbial form. However, this terminology

    obscures two dierent functions that such a form might have. As argued in Cho and Sells (1995)

    and Sells (1995), the general case in Korean is that all words bear a nal morpheme indicating how

    they participate in the structure. For verbs, for example, there are the prenominal endings like

    -(n)un which indicate that the verb is in construction with a following noun or nominal constituent.

    Similarly, the endings like -key , -e/-a, and so on, indicate that the verb is in construction with

    another verb; in Cho and Sells (1995) we referred to such elements as COMPs.

    This purely syntactic function is independent of the semantic relation between the verb bearing

    a COMP and the following verb that it is in construction with; in this syntactic sense, any COMP

    produces a form that is an ad-verb. However, on the semantic side of things, there seems to be

    fairly good evidence that -key-forms are not semantic adverbs or adverbials. I base what follows on

    the arguments in Jang (1997), who argues \that while ppalli -type is a real adverb, ppalu-key-type

    is in fact a (secondary) predicate and not an adverb. I provide evidence for this claim from small

    clause, depictive clause, purposive clause, resultative clause, and causative clauses" (p. 153). The

    examples are shown in (6){(10).

    3

  • (6) na-nun ku-lul solcikha-key (*solcikhi) sayngkakha-n-ta

    I-TOP he-ACC honest (*honestly) think-PROC-DECL

    `I consider him honest.'

    (7) John-i mwulkoki-lul sinsenha-key (*sinsenhi) mek-ess-ta

    John-NOM sh-ACC fresh (*freshly) eat-PAST-DECL

    `John ate the sh fresh.'

    (8) John-un kkangthong-ul napcakha-key (*napcaki) nwul-ess-ta

    John-TOP can-ACC at (*atly) press-PAST-DECL

    `John pressed the cat at.'

    (9) pang-i siwenha-key (*siwenhi), mwun-ul yele-noha-la

    room-NOM cool (*coolly), door-ACC open-IMP

    `To make the room cool, leave the door open.'

    (10) twu kitwung sai-uy keli-lul mel-key (*melli) hay-la

    two pole between-GEN distance-ACC distant (*distantly) do-IMP

    `Make the distance between the two poles great.'

    Putting the semantic distinction crudely, we can say that the true adverbs are predicated of an

    event, while the -key-predicates are predicated of an individual.

    In additional to causative and resultative constructions, -key also appears in other constructions,

    as in (11) and (12).

    (11) a. komap-key kwul-ta

    kind-COMP treat-DECL

    `treat (someone) kindly'

    b. mip-key kwul-ta

    hateful-COMP treat-DECL

    `behave hatefully (towards someone)'

    The complex predicate -key poi-ta (`seem/look like') alternates with -e/-a on V

    1

    , though the two

    have slightly dierent meanings (examples based on those in Martin (1992, 613)).

    (12) a. alay salam-tul-i cham cak-key poi-n-ta

    there person-PLU-NOM very small-COMP look-PROC-DECL

    `The people down there look very small.'

    b. alay salam-tul-i cham cak-a poi-n-ta

    there person-PLU-NOM very small-COMP look-PROC-DECL

    `The people down there look very small.'

    The second type shows evidence of being a closer unit; for example, short form negation may not

    intervene between the two parts.

    (13) a. alay salam-tul-i cham cak-key an poi-n-ta

    there person-PLU-NOM very small-COMP NEG look-PROC-DECL

    `The people down there do not look very small.'

    4

  • b. *alay salam-tul-i cham cak-a an poi-n-ta

    there person-PLU-NOM very small-COMP NEG look-PROC-DECL

    `The people down there do not look very small.'

    One generalization that will emerge from this paper is that V

    1

    -key is never forced to be absolutely

    adjacent to V

    2

    , as illustrated by this `negative intervention', unlike what is generally the case with

    the other complex predicates. This kind of dierence between complex predicates with -key and

    those with some other COMP will gure prominently in this paper.

    What should we call -key? Jang calls it a `predicativizer', to emphasize that V-key is predicated of

    individuals, but this seems misleading to me. My reason for this is that V in V-key is predicated

    of individuals anyway, so -key is not adding this property. I will argue below that -key is the least

    marked COMP available for a verb|that is, it marks the minimal morphological operation on a

    verb allowing it to appear in construction with another verb, which is a purely syntactic function.

    In order to distinguish it from other COMPs, I will refer to -key as a `marker'.

    1.2. Freely Composed Complex Predicates vs. Lexically Governed Complex

    Predicates

    One parameter of complex predicates that I wish to focus on in this paper concerns whether the

    relation between V

    1

    and V

    2

    is one of lexical government (from V

    2

    ), or whether the argument

    structures of the two verbs are composed by a more general mechanism.

    We have seen that Korean uses -key in causative and resultative constructions, among others, and

    that -key marks a basic predicate of individuals. I will propose in this paper that these complex

    predicates are freely composed, rather than being lexically governed. This makes obvious sense

    for the resultatives, and I will argue that it is the resultatives which show the true nature of -key .

    This marker simply provides the syntactic linking mechanism for V

    1

    ; all the other Korean COMPs

    provide more information in, and constraints on, the structure in which they appear.

    All other complex predicates involve true constructional combinations: all true COMPs fall in these

    constructions, where the argument structure and other semantic properties of the combination are

    not really independently motivated outside of the complex predicate in question. The complex

    predicates expressing ability, desire, and many others fall into this class, and the evidence for this

    claim comes from various types of argument structure sharing, and from case alternations on the

    object of V

    1

    .

    1.3. Constructionally-Determined Argument Structures

    1.3.1. General Considerations

    It seems uncontroversial that many complex predicates involve some kind of constructional merger

    of the argument structures of V

    1

    and V

    2

    .

    1

    Let us look at a simple illustrative example in (14); the

    structure that I will assume is shown as well. (The syntactic details of the structure are discussed

    below.)

    1

    For an in-depth semantic account of certain complex predicates, using the Jackendo's framework of Conceptual

    Structures, see Butt (1997).

    5

  • (14) a. inho-nun chayk-ul ilk-e po-ass-e-yo

    Inho-TOP book-ACC read-COMP try-PAST-POL

    `Inho tried reading a book.'

    b. V

    0

    N

    0

    V

    0

    inho-nun N

    0

    V

    0

    chayk-ul V

    0

    V

    0

    ilk-e po-ass-e-yo

    Here it seems that po-ta takes an agentive subject but inherits its other arguments from V

    1

    , and

    hence the object chayk-ul appears not as the object of ilk-e, but rather as the object of the verbal

    complex.

    The examples in (15) involve aspectual properties in which the major semantic properties of V

    1

    are directly inherited to the whole verbal complex. In these cases, it is clear that while V

    2

    is the

    syntactic head of the formation, V

    1

    is in some sense the semantic head.

    (15) a. ilk-ko iss-e-yo

    read-COMP PROG-POL

    `is reading'

    b. kh-e ci-ta

    big-COMP become-DECL

    `becomes big'

    Moreover, even apparently simple cases mask signicant complexity. For example, while ci-ta in

    (15)c indicates a change of state with a stative predicate as V

    1

    , the same form can appear with

    intransitive and transitive active predicates, as in (16) (from Lee (1993, 112.)).

    (16) a. ku mwul-i malk-a ci-ta

    that water-NOM clear-COMP become-DECL

    `That water becomes clear.'

    b. ku manh-un salam-tul-i ku pang-eyse ca ci-ess-ta

    that many people-PLU-NOM that room-in sleep.COMP become-PAST-DECL

    `That many people managed to sleep in the room.'

    c. kangmwul-i mak-a ci-ess-ta

    river-NOM block-COMP become-PAST-DECL

    `The river became blocked.'

    While there is some general `change of state' notion that can be applied to all three of these

    examples, it is far from trivial to simply say that V+ci-ta has a straightforward and predictable

    semantics.

    Some very interesting examples of the compositional nature of some complex predicates are dis-

    cussed in Lee (1998a). Looking at what kinds of argument structures allow for Locative Alternation

    6

  • in Korean and English, she observes that neither noh-ta as a main predicate, nor ssah-ta as a main

    predicate, allow the locative alternation, illustrated in (17) for the latter.

    (17) a. minswu-nun enehak chayk-tul-ul chaykcang-ey ssah-ass-ta

    Minsoo-TOP linguistics book-PLU-ACC bookshelf-LOC pile-PAST-DECL

    `Minsoo piled Linguistics books on the bookshelf.'

    b. *minswu-nun chaykcang-ul enehak chayk-tul-lo ssah-ass-ta

    Minsoo-TOP bookshelf-LOC linguistics book-PLU-ACC pile-PAST-DECL

    `Minsoo piled the bookshelf with Linguistics books.'

    However, using V-e noh-ta (`do in preparation for the future'), the complex predicate ssah-a noh-ta

    does allow Locative Alternation, as the two examples in (18) show.

