A Computational Treatment of Korean Serial Verb Constructions

A Computational Treatment of Korean Serial Verb Constructions⋆ Jong-Bok Kim a, Jaehyung Yang b, and Sanghoun Song c aSchool of English, Kyung Hee University jongbok@khu.ac.kr bDept. of Computer Science, Kangnam Univeristy jaehyung@kangnam.ac.kr cDept. of Linguistics, Univ. of Washington sanghoun@gmail.com Abstract. The so-called serial verb construction (SVC) is a complex predicate structure consisting of two or more verbal heads but denotes one single event. This paper first dis- cusses the grammatical properties of Korean SVCs and provides a lexicalist, construction- based analysis couched upon a typed-feature structure grammar. We also show the results of implementing the grammar in the LKB (Linguistics Knowledge Building) system couched upon the existing the KRG (Korean Resource Grammar) which has been developed since 2003. The implementation results provides us with a feasible direction of expanding the analysis to cover a wider range of relevant data. Keywords: serial verb construction, Korean Resource Grammar, LKB (Linguistic Knowledge Building) 1 Introduction Korean is one of the languages that employ the so-called serial verb construction (SVC) in which more than one verb occurs without any specific coordination or subordination markings: (1) a. Mia-ka hakkyo-ey kel-e ka-ss-ta Mia-NOM school-to walk-COMP go-PAST-DECL ‘Mia walked to school.’ b. Mia-ka cwul-ul cap-a tangki-ess-ta Mia-NOM rope-ACC draw-COMP pull-PAST-DECL ‘Mia pulled a rope, drawing it.’ Both sentences here, though including two serial verbs with their own predicate relations, seman- tically represent only a single event. These