The Sino-Korean light verb construction and lexical argument structure

In Korean, a class of lexemes of Chinese origin exhibit both nominal and verbal behavior. Specifically, they can assign lexically idiosyncratic case, but require a semantically vacuous light verb in order to form a sentence and are themselves marked with accusative case. In this paper, we propose a TAG-based account of this behavior, and propose some generalizations towards a pure representation of lexical argument structure. 1. Linguistic Facts and Issues In this paper, we provide a syntactic analysis of Sino-Korean light verb constructions (LVC henceforth) that are composed of the light verb ha and an activity-denoting noun of Chinese origin.1 We will refer to this activity-denoting noun as the ‘base’ of the LVC. The argument structure of LVCs come from the base, and the light verb is semantically vacuous and does not assign any theta roles. This is shown by the fact that although the examples in (1) all contain ha, they have different argument structures. (1) a. John-i John-Nom swuhak-ul math-Acc yenkwu-lul research-Acc ha-yess-ta. HA-Past-Decl ‘John researched math.’ b. Kicha-ka train-Nom Seoulyek-ey Seoul-station-at tochak-ul arrival-Acc ha-yess-ta. HA-Past-Decl ‘The train arrived at Seoul station.’ c. Kicha-ka train-Nom Seoulyek-eyse Seoul-station-from chwulpal-ul departure-Acc ha-yess-ta. HA-Past-Decl ‘The departed from Seoul station.’ For instance, the arguments in (1a) are agent and goal, those in (1b) are patient and goal, and those in (1c) are patient and source. If, however, the theta roles in LVCs are assigned by the base, it is puzzling why the argument NPs are syntactically realized outside of the base NP. The case postpositions such as Acc, -ey and -eyse on the argument NPs indicate that they are daughters of VP, and not the base NP. An 1Han has been partially funded by the Army Research Lab via a subcontract from CoGenTex, Inc., and by NSF Grant SBR 8920230. We would like to thank Aravind Joshi, Tony Kroch, Martha Palmer, and Anoop Sarkar for useful discussions. Chung-hye Han and Owen Rambow NP that is a daughter of another NP requires genitive or null case postposition in Korean. We will refer to the first kind of case as VERBAL CASE, and the second as NOMINAL CASE. Moreover, as noted by (Grimshaw & Mester, 1988), there are restrictions on argument realization, which can be clearly shown with ditransitive LVCs, as in (2). (2) a. John-i John-Nom Mary-eykey Mary-to inhyung-ul doll-Acc senmwul-ul gift-Acc ha-yess-ta. HA-Past-Decl ‘John gave a gift of a doll to Mary.’ b. John-i John-Nom Mary-eykey Mary-to inhyung(-uy) doll(-Gen) senmwul-ul gift-Acc ha-yess-ta. HA-Past-Decl c. * John-i John-Nom inhyung-ul doll-Acc Mary-eykey-uy Mary-to-Gen senmwul-ul gift-Acc ha-yess-ta. HA-Past-Decl d. * John-i John-Nom Mary-eykey-uy Mary-to-Gen inhyung(-Gen) doll(-Gen) senmwul-ul gift-Acc ha-yess-ta. HA-Past-Decl The base senmwul (‘gift’) assigns agent, goal and theme. In (2a), all the argument NPs are realized outside of the base NP. In (2b), the agent and goal arguments are realized outside of the base NP, but the theme argument is realized inside the base. However, it is not possible to realize theme argument outside of the base when the goal argument is realized inside the base, as shown in (2c), and it is not possible to realize both theme and goal arguments inside the base, as shown in (2d). (Grimshaw & Mester, 1988) (G&M henceforth) summarize the restrictions on argument realization as follows: (i) the subject argument must always be outside the base NP; (ii) at least one argument apart from the subject must be outside the base NP; and (iii) for nouns that take a theme and a goal, if the theme argument is realized outside the base NP, the goal must also be realized outside the base NP. In what follows, we first briefly discuss some previous analyses and their shortcomings, and present our own analysis using the framework of Feature Based Lexicalized Tree Adjoining Grammar. We discuss English data in comparison, and conclude with a discussion of noun phrases. 2. Previous Analyses According to G&M, a light verb such as ha has no argument structure on its own and it occurs with a noun which is ‘theta-transparent.’ Theta-transparent nouns can transfer some or all of their arguments to the argument structure of the light verb. This mechanism allows the light verb to directly assign theta roles to the argument NPs in syntax and such argument NPs are realized outside the base NP. They further assume (following much previous work) that arguments have a hierarchy according to prominence. For instance, the agent is more prominent than the goal, which is more prominent than the theme. Based on this assumption, they propose that when a theta role is transfered (e.g., the theme), any theta roles that are higher in prominence must transfer as well (i.e, the agent and goal). This explains the ungrammaticality of (2c). G&M also stipulate that the base noun must transfer at least one internal argument in order to be licensed. Otherwise, the theta-criterion is violated, since the base noun does not receive a theta role from anywhere. This is why (2d) is ungrammatical under G&M’s system. G&M wrongly predict that intransitive LVCs do not exist, since there is no internal argument to participate in the transfer. But intransitive LVCs clearly do exist, as shown in (3) and (4). Note that while (4) may be ambiguous between a heavy and light verb reading of ha, (3) is not, since the subject is not an agent. Light Verbs and Lexical Argument Structure (3) John-i John-Nom samang-ul death-Acc ha-yess-ta. HA-Past-Decl ‘John died.’ (4) John-i John-Nom swuyuong-ul swimming-Acc ha-yess-ta. HA-Past-Decl ‘John was swimming.’ For this reason, (Yoon, 1991) rejects G&M’s argument transfer theory and proposes ‘argument sharing’ mechanism. He argues that the light verb is thematically underspecified and so unsaturated. This forces the base noun which has theta structure and the light verb to undergo the operation of Theta Identification, allowing the argument structure of the base noun and that of the light verb to be shared. This sharing is viewed as the unification of the argument structure of the base noun into the underspecified argument structure of the light verb. Yoon’s theory predicts that when there are more than one internal arguments, they must all be realized outside of the base NP. But this is an incorrect prediction: in ditransitive LVCs, while the goal argument is realized outside the NP, the theme argument can be realized inside NP, as shown in (2b). The same problem persists in (Park, 1992). He argues that the categorial status of the base is not a noun, but a verb. Thus, it assigns theta-roles just as any other verbs. The light verb is simply an auxiliary verb that supports inflection. But if the base is simply a verb, then (2b) is wrongly predicted to be ungrammatical. 3. TAG Analysis The key to our analysis is the assumption that the base is underspecified with respect to word class (verb or noun). We propose that this base is the anchor of an elementary tree with all its arguments and that it acquires a noun status only after the light verb adjoins into the elementary tree. The assumption that the category of the base is unspecified is well-motivated for two reasons: (i) The base form originates from Chinese, in which the same form is used both as a noun and a verb, (ii) there is no consensus in the literature as to what the category of the base is and positing that it is either a noun or a verb leads to difficulties, as discussed in x2. We represent this by using the label X for its category (which projects to XP). We also assume that each node in a tree is associated with a category feature CAT with values such as V(ERB) and N(OUN). The CAT feature of nodes labeled V, VP, or S is necessarily V for both the top and bottom feature structures, while nodes labeled N or NP necessarily have [CAT:N].2 But the CAT feature of the base of LVC is unspecified. In addition, we assume that nodes in a projection have a full set of morpho-syntactic features. In this paper we use only the binary feature [TENSED: ]. We assume that the base is [TENSED:-] (since it carries no tense morphology), that the S node is marked [TENSED:+], and that the TENSED feature is shared among the nodes of a projection. We assume that when a lexeme (of any category) forms a syntactic predication structure it projects to a maximal verbal projection (VP) and we refer to this VP as the PREDICATE. Furthermore, following (Heycock & Lee, 1989), we assume that in Korean, nominative case is assigned by the predicate, not by Infl. (Heycock & Lee, 1989) use as evidence the presence of multiple nominative constructions and the fact that infinitivals can have nominative case-marked subject. As a result, all clausal structures need a VP node as a sister to the subject argument to license nominative case.3 We also assume that the lexeme projects all of its argument positions in canonical order according to theta hierarchy. That is, the most prominent argument attaches 2The node labels are not actually used in our analysis, and we could also label all nodes XP. We retain the traditional labels for clarity. 3This is compatible with the XTAG analysis of the predicative use of nouns and adjectives in English, the trees for which project from N (or A) to S via NP (AP) and VP (though perhaps the VP is less motivated in English than in Korean because the adjoined auxiliary provides the nominative case in English, not the predication structure itself). Chung-hye Han and Owen Rambow to the highest projection, and the least prominent attaches to the lowest projection.4 We assume that each lexeme idiosyncratically fixes a case grid for its arguments,5 which is only realized in appropriate syntactic contexts. (Thus, rather than speak of unified case assignment, we will henceforth speak of case assignment by the lexical head and subsequent case realization in a particular syntactic context.)