    (18) a. minswu-nun enehak chayk-tul-ul chaykcang-ey ssah-a noh-ass-ta

    Minsoo-TOP linguistics book-PLU-ACC bookshelf-LOC pile-COMP put-PAST-DECL

    `Minsoo piled Linguistics books on the bookshelf.'

    b. minswu-nun chaykcang-ul enehak chayk-tul-lo ssah-a noh-ass-ta

    Minsoo-TOP bookshelf-LOC linguistics book-PLU-ACC pile-COMP put-PAST-DECL

    `Minsoo piled the bookshelf with Linguistics books.'

    From these kinds of examples, we have straightforward evidence of the verbal complex having

    properties which are not directly attributable to V

    1

    , but rather seem to come from the combination

    of V

    1

    with the specic V

    2

    which selects for it; and V

    2

    with this particular interpretation may not

    have any other usage outside of complex predicate in question.

    1.3.2. Desideratives

    There are other complex predicates which also show unusual semantic and/or syntactic properties.

    Desideratives, in both Korean and Japanese, show a case alternation on the accusative object of a

    transitive V

    1

    , even though this is not possible with V

    1

    alone.

    (19) a. ku yenghwa-lul/-ka po-ko siph-ta

    that movie-ACC/-NOM see-COMP want-DECL

    `(I) want to see that movie.'

    b. eego-o/-ga hanasi-tai

    English-ACC/-NOM speak-want.PRES

    `(I) want to speak English.'

    This alternation in case arises when an active predicate is derived into a stative one, largely re-

    gardless of the structural formation. The Japanese desiderative in (19)b is derived morphologically,

    while the Korean one is a verbal complex. It has been assumed that the version with the nominative

    object involves some kind of `restructuring' or `incorporation', but there is no evidence for this,

    apart from the case marking facts, even in examples with the nominative object. Specically, by

    standard tests of lexicality, there can be no incorporation involved, although the two verbs do form

    7

  • a tight unit (Cho (1988)); V

    1

    is followed by the complementizer -ko, which can in turn be followed

    by various emphatic particles, such as focus -nun, or -man, meaning `only'. These particles cannot

    appear inside true (= morphologically formed) words (Cho and Sells (1995)).

    In fact, the position of the `short form' negative an in (20) can perhaps be taken to indicate that

    the structural relation between the embedded verb and the `want' part is the same regardless of

    the case on the object. The crucial point of these examples is that the negation has scope over the

    nal verb siph-ta in either case.

    (20) a. na-nun pap-ul an mek-ko siph-ta

    I-TOP rice-ACC NEG eat-COMP want-DECL

    `I do not want to eat rice.'

    b. na-nun pap-i an mek-ko siph-ta

    I-TOP rice-NOM NEG eat-COMP want-DECL

    `I do not want to eat rice.'

    We see then that V

    1

    has dierent case marking properties in the complex predicate than it does

    alone.

    1.3.3. Abilitatives/Potentials

    Broadly speaking, the expression of ability, sometimes referred to as the `potential', shows similar

    case properties to the desiderative. However, the expression of ability takes the form of a very

    complex complex predicate in Korean, which does not allow for a case alternation on the object.

    (21) chelswu-nun ku chayk-ul/*-i ilk-ul swu iss-e-yo

    Chelsoo-TOP that book-ACC/*-NOM read be.able-POL

    `Chelsoo can read that book.'

    As noted for (2) above, the Japanese complex periphrastic potential does allow a case alternation

    on the object; and so does the morphological potential:

    (22) eego-o/ga hanas-e-ru

    English-ACC/-NOM speak-can-PRES

    `(I) can speak English.'

    Again, the same argument structure properties|whatever they are!|are manifest in dierent types

    of morpho-syntactic expression of the complex predicate.

    1.3.4. Negation

    Finally, as is also well-known, in long form negation the V

    2

    anh-ta shares the value of stativity

    that V

    1

    has. A non-stative predicate like mek-ta must have the processive morpheme, whether it

    8

  • is negated or not; a stative predicate like coh-ta must not have the processive morpheme.

    2

    These

    facts are shown in (23).

    (23) a. mek-ci anh*(-nun)-ta

    eat-COMP NEG*(-PROC)-DECL

    `does not eat'

    b. coh-ci anh(*-nun)-ta

    good-COMP NEG(*-PROC)-DECL

    `is not good'

    1.4. Summary

    In summary, all of the complex predicates discussed in section 1.3 show syntactic and/or semantic

    properties that are somewhat unexpected and are also somewhat specic to the construction in

    question. I take these properties to be criterial of lexically governed complex predicates.

    2. Subtypes of Lexically Governed Complex Predicate

    In this section I want to consider what the types of semantic relationship between V

    2

    to V

    1

    can be,

    in a governed complex predicate. Having established these semantic dierences, I will then argue

    in section 3 that all governed complex predicates behave alike with respect to their interaction with

    short form negation, and that this interaction is indicative of a uniform syntactic structure.

    2.1. Honorication

    In this subsection I will look at the appearance of the honoric marker -(u)si- on the two verbs

    in the complex.

    3

    I will not present data with the causative -key hata here, but, as is well-known,

    honorication is possible in both places, and dierentially honors the causee or the causer. In certain

    complex predicates, the honoric morpheme cannot be attached to V

    1

    , due to a morphological

    condition. This is the well-known condition that if verb in the V

    1

    position has the `continuative'

    COMP -e/-a, there can be no marker of honorication, as the two morphemes compete for the

    same slot, as seen in (24)a. However, qualied by this restriction, many of the verbal complexes

    allow the honoric morpheme in both places, as in (24)b/c, with the COMPs -ko or -ci . In all cases

    honorication must be present on V

    2

    (if the subject is to be honored), and optional on V

    1

    if it is

    morphologically allowed.

    (24) Honoric: V

    1

    = blocked, V

    2

    = obl.

    ilk(*-usi)-e po-si-ta

    read(*-HON)-COMP try-HON-DECL

    `[someone honorable] tries to read'

    What this data shows is that the honoree is a dependent of both verbs; roughly, the subject

    argument is shared between the two of them. From our intuitions of the semantics, we might say

    2

    If the general approach to the semantics of negation outlined below is correct, in which the CONTENT of the

    negative verb is identied with the content of V

    1

    , this property of sharing the value of stativity would follow as a

    consequence.

    3

    See also Sells (1991) and Martin (1992, 226.).

    9

  • that po-ta in (24) is a `control' predicate: it seems reasonable that the subject volitionally tries out

    the action denoted by V

    1

    .

    Let us look at some other examples now.

    (25) Honoric: V

    1

    = opt., V

    2

    = obl.

    a. ilk(-usi)-ko kyeysi-ta

    read(-HON)-COMP be.HON-DECL

    `[someone honorable] is reading'

    b. ilk(-usi)-ci anh-usi-ta

    read(-HON)-COMP NEG-HON-DECL

    `[someone honorable] does not read'

    Here, the V

    2

    predicates are dierent from po-ta in (24), and probably are not predicates at all, in

    the semantic sense. Rather, they just provide aspectual or polarity information. As suggested to

    me by Ivan Sag, it seems reasonable to propose for these cases that the semantic content of V

    1

    and V

    2

    are identied, with V

    2

    just providing a little extra information.

    4

    If this is correct, the HON

    marking again reects a semantic relationship between the predicate on which it is marked, and

    the semantic subject of that predicate. For want of a better term, I will refer to these types as

    `modier' complex predicates.

    5

    Honorication is also possible on the V

    2

    s in (26) (see also (12) above).

    (26) Honoric: V

    1

    = blocked, V

    2

    = obl.

    coh-a poi-si-ta

    good-COMP seem-HON-DECL

    `[someone honorable] seems good'

    This looks like a raising predicate, in which case poi-ta would assign no semantic role to its subject.

    I will discuss true raising predicates below, but, at least for this specic construction, it seems

    as though the best English translation might be `gives the appearance of', rather than `seems',

    suggesting a weak thematic relationship between poi-ta and the subject. Martin translates V-e poi-

    ta as `looks like, appears to be', while he translates V-key poi-ta as `looks, seems'. (27) would then

    conform to the generalization that honoric marking on V

    2

    correlates with a semantic relationship

    between the semantic content of V

    2

    and its surface subject.

    The patterns of honoric marking become more interesting when we look at other complex predi-

    cates which intuitively involve a propositional operator such as `seem' or `must'. In these cases, in

    contrast to (26), the honoric sux must be attached to V

    1

    , and cannot appear on V

    2

    , as shown

    in (27).

    6

    4

    For example, Bratt (1996, 225) species for an that \the entire CONTENT of the negated verb is unied with

    the CONTENT of the . . . negative particle, except for the POLARITY", which is changed from 1 (positive) to 0

    (negative).

    5

    Probably the pure inchoative uses of V-e ci-ta fall into this category too.

    6

    In Sells (1991) I drew a distinction between `individual raising' for the V-e poi-ta type and `propositional raising'

    for the V-na po-ta type. I now think that this was mistaken.

    10

  • (27) Honoric: V

    1

    = obl., V

    2

    = *

    a. ilk-usi-eya ha(*-si)-ess-ta

    read-HON-COMP must(*-HON)-PAST-DECL

    `[someone honorable] had to read'

    b. ilk-usi-na po(*-si)-ta

    read-HON-COMP seem(*-HON)-DECL

    `[someone honorable] seems to read'

    I think the obvious analysis of these cases is that the V

    2

    verb represents a propositional operator

    over V

    1

    , with the subject `lowered' semantically to be the subject of V

    1

    . Hence, the subject

    has no semantic relation to the CONTENT of V

    2

    , and honoric marking therefore cannot appear.

    Interestingly, the COMP -e/-a does not seem to appear in any pure raising complex predicate.

    2.2. Summary

    From these and other considerations, we can identify at least the following dierent types of gov-

    erned complex predicate: control, raising, and modier.

    7

    These are represented in (28){(30).

    (28) Control Type (e.g., V-e po-ta)

    2

    6

    6

    6

    6

    6

    6

    6

    6

    6

    6

    6

    6

    6

    6

    6

    4

    HEAD verb

    SUBJ

    D

    NP

    i

    E

    COMPS

    *

    V

    2

    6

    6

    4

    SUBJ

    D

    PRO

    i

    E

    COMPS

    2

    CONT

    3

    3

    7

    7

    5

    +

    2

    CONT

    2

    4

    RELN control rel

    ROLE i

    SOA-ARG

    3

    3

    5

    3

    7

    7

    7

    7

    7

    7

    7

    7

    7

    7

    7

    7

    7

    7

    7

    5

    Here I use the idea of PRO on a list to express an unrealized argument, following Manning and

    Sag (1998) and Manning, Sag and Iida (1998).

    8

    (28) says that verbs of this type express a relation

    which assigns a role to their highest argument (index i), and that this argument is coindexed with

    the subject of the verbal complement (although the relation may be quite a `weak' one, such as

    `show signs of'). Any complements (indicated by

    2

    ) of the verbal complement are inherited as

    complements of the selecting verb, and appear outside of the verbal complex, as seen in (14)b

    above.

    The raising type is essentially identical, except that the relation that is involved|glossed here as

    an `operator' relation|does not assign a role to the subject of verb (indicated by

    1

    ), and that

    7

    In this paper, I treat `NP', `N

    0

    ', and `N[BAR 1]' as equivalent, as I follow Fukui (1986) in assuming that phrases

    in Korean (and Japanese) only project to a single-bar level. I use `NP' in the HPSG lexical entries and rules for

    continuity with other HPSG work.

    8

    Lee (1998b) presents an account of Korean causatives based on the analysis of Japanese in Manning, Sag and

    Iida (1998).

    11

  • subject is also the subject of the verbal complement.

    (29) Raising Type (e.g., V-na po-ta)

    2

    6

    6

    6

    6

    6

    6

    6

    6

    6

    6

    6

    6

    6

    4

    HEAD verb

    SUBJ

    D

    1

    E

    COMPS

    *

    V

    2

    6

    6

    4

    SUBJ

    D

    1

    E

    COMPS

    2

    CONT

    3

    3

    7

    7

    5

    +

    2

    CONT

    RELN operator rel

    SOA-ARG

    3

    3

    7

    7

    7

    7

    7

    7

    7

    7

    7

    7

    7

    7

    7

    5

    The modier type has yet dierent semantics: the semantics of the verb (such as iss-ta) are iden-

    tied with the semantics of the verbal complement, except for some small modication, indicated

    informally by `n[ . . . ]'. This small modication may be the expression of progressive aspect, the

    reversal of polarity from positive to negative, and so on.

    (30) Modier Type (e.g., V-ko iss-ta)

    2

    6

    6

    6

    6

    6

    6

    6

    6

    6

    6

    6

    4

    HEAD verb

    SUBJ

    D

    1

    E

    COMPS

    *

    V

    2

    6

    6

    4

    SUBJ

    D

    1

    E

    COMPS

    2

    CONT

    3

    3

    7

    7

    5

    +

    2

    CONT

    3

    n[ . . . ]

    3

    7

    7

    7

    7

    7

    7

    7

    7

    7

    7

    7

    5

    In the remainder of this paper, I want to show how these three types of complex predicate nev-

    ertheless show a unity, which distinguishes them from structures involving the marker -key , even

    though they show dierent kinds of semantic relationships within the complex predicates.

    3. Syntactic Connection in Governed Complex Predicates

    3.1. Negative Placement

    There is a very interesting property of all complex predicates except for the causative, which I

    discussed in Sells (1991): with the exception of the causative, none of them allow the short form

    negation an to intervene between V

    1

    and V

    2

    .

    9

    In fact, the generalization as I stated it in that paper

    is not quite correct: the real generalization is that only if V

    1

    is marked with -key can an immediately

    precede V

    2

    , regardless of the semantic relation between the verbs. An abstract characterization of

    the facts is given schematically in (31).

    9

    This property is noted for the desiderative complex predicate in Martin (1992, 336).

    12

  • (31) Negative Placement (illustrated with an)

    a. an V

    1

    - V

    2

    :

    possible with = COMP or -key ; negation scopes over V

    1

    or the whole verbal complex.

    b. V

    1

    - an V

    2

    :

    possible with = -key ; bad with any COMP; negation scopes over V

    2

    .

    c. V

    1

    -ci anh- V

    2

    :

    possible with = COMP or -key ; negation scopes over V

    1

    .

    d. V

    1

    - V

    2

    -ci anh-ta:

    possible with = COMP or -key ; negation scopes over V

    2

    .

    We can see from (c) and (d) that long form negation is always possible, so that is less interesting

    with regard to dierences in complex predicates. To see how to interpret (31)a{b, let us begin with

    the causative -key hata examples in (32). Here, either verb may be negated (see Song (1988)) by

    short form negation.

    (32) a. an ilk-key ha-ta (negating V

    1

    )

    NEG read-MRKR do-DECL

    `cause someone to not read'

    b. ilk-key an ha-ta (negating V

    2

    )

    read-MRKR NEG do-DECL

    `not cause someone to read'

    c. an ca-key mos ha-ta (independent negation of both)

    NEG sleep-MRKR cannot do-DECL

    `be unable to cause someone to not sleep'

    d. kimchi-lul mos mek-key ha-ta (negating V

    2

    )

    kimchee-ACC cannot eat-MRKR do-DECL

    `cannot make someone eat kimchee' (from Bratt (1996, 228))

    Note that negation of V

    2

    can either immediately precede V

    1

    , as in (d), or V

    2

    , as in (b). The possible

    intervention as in (b) is not a fact about the causative per se; for the other constructions with the

    MRKR -key , intervening negation is also possible, as in (33) and (13)a above.

    (33) komap-key an kwul-ta

    kind-MRKR NEG treat-DECL

    `does not treat (someone) kindly'

    In sharp contrast, with complex predicates which have a COMP on V

    1

    , the short form negation may

    not intervene.

    (34) Negating V

    2

    13

  • a. *ilk-e an po-ta

    read-COMP NEG try-DECL

    `not try to read'

    b. *ilk-e an cwu-ta

    read-COMP NEG give-DECL

    `not give the favor of reading'

    This suggests that negation somehow cannot `interrupt' a governed complex predicate. Importantly,

    intervening short form negation is bad, regardless of the semantic type of the complex predicate.

    (34){(35) illustrate this for control complex predicates; (35){(36) illustrate it for the other types of

    complex predicate.

    (35) Control Complex Predicate

    a. mek-e po-ta `tries eating'

    b. *mek-e an po-ta `does not try eating'

    c. coh-a poi-ta `seems good'

    d. *coh-a an poi-ta `does not seem good'

    (36) Raising Complex Predicate

    a. phyenci-lul ssu-na po-ta `seems to write a letter'

    b. *phyenci-lul ssu-na an po-ta `does not seem to write a letter'

    c. ilk-eya ha-yess-ta `had to read'

    d. *ilk-eya an ha-yess-ta `did not have to read'

    (37) Modifer Complex Predicate

    a. ilk-ko iss-ta `is reading'

    b. *ilk-ko an iss-ta `is not reading'

    c. yeypp-e ci-ess-ta `became pretty'

    d. *yeypp-e an ci-ess-ta `did not become pretty'

    However, it is not the case that V

    2

    cannot be negated; long form negation is always possible. More-

    over, examples like those above do have grammatical variants, but with the short form negation in

    front of V

    1

    , as shown in (38), even though the scope of negation includes V

    2

    . All of the ungram-

    matical examples in (34){(37) become grammatical if the short form negation is placed in front of

    the whole complex.

    (38) Negation preceding V

    1

    a. an ilk-e po-ta

    NEG read-COMP try-DECL

    `not try to read'

    b. an ilk-ko iss-ta

    NEG read-COMP PROG-DECL

    `is not reading'

    c. an coh-a poi-ta

    NEG good-COMP seem-DECL

    `does not seem good'

    d. an ilk-eya ha-ta

    NEG read-COMP must-DECL

    `must not read'

    14

  • Strictly speaking, in these examples, there is an ambiguity of the attachment of negation: is it just

    V

    1

    that is negated, or is it the constituent which consists of the two verbs? In general, it is possible

    for negation preceding V

    1

    to take scope over the whole complex predicate, as a complement of V

    2

    ,

    though one interpretation may be preferred over the other. In (38)d, though, negation only has

    narrow scope, though this fact is not peculiar to Korean; in English, while modals and negation

    generally enter into scope ambiguities, the example John must not read also only allows the reading

    with narrower scope for the negation (Kim (1995, 141.)).

    This leads us to the following general conclusions for governed complex predicates.

    (39) In a V

    1

    {V

    2

    governed complex predicate:

    a. Short form negation of V

    1

    immediately precedes V

    1

    .

    b. Short form negation of V

    2

    immediately precedes V

    1

    .

    In constrast, with a free complex predicate, using the MRKR -key , negation of V

    2

    can immediately

    precede V

    2

    , as would normally be expected.

    Turning to long form negation, it turns out that long form negation is always possible. The examples

    in (40) have the negation following V

    2

    , which is interpreted as the negation scoping over the verbal

    complex.

    (40) Negation following V

    2

    a. ilk-e po-ci anh-ta

    read-COMP try-COMP NEG-DECL

    `not try to read'

    b. ilk-e cwu-ci anh-ta

    read-COMP give-COMP NEG-DECL

    `not give the favor of reading'

    c. ilk-eya ha-ci anh-ta

    read-COMP must-COMP NEG-DECL

    `does not have to read'

    Alternatively, long form negation can immediately follow V

    1

    , in which case it takes scope over just

    V

    1

    .

    (41) Negation following V

    1

    a. ilk-ci anh-a po-ta

    read-COMP NEG-COMP try-DECL

    `try not to read'

    b. ilk-ci anh-a cwu-ta

    read-COMP NEG-COMP give-DECL

    `give the favor of not reading'

    15

  • c. ilk-ci anh-a-ya ha-ta

    read-COMP NEG-COMP must-DECL

    `must not read'

    As far as I am aware, long form negation is always possible, and the scope of the negation is

    unambiguously over the verb that the negation immediately follows.

    3.2. The Analysis of Short Form Negation

    Now let us consider in more detail the analysis of short form negation. The key facts to be account

    for are summarized again in (42):

    (42) Negative Placement (illustrated with an)

    a. an V

    1

    - V

    2

    :

    possible with = COMP or -key ; negation scopes over V

    1

    or the whole verbal complex.

    b. V

    1

    - an V

    2

    :

    possible with = -key ; bad with any COMP; negation scopes over V

    2

    .

    We see in (42)b that if V

    1

    bears a COMP and is selected by V

    2

    , then an cannot appear between

    them. In contrast, if V

    1

    is marked with -key , negation can intervene, suggesting that V

    1

    is not

    selected by V

    2

    . Accepting these ideas, the question now becomes, why would an interfere with

    verbal selection?

    Let us rst briey consider previous approaches to an. I have argued elsewhere (Sells (1994), Sells

    (1998)) that an should be treated as a negative adjunct, attaching at the V

    0

    level. In case two or

    more verbs are combined in a verbal complex, this approach would predict an should be able to

    appear as an adjunct to either V

    0

    part, or to the whole; an could negate any of the three verbal

    nodes in (4).

    Kim (1995) argues that an has an even tighter relationship, as verbal prex, which would mean

    that an ought to be able to appear before either verb in a complex predicate, negating that verb,

    but could never negate the whole complex.

    Government-Binding and Minimalist approaches using NegP would treat an typically as a head

    of NegP, selecting a VP as its complement (for a recent account, see Hagstrom (1996)). Broadly

    speaking, it seems to me that one problem with such a proposal is that short form negation shows

    no evidence whatsoever of taking (syntactic) scope over a VP; short form negation never appears

    before phrasal complements of a verb, but rather follows them.

    All of these approaches have nothing to say about the intervention eect summarized in (42),

    and nothing to say about the dierence between -key and the COMPs. This suggests that these

    proposals|that an is an adjunct, a prex, or a higher head|are all wrong. Rather, there is a

    fourth alternative, which provides a straightforward account of the interaction in (42). This is the

    idea that I will pursue here, namely that the negative is a complement of the verb. This is the

    direct opposite of the NegP approach, in which the content verb is a complement to negation; of

    course, for long form negation, the content verb is a complement.

    The idea that negation can be expressed as a complement rather than as an adverb is explored for

    English and French in Kim (1995). Here I would like to extend that hypothesis to Korean short

    16

  • form negation. In fact, I would like to suggest that an is never an adjunct (in contrast to what I

    have argued previously), but is always a complement.

    3.3. Negation as a Complement

    For a language like English, the prototypical expression of negation is as a VP-modier, like many

    other adverbs. For example, this is a plausible analysis of examples like those in (43), with the

    structure shown:

    (43) a. Max has [

    VP

    not [

    VP

    eaten kimchee]].

    b. Max has [

    VP

    never [

    VP

    eaten kimchee]].

    Taking a view of syntax which does not use empty categories, this straightforwardly explains why

    an example like (44) is bad: never as no VP to modify.

    (44) *Sam has eaten kimchee, but Max has never.

    However, not can be stranded preceding a `VP-Deletion' site, in contrast to never :

    (45) Sam has eaten kimchee, but Max has not.

    To account for this, Kim (1995) and Kim and Sag (1996) proposed that not can function either as

    a VP-adverbial, as in (43)b, or as a complement selected by an auxiliary, in (45). Hence not would

    be sanctioned in the structure in (45) by the preceding auxiliary verb. To allow for this possibility,

    there is a lexical rule which adds negation as a complement to the auxiliary verb, even though the

    negation functions semantically as a modier on the verb's meaning. As an auxiliary in English

    already takes a non-nite VP complement, the output of the lexical rule will be an auxiliary verb

    which takes two complements, though only the VP will be a semantic argument of the auxiliary.

    10

    In contrast to English, in French all nite verbs may take a negative complement (pas in French).

    In Korean, negation is expressed in a rather dierent way. As noted above, there is no evidence

    that negation is a VP-adverbial of any kind; to be precise, I should say that this is true for adults:

    there is some evidence that some children produce VP-modifying negative structures (see Cho and

    Hong (1988) and Kim (1997)), but then retreat from this position. Instead, negation in Korean is

    expressed either as a verb which takes a preceding non-nite V complement, with the long form

    anh-ta, or it is the short form an, which I claim can only enter into the structure in one way: as a

    complement. To account for the latter type, we need a lexical rule which adds a negative element

    as an X

    0

    complement to any verb.

    10

    There have been many developments in HPSG since Pollard and Sag (1994) regarding the way that valence

    information is determined and represented. These developments do not directly bear on my more general proposals

    here, and for ease and clarity of presentation, I will assume that each verb has a SUBJ list (of maximum length 1)

    for its subject and a COMPS list for the non-subject dependents.

    17

  • (46) Korean Short Form Negation Lexical Rule

    2

    4

    HEAD verb

    COMPS

    1

    CONT

    2

    3

    5

    )

    2

    6

    4

    COMPS

    D

    Neg

    0

    :

    3

    E

    1

    CONT

    3

    h

    ARG

    2

    i

    3

    7

    5

    This rule adds a Neg

    0

    such as an or mos as the rst element on the COMPS list, and applies the

    semantics of that negation to the verb's CONTENT.

    Consider now the case where we have a complex predicate of the form V

    1

    {V

    2

    , where V

    1

    is a

    complement of V

    2

    , and we want to negate V

    2

    . By assumption, we use the lexical rule to add an as

    a complement to V

    2

    , so then V

    2

    will have two complements, and in fact both complements are X

    0

    categories.

    (47) COMPS list of negated Korean V

    2

    COMPS < Neg

    0

    , V

    0

    >

    At this point, we need to make one assumption that is not (yet) motivated independently of the

    facts that we are trying to explain. That assumption is that a head combines with all of its X

    0

    complements at once; it is certainly a reasonable assumption, one that feeds directly into a nice

    account of the negative intervention facts. If a verb with a COMPS list like that shown in (47)

    takes all of its X

    0

    complements at once, we will have a ternary structure, and the facts about

    negative intervention simply reduce to the fact that the Neg

    0

    complement of V

    2

    must precede the

    V

    1

    complement.

    11

    This will give the structure shown in (48), assuming that the negation can be

    properly ordered before V

    1

    .

    (48) Negating V

    2

    : V

    0

    Neg

    0

    V

    0

    V

    0

    an mek-e po-ta

    On the other hand, if only V

    1

    in a complex predicate is negated, then it will combine with its single

    (negative) complement, and then V

    2

    will combine with that syntactically formed V

    0

    complement,

    as shown in (49).

    11

    There is a technical question within HPSG about the relative order of complements on the COMPS list and

    the order in which those elements are cancelled o in the syntax. Here I simply assume that the structures which I

    propose can indeed by licensed by (some elaboration of) the lexical entries and schemata I give.

    18

  • (49) Negating V

    1

    : V

    0

    V

    0

    V

    0

    Neg

    0

    V

    0

    po-ta

    an mek-e

    As far as I can tell, the key to nding an explanation for the puzzling fact of the negative intervention

    eect relies crucially on the availability of a structure like that in (48): it is denitely possible to

    negate V

    2

    , but when you do that, the negation does not appear directly in front of V

    2

    . I cannot

    see that any prexal, adjunct, or higher head analysis of an would give this property; eectively,

    the central feature of my proposal here is that negation is lower than V, and therefore enters into

    the structure as a complement.

    3.4. The Theory of Syntactic Structures

    Let us now consider how to correctly license the simple syntactic structures just discussed. Pre-

    vious work (Sells (1994), Abeille and Godard (1998), Sadler and Arnold (1994)) has shown that

    it is necessary to distinguish three types of element in the syntax, which are illustrated in (50)

    along with the notation I will use to distinguish them. I assume the single-bar theory of regular

    Korean phrases (see Fukui (1986), Sells (1994), Sells (1995)), and also assume that the only major

    syntactic constituents are built around N or V, without any attendant functional categories such as

    INFL or COMP.

    12

    In between familiar words ((50)a) and phrases ((50)c), we have small syntactic

    constructions, such as verbal complexes.

    (50) Types in the Syntax

    a. Items drawn directly from the lexicon:

    LEX +

    BAR 0

    b. Items forming `small' syntactic phrases, e.g., verbal complexes:

    LEX

    BAR 0

    c. Regular full phrases:

    LEX

    BAR 1

    Only items directly drawn from the lexicon are [LEX +]. However, the small constructions share

    some properties with lexical items, captured here by treating them all as X

    0

    s, represented by [BAR

    0]. For example, lexical items and syntactically-formed X

    0

    s enter into blocking relationships, while

    full phrases do not (see Poser (1992), Sells (1998)). Anything composed in the syntax is a phrase

    in the sense that it is [LEX ]; but there are two types of phrase, depending on the specication

    of the BAR level.

    12

    The COMP elements discussed in this paper are all treated as verbal inectional axes, following Cho and Sells

    (1995).

    19

  • Now we need the following ID-schemata to license the necessary structures. Verbal complexes in

    HPSG analyses of Korean have been proposed by Chung (1993), Kim (1995) and Bratt (1996); the

    notation in each diers slightly, but the structures built are almost the same.

    (51) a. X[LEX ] ! C[BAR 0]*, H[LEX +]

    b. X[LEX ] ! C[BAR 1]*, H

    By schema (a), once a lexical item has combined with any [BAR 0] complements, it is specied as

    [LEX ] (by the left-hand side of the schema). Hence, the resulting structure can no longer be the

    head in recursive application of (a). The result of this is that a head must combine with all of its

    [BAR 0] complements at the same time.

    Schema (b) only allows combination with phrases, and provides the regular structures in the sen-

    tential syntax.

    13

    These two schemata will license the structures I propose in this paper. Although I will not discuss

    linear precedence conditions until section 6, it is crucial to my account of the negative intervention

    facts that all X

    0

    complements of a head are introduced at once, in a at structure, so that linear

    precedence statements will apply to them all at once.

    3.5. Summary

    To summarize this section briey, there are some important conclusions from the analysis of complex

    predicates for the theory of syntactic structures, enumerated in (52).

    (52) a. Negation can be the complement of V; negation is not always `higher' than V in the

    structure.

    b. A verb must be able to combine with more than one complement simultaneously; struc-

    tures cannot be limited to binary formations.

    Although I will not comment on the larger theoretical implications of these conclusions, they

    challenge long-held assumptions about the variety of ways in which negation can be expressed,

    and also current assumptions within the Minimalist Program (e.g., Chomsky (1995)) that the only

    process of structural formation is necessarily binary.

    4. Freely Composed Complex Predicates

    So far, I have concentrated on governed complex predicates. Now, I turn to complex predicates

    involving -key . My claim in this paper is that complex predicates with -key are not lexically

    governed, but rather are freely composed. As such complex predicates allow negative intervention,

    this shows us that V

    1

    is not a complement, or more precisely not a V

    0

    complement of V

    2

    in the

    13

    As stated, (51)b allows any conguration for fully phrasal structures, ranging from binary to at structures, and

    any intermediate congurations. I will not address this further, as I am concentrating almost exclusively on verbal

    complexes in this paper.

    20

  • way that the selected V

    0

    complements discussed above are complements. I will return presently to

    examples which seem to have selected complements, such as (11) above with kwul-ta.

    Each V

    2

    which appears in a governed complex predicate has a COMPS list of the form shown in

    (53)a: it selects for a zero-level V complement, bearing a specic COMP expressing the VFORM

    value . On the other hand, the heads which enter into free composition select for nothing more

    than a verbal complement in the rst instance|roughly speaking, this could be a verb, a VP, or a

    sentence|anything of category V.

    (53) Selectional types:

    a. Lexically Governing Heads

    COMPS:

    1

    b. Freely Composing Heads

    COMPS: < . . . ,

    h

    CAT V

    i

    , . . . >

    No VFORM is specied in (53)b, and as verbs bearing COMPs do not randomly appear in Korean

    sentences, it must be the case that VFORM is only present when sanctioned by a selecting predicate.

    As my assumption is that -key is not a COMP and therefore does not mark VFORM, -key cannot

    appear on the V

    0

    complement in (53)a, and no true COMP can appear on the free verbal complement

    in (53)b.

    4.1. Causatives

    I am now going to address the issue of the correct analysis of causatives presently, after relating

    that to the issue of the analysis of resultatives. However, to begin, with the canonical key ha-ta

    causative, there are clearly three options for the complement of ha-ta: it can be a full sentence

    (with a nominative causee), it can be VP complement, or it can be a V

    0

    complement appearing

    in a verbal complex (Bratt (1996) provides extensive evidence for the verbal complex possibility).

    In the latter two cases the causee can be marked with the accusative or dative. Assuming that

    we can get the semantics right, we can say that the causative ha-ta really does not care what its

    syntactic complement is. As the construction is not lexically governed, there must be some more

    general mechanism for providing the semantic connection between ha-ta and its complement. It is

    this mechanism that I refer to by `free composition'. Let us assume for the moment that ha-ta just

    requires some kind of verbal complement, without being more specic.

    Now it certainly is possible for a causative to be expressed as a verbal complex, and in some cases

    it is necessary, as with (54). Here, under the reading where negation scopes over both verbs, the

    structure must be that shown in (55), with negation and V

    1

    appearing as X

    0

    complements.

    (54) an [mek-key hay-ss-ta]

    NEG [eat-MRKR do-PAST-DECL]

    `did not cause to eat'

    21

  • (55) V

    0

    Neg

    0

    V

    0

    V

    0

    an mek-key hay-ss-ta

    A similar semantic interpretation can be obtained by having the negation immediately preceding V

    2

    ,

    and this example shows us that the V

    1

    complement is not a V

    0

    complement, but rather projects to

    the V

    0

    level, as shown in (56) for the example in (32)b. This is precisely the structure that governed

    complex predicates lack.

    (56) V

    0

    V

    0

    V

    0

    (chayk-ul) ilk-key Neg

    0

    V

    0

    an ha-ta

    From these examples we see that the causative can be expressed as a verbal complex, but does not

    need to be, indicating a certain `freedom' of composition not found with all the complex predicates

    discussed above.

    What is further interesting about the `freedom' of the causative is that it can be expressed via a

    variety of predicates, including mantul-ta and sikhi-ta:

    (57) yumi-ka inho-eykey pap-ul mek-key mantul-ess-ta

    Yumi-NOM Inho-DAT rice-ACC eat-MRKR cause-PAST-DECL

    `Yumi made Inho eat the rice.' (from Kim (1990, 162))

    (58) sensayngnim-i haksayng-eykey kongpwu-lul ha-key sikhi-ess-ta

    teacher-NOM student-DAT study-ACC do-MRKR cause-PAST-DECL

    `The teacher made the students study.'

    No governed complex predicate allows dierent verbs to be substituted in, preserving meaning.

    4.2. Resultatives

    As noted above in section 1, another use of -key is in resultative constructions. Interestingly, the

    apparently resultative examples in (59), from Lee (1993, 159, 371), are presented under the heading

    of `causative' formations|for good reason, I will argue.

    (59) a. na-nun ai-lul mengi tul-key ttayli-ess-ta

    I-TOP child-ACC black.and.blue-MRKR beat-PAST-DECL

    `I beat the child black and blue.'

    22

  • b. na-nun ai-lul wul-ci anh-key ttayli-ess-ta

    I-TOP child-ACC cry-COMP NEG-MRKR `beat'-PAST-DECL

    `I calmed the child to stop him from crying.'

    c. na-nun ku-lul nolla-key mil-ess-ta

    I-TOP he-ACC frighten-MRKR push-PAST-DECL

    `I pushed him to frighten him.'

    Why are these classied as resultatives, while the examples in the previous subsection are classied

    as causatives? Note that (59)b{c do not involve stative predicates in the true sense of stativity,

    and these examples do not work in English (e.g., *I stroked the child not crying , *I pushed him

    frightened), showing that they are not `canonical resultatives', if I can use that term. Is there really

    a dividing line between causatives and resultatives (in Korean), or are they all part of the same

    family of closely-related constructions?

    There are other similarities between resultatives and causatives that we can nd. As with causatives,

    resultatives can be expressed with a full sentence result complement containing an internal nom-

    inative subject (see Kim and Maling (1997) for a recent discussion). Kim and Maling treat the

    resultative complement as an adjunct in Korean, while I will treat it as a free complement, using

    the lexical rule in the following subsection.

    14

    In fact, resultatives in Korean look just like causatives, except that the nal verb is not ha-ta. Why

    should there be such a strong resemblance? My answer to this will be that a causative actually is

    a resultative, with ha-ta used, indicating the most general type of main (action) predicate.

    4.3. Free Composition

    From the HPSG perspective adopted here, to allow for resultatives, we need to have a lexical

    rule which adds a verbal complement to the COMPS list of any verb, and species the semantic

    relationship between the verb and the new complement. For simplicity, I restrict my discussion

    here to main verbs which are transitive, such as those in (59), and where the object controls the

    subject of the resultative complement; a fuller account should give a more generalized rule. The

    lexical rule is given in (60).

    15

    14

    I agree with Kim and Maling that resultatives in Korean do not radically alter the argument structure of main

    predicate as they do in English. To my mind, the key dierence is that, for all the freedom that Korean resultatives

    show, they precisely do not show the real argument-structure altering type with a `non-thematic object' like drink

    the bottle empty.

    15

    As stated, this rule allows the main verb to inherit any unsaturated complements of the added verbal complement:

    this corresponds to allowing either a verbal complex structure for the two verbs, or a structure in which the main

    verb takes the resultative as a VP complement.

    23

  • (60) Free Composition Lexical Rule: Resultative of Transitive

    2

    6

    6

    4

    HEAD verb

    COMPS

    D

    NP

    j

    E

    1

    CONT

    3

    3

    7

    7

    5

    )

    2

    6

    6

    6

    6

    6

    6

    6

    6

    6

    6

    4

    COMPS

    *

    V

    2

    6

    6

    4

    SUBJ

    D

    PRO

    j

    E

    COMPS

    2

    CONT

    4

    3

    7

    7

    5

    , NP

    j

    +

    1

    2

    CONT

    2

    4

    RELN cause

    CAUSE

    3

    EFFECT/RESULT

    4

    3

    5

    3

    7

    7

    7

    7

    7

    7

    7

    7

    7

    7

    5

    The semantics here follows the ideas in Dowty (1979) and Foley and van Valin (1984) where

    causation is a relation between two actions, or an action and a state. In (60), I use EFFECT to

    denote a caused action, and RESULT to denote a caused change of state; that is, a causative is

    standardly treated as [ CAUSE ], while a resultative is [ CAUSE [BECOME ]], and I intend

    (60) to cover both cases.

    Now, should we consider the examples in (59) to be complex predicates? I think that most linguists

    would feel that the answer is `no', at least, not compared to the complex predicates that I discussed

    in sections 1{3. So, (60) provides structures with a semantic relation between a verb and a verbal

    complement where the combination is not considered to be a complex predicate.

    What would happen if we had a very general verb like ha-ta, and let (60) apply to it? Then we

    would get a verb expressing a relation between X and Y, meaning that X did something to Y

    causing some eect on Y. This is precisely the standard meaning assigned to the causative, and it

    can be derived without treating the causative as a complex predicate. Naturally, if this is possible,

    with ha-ta, it is also possible with some other very general verbs such as sikhi-ta or mantul-ta.

    16

    4.4. The Argument Structure(s) of ha-ta

    Although I cannot defend it strongly here, I think it is quite plausible to think of ha-ta as being able

    to express the generalized action of doing something (to someone/something). Looking broadly at

    Korean, I think we could say that ha-ta can be a generalized verb of action, a generalized verb of

    mental process, and of course a pure supporting verb in forms like kkaykkus ha-ta (`be clean').

    For instance, all of the examples below illustrate ha-ta as a general verb of mental process, and of

    course with the quotative complementizer -ko, it means `say'.

    (61) na-nun [inho-ka o-ci anh-na] ha-n-ta

    I-TOP [Inho-NOM come-COMP NEG-COMP] do-PROC-DECL

    `I am afraid that Mary might come.' (Ki-Sun Hong, p.c.)

    All of the examples in (62) are from Lee (1993, 159).

    (62) a. il-i cal toy-ess-umyen ha-n-ta

    event-NOM well become-PAST-CONDIT do-PROC-DECL

    `I hope everything will turn out well.'

    16

    In (67) below, I assume that sikhi-ta is lexically specied as a causative predicate; but mantul-ta clearly has a

    use as a simple transitive verb.

    24

  • b. il-i kkuthna-ss-nunka ha-n-ta

    work-NOM end-PAST-Q do-PROC-DECL

    `I wonder if the work is nished.'

    c. nayil ku pun-i o-si-na ha-n-ta

    tomorrow that man-NOM come-HON-COMP do-PROC-DECL

    `I suspect that that man will come tomorrow.'

    Although these are in a dierent domain from pure action, I think they add plausibility to my

    claim that ha-ta is a verb expressing generalized activity in at least two canonical domains: action

    and mental process.

    17

    4.5. Summary: Freely Composed Predicates

    As I mentioned above, my steps towards an analysis in this section are based on the idea that

    the causative is a limiting case of the resultative. There are some important consequences of this

    assumption, following from the nature of resultative constructions. On the one hand, it is obvious

    that the way that a resultative construction is expressed varies from language to language. On

    the other hand, as far as I am aware, there are three striking universal properties of resultative

    constructions, enumerated in (63).

    (63) Universal properties of resultative constructions

    a. The main verb takes no special morphology marking `resultative' (unlike the causative

    in many languages).

    b. The resultative (secondary) predicate is in the simplest non-nite form that the language

    makes available.

    c. The syntactic formation is parasitic on other structures in the language; there is no

    special syntax for resultatives.

    These are all facts about simplicity: resultatives are simple. Generalization (a) seems to me to be

    the most obviously robust, and in some sense it is perhaps the most surprising; resultatives often

    seem to involve a change in argument structure, but no verb marks it. However, just accepting

    (a{c), we know that if we can nd a resultative in a given language, we can learn about the basic

    conditions under which that language can function. In particular, we can infer something about

    -key : it is used to indicate the simplest form of the (non-nite) verb. (I think this could also be said

    to be the conclusion of Jang (1997).) Due to the fact that Korean verb roots are morphologically

    bound, they cannot appear `bare'; -key is the least marked inectional ending, and because of this

    fact, the language uses it in resultatives.

    Even though -key is `unmarked' with respect to other non-nite endings, it still may appear to be

    selected by a higher predicate. For instance, we can contrast the output of the lexical rule in (60)

    with the entry of a verb like kwul-ta (`treat'), which presumably takes an object and a predicative

    property-denoting complement.

    18

    17

    The form ha-ta does appear as a true governing predicate in the complex predicate V-eya ha-ta, meaning `must

    V'.

    18

    The object can be marked with either accusative or dative case, depending in part on the nature of the property-

    denoting complement (Hanjung Lee, p.c.).

    25

  • (64) Lexical entry for kwul-ta

    2

    6

    6

    6

    6

    6

    6

    6

    6

    6

    6

    6

    6

    6

    4

    SUBJ

    D

    NP

    i

    E

    COMPS

    *

    V

    2

    4

    SUBJ

    D

    PRO

    j

    E

    CONT

    1

    3

    5

    , NP

    j

    +

    CONT

    2

    6

    6

    4

    RELN treat

    ACTOR i

    PATIENT j

    PROPERTY

    1

    3

    7

    7

    5

    3

    7

    7

    7

    7

    7

    7

    7

    7

    7

    7

    7

    7

    7

    5

    Here, the verb is specied to take a verbal complement. As a COMP expressing VFORM only

    appears when specically selected, no COMP will be sanctioned: the simplest (non-nite) verbal

    form is marked with -key , and so -key is what appears.

    5. Free Complements and Verbal Noun Complements

    The idea that -key diers from other COMPs in not expressing VFORM can be bolstered by looking

    at other data. In this section I want to consider briey the similarities of -key-complements and

    verbal noun (VN) complements. Crucially, both types canonically appear with ha-ta, but they

    also both appear with sikhi-ta (`cause'), and the passive is formed directly with toy-ta (`become'),

    rather than the passive of ha-ta.

    5.1. Verbal Nouns

    VNs with ha-ta have received a great deal of attention in the linguistic literature, and I will not

    attempt to address the nature of that specic combination here, though it could be thought of as

    a yet more generalized version of free composition. As just noted, VNs may also combine with the

    verb sikhi-ta, which always gives a causative reading, as seen in the examples in (65){(66).

    (65) a. hwasal-i kwanyek-ey myengcwung hay-ss-ta

    arrow-NOM target-at hit.the.mark do-PAST-DECL

    `The arrow hit the target.' (from Jung (1997))

    b. John-i hwasal-ul myengcwung sikhi-ess-ta

    John-NOM arrow-ACC hit.the.mark cause-PAST-DECL

    `John caused the arrow to hit (the target).'

    (66) a. sensayngnim-i yenge-lul kongpwu hay-ss-ta

    teacher-NOM English-ACC study do-PAST-DECL

    `The teacher studied English.'

    b. sensayngnim-i yenge-lul kongpwu sikhi-ess-ta

    teacher-NOM English-ACC study cause-PAST-DECL

    `The teacher made someone study English.'

    26

  • The verb sikhi-ta can combine either with V-key or with a VN, and gives a canonical causative

    interpretation. This is encoded in its lexical entry in (67).

    (67) Lexical entry for sikhi-ta

    2

    6

    6

    6

    6

    6

    6

    6

    6

    6

    6

    6

    6

    6

    6

    6

    4

    SUBJ

    D

    NP

    i

    E

    COMPS

    *

    V(N)

    2

    6

    6

    4

    SUBJ

    D

    PRO

    j

    E

    COMPS

    1

    CONT

    2

    3

    7

    7

    5

    , NP

    j

    +

    1

    CONT

    2

    6

    6

    4

    RELN cause

    ACTOR i

    PATIENT j

    EFFECT

    2

    3

    7

    7

    5

    3

    7

    7

    7

    7

    7

    7

    7

    7

    7

    7

    7

    7

    7

    7

    7

    5

    This verb can take either a simple verbal complement, or a verbal noun complement, and, really,

    there is no dierence between the two. As one would not want to say that VNs carry a VFORM value,

    so too it now seems odd to think of V-key carrying such a value. Thus, sikhi-ta really just wants

    some kind of predicative complement as its rst complement. It expresses a causative relation

    between its two thematic NP arguments, and its object controls the subject of the predicative

    complement.

    Moving now to toy-ta, the passive of piphan ha-ta (`criticize') is piphan toy-ta (`be criticized'), just

    like the passive of the causative haykyel ha-key ha-ta (`cause to solve') is haykyel ha-key toy-ta (`be

    caused to solve'). It has been often noted that the toy-ta form tends to lose the passive of the

    causative reading, and typically has more of an `it came to be the case that . . . ' interpretation.

    This is consistent with the idea that all of these freely composed predicates just compose some

    main verb with a preceding complement: in the case of toy-ta, the main verb would mean just

    what toy-ta means, namely a change of state or circumstance.

    5.2. The marker -key vs. the COMPs

    A bare VN is an X

    0

    element, but a VN can also appear as the head of a phrase, when case marked.

    In Sells (1996) I tried to draw some generalizations that emerged from treating case as correlating

    with projection to N

    0

    , and tense as correlating with projection to V

    0

    . Above, I have discussed the

    fact that V-key can appear as the head of a V

    0

    , a VP, or an S constituent, so V-key can project to

    V

    0

    (a fully phrasal projection of V). However, verbs hosting COMPs never project above the [BAR

    0] level: they must appear in verbal complexes. In other words, a verb with a VFORM feature is

    limited to a [BAR 0] specication.

    So a verb marked for tense or marked with -key can project to V

    0

    ; a verb bearing a COMP does

    not project. From this perspective, it makes sense to think that verbs normally can project, but

    that COMPs have the specic function of limiting that projection, and hence a COMP-bearing verb

    must appear in a verbal complex. In contrast, -key is the least marked (non-nite) ending, and

    consistent with that, does not limit (or force) the projection of a verb to V

    0

    .

    27

  • 6. Ordering Principles

    Finally, let me briey consider one crucial aspect of my syntactic proposal, missing so far. A verb

    may combine with several X

    0

    elements, as in the slightly forced example (68).

    (68) haksayng-tul-i swukcey-lul cey ttay

    student-PLU-NOM homework-ACC proper time

    ceychwul cal an hay cwu-ess-ta

    submit well NEG do give-PAST-DECL

    `The students didn't hand in their homeworks in time (well).'

    Here the sequence ceychwul cal an hay cwu-ess-ta is a sequence of 5 X

    0

    elements, headed by cwu-

    ess-ta. Although Korean is a scrambling language, X

    0

    elements such as these come in a xed order.

    This xed order distinguishes the [BAR 0] structures in Korean from the [BAR 1] phrasal syntax,

    where scrambling is generally free. I discussed such facts as these in Sells (1994), and proposed

    that elements such as cal and an were restricted adverbial modiers. In the present paper, I have

    shown that there is motivation to think of an as a complement (of cwu-ta in (68)), and it certainly

    is reasonable to think of the VN ceychwul as a complement of hay . If we accept this, then it

    looks very odd to treat the restricted adverb cal as an adjunct, in the middle of a string of clear

    complements.

    Given the view I have been developing here, it makes sense to think of all non-head X

    0

    elements

    as being introduced as complements, and that this X

    0

    complement domain is precisely the `sub-

    phrasal' domain described in Sells (1994).

    19

    We can specify a lexical rule for the few restricted

    adverbs such as cal or cokum in Korean; like the negation lexical rule, it adds a restricted adverb

    as the complement to any verb.

    (69) Korean Restricted Adverb Lexical Rule

    2

    4

    HEAD verb

    COMPS

    1

    CONT

    2

    3

    5

    )

    2

    6

    4

    COMPS

    D

    Adv

    0

    :

    3

    E

    1

    CONT

    3

    h

    ARG

    2

    i

    3

    7

    5

    Now, let us look at the [BAR 0] structure of (68), shown in (70). Strictly speaking, the VN is a

    complement of hay , and cal and an are complements of cwu-ta. However, as cwu-ta selects for

    hay , it will inherit all of the complements, and so everything except cwu-ta in (70) will be licensed

    syntactically as a complement to cwu-ta.

    (70) V

    0

    VN

    0

    Adv

    0

    Neg

    0

    V

    0

    V

    0

    ceychwul cal an hay cwu-ess-ta

    The generalizations for all [BAR 0] structures are shown in (71).

    19

    The restricted nature of elements like cal and an is also discussed in Lee (1993).

    28

  • (71) a. The head is nal.

    b. All verbs follow all other elements.

    c. Negation immediately precedes a verb.

    d. The adverb cal precedes negation.

    e. The (bare) verbal noun precedes everything else.

    Generalization (e) is quite striking, as the bare VN indeed seems to enter into some kind of argument

    structure sharing construction with the verb ha-ta, even though the two may not be adjacent. This

    is one place where VNs and Vs marked with -key dier: Vs marked with -key must follow restricted

    adverbs and negation under a dominating V

    0

    , while VNs are initial.

    Speculatively, it would seem that an interesting way to approach these perhaps quite parochial facts

    would be to use the idea from Optimality Theory (OT) of ranked and conicting constraints. Work

    in OT phonology has shown that quite unusual and expected constellations of facts can emerge from

    universal constraints under a language-specic ranking. The informal linear precedence statements

    in (72), if taken as ranked in the order given, predict exactly the generalizations in (71).

    (72) Ranked Linear Precedence Constraints:

    a. Head is nal.

    b. V is nal.

    c. Neg

    0

    immediately precedes V.

    d. Adv

    0

    immediately precedes V.

    Here (a) is uncontroversial, and (c) is surface true: short form negation always immediately precedes

    a verb. (d) is almost surface true: a restricted adverb like cal always appears immediately before a

    verb, unless short form negation intervenes. Hence, these constraints have some plausibility. If we

    now interpret them as ranked as shown, by (b) verbs will follow anything else, but the head will

    be absolutely nal, by (a), the strongest requirement. Short form negation will be closer to a verb

    than a restricted adverb, as (c) is a stronger requirement than (d). And if there is a verbal noun to

    be ordered, it will be forced to precede any element mentioned in (a{d), as (a{d) eectively force

    the elements they mention to be as far to the right as possible.

    Conclusion

    In conclusion, I have based my observations in this paper around the special status of the marker -

    key , which appears only in freely composed complex predicates. Such complex predicates dier from

    lexically governed ones, which in some sense represent the majority of complex predicates in Korean.

    These governed complex predicates are expressed in verbal complexes, which also allow short form

    negation, restricted adverbs, and bare verbal nouns within them. Within Korean, the approach

    that I have taken here might perhaps provide a foundation for considering what properties of

    structure correlate with the various COMPs; although the various COMPs seem specic to particular

    29

  • governed complex predicates, it hardly seems plausible that their distribution is actually random. If

    their distribution is not random, then there must be principles which determine that distribution.

    Investigating this further should give us clues to the relation between the specic expression of

    each governed complex predicate and its specic semantic properties. On a wider level, the account

    that I have suggested here draws attention to the need for a better understanding of the relation

    between causatives and resultatives.

    In terms of the theory of syntactic structure, the interactions of the governed complex predicates

    with short form negation show clearly that negation must be introduced subordinate to the verb

    it negates: it must be a complement. Additionally, the xed ordering we nd among X

    0

    elements

    seems to follow most naturally from a at n-ary structure.

    References

    Abeille, Anne, and Daniele Godard. 1998. A lexical approach to quantier oating in French. To

    appear in A. Kathol, J.-P. Koenig, and G. Webelhuth (eds.), Lexical and Constructional Aspects

    of Linguistic Explanation, Stanford, CSLI Publications.

    Bratt, Elizabeth Owen. 1996. Argument Composition and the Lexicon: Lexical and Periphrastic

    Causatives in Korean. Doctoral dissertation, Stanford University.

    Butt, Miriam. 1997. Complex predicates in Urdu. In Alex Alsina, Joan Bresnan, and Peter Sells

    (eds.), Complex Predicates, 107{149. Stanford, CSLI Publications.

    Cho, Jae Ohk. 1988. Suxed verb forms and compound verb constructions. In Eung-Jin Baek (ed.),

    Papers from the Sixth International Conference on Korean Linguistics, 77{106. International

    Circle of Korean Linguistics, Seoul, Hanshin.

    Cho, Young-Mee Yu, and Ki-Sun Hong. 1988. Evidence for the VP constituent from child Korean.

    Papers and Reports on Child Language Development 27, 31{38.

    Cho, Young-mee Yu, and Peter Sells. 1995. A lexical account of inectional suxes in Korean.

    Journal of East Asian Linguistics 4, 119{174.

    Chomsky, Noam. 1995. The Minimalist Program. Cambridge, MIT Press.

    Chung, Chan. 1993. Korean auxiliary verb constructions without VP nodes. In S. Kuno et al. (ed.),

    Harvard Studies in Korean Linguistics, Vol. 5, 274{286.

    Dowty, David. 1979. Word Meaning and Montague Grammar. Dordrecht, Reidel.

    Foley, William, and Robert van Valin. 1984. Functional Syntax and Universal Grammar. New

    York, Cambridge University Press.

    Fukui, Naoki. 1986. A Theory of Category Projection and Its Applications. Doctoral dissertation,

    MIT.

    Hagstrom, Paul. 1996. do-support in Korean: Evidence for an interpretive morphology. In Hee-Don

    Ahn, Myung-Yoon Kang, Yong-Suck Kim, and Sookhee Lee (eds.),Morphosyntax in Generative

    Grammar (Proceedings of 1996 Seoul International Conference on Generative Grammar), 169{

    180. Seoul, The Korean Generative Grammar Circle, Hankwuk Publishing Co.

    30

  • Jang, Youngjun. 1997. On the so-called adjunct predicates in Korean. In Ralph Blight and Michelle

    Moosally (eds.), The Syntax and Semantics of Predication (Texas Linguistic Forum 38), 149{

    159. Austin, Texas, Department of Linguistics.

    Jung, Yunsun. 1997. Argument selection of Sino-Korean verbal nouns. In Ho-Min Sohn and

    John Haig (eds.), Japanese/Korean Linguistics, Vol. 6, 437{453. CSLI, Stanford Linguistics

    Assocation.

    Kim, Jong-Bok. 1995. The Grammar of Negation: A Lexicalist, Constraint-Based Perspective.

    Doctoral dissertation, Stanford University.

    Kim, Jong-Bok, and Ivan Sag. 1996. The parametric variation of English and French negation. In

    J. Camacho et al. (ed.), Proceedings of WCCFL, Vol. 14, 303{317. CSLI, Stanford Linguistics

    Association.

    Kim, Soowon, and Joan Maling. 1997. A cross-linguistic perspective on resultative formation. In

    Ralph Blight and Michelle Moosally (eds.), The Syntax and Semantics of Predication (Texas

    Linguistic Forum 38), 189{204. Austin, Texas, Department of Linguistics.

    Kim, Young-joo. 1990. The Syntax and Semantics of Korean Case: The Interaction between Lexical

    and Syntactic Levels of Representation. Doctoral dissertation, Harvard University.

    Kim, Young-joo. 1997. Acquisition of Korean. In Dan Slobin (ed.), The Crosslinguistic Study Of

    Language Acquisition. Hillsdale, New Jersey, Lawrence Erlbaum Associates.

    Lee, Hanjung. 1998a. On the semantic and conceptual basis of locative alternation. Ms. Stanford

    University.

    Lee, Hanjung. 1998b. Korean causatives in a constraint-based lexicon. ICKL presentation, 1998.

    Lee, Jae Hong. 1993. Postverbal adverbs and verb movement in Korean. In P. Clancy (ed.),

    Japanese/Korean Linguistics, Vol. 2, 429{446. CSLI, Stanford Linguistics Association.

    Lee, Keedong. 1993. A Korean Grammar on Semantic-Pragmatic Principles. Seoul, Hankwuk

    Mwunhwasa.

    Manning, Christopher, and Ivan Sag. 1998. Dissociations between argument structure and gram-

    matical relations. To appear in A. Kathol, J.-P. Koenig, and G. Webelhuth (eds.), Lexical and

    Constructional Aspects of Linguistic Explanation, Stanford, CSLI Publications.

    Manning, Christopher, Ivan Sag, and Masayo Iida. 1998. The lexical integrity of Japanese

    causatives. To appear in Robert Levine and Georgia Green (eds.), Readings in Modern Phrase

    Structure Grammar , Cambridge, Cambridge University Press.

    Martin, Samuel. 1992. A Reference Grammar of Korean. Rutland, Vermont, Charles E. Tuttle.

    Pollard, Carl, and Ivan Sag. 1994. Head-Driven Phrase Structure Grammar. Chicago, University

    of Chicago Press and Stanford, CSLI Publications.

    Poser, William. 1992. Blocking of phrasal constructions by lexical items. In Ivan Sag and Anna

    Szabolcsi (eds.), Lexical Matters, 111{130. Stanford, CSLI Publications.

    Sadler, Louisa, and Doug Arnold. 1994. Prenominal adjectives and the phrasal/lexical distinction.

    Journal of Linguistics 30, 187{226.

    Sells, Peter. 1991. Complex verbs and argument structures in Korean. In S. Kuno et al. (ed.),

    Harvard Studies in Korean Linguistics, Vol. 4, 395{406.

    31

  • Sells, Peter. 1994. Sub-phrasal syntax in Korean. Language Research 30, 351{386.

    Sells, Peter. 1995. Korean and Japanese morphology from a lexical perspective. Linguistic Inquiry

    26, 277{325.

    Sells, Peter. 1996. Case, categories, and projection in Korean and Japanese. In Hee-Don Ahn,

    Myung-Yoon Kang, Yong-Suck Kim, and Sookhee Lee (eds.), Morphosyntax in Generative

    Grammar (Proceedings of 1996 Seoul International Conference on Generative Grammar), 47{62.

    Seoul, The Korean Generative Grammar Circle, Hankwuk Publishing Co.

    Sells, Peter. 1998. Optimality and economy in Japanese and Korean morphosyntax. In Noriko Akat-

    suka et al. (ed.), Japanese/Korean Linguistics, Vol. 7, 499{514. CSLI, Stanford Linguistics

    Assocation.

    Sohn, Ho-Min. 1994. Korean. London and New York, Routledge.

    Song, Seok Cheong. 1988. Explorations in Korean Syntax and Semantics. Berkeley, Institute of

    East Asian Studies.

    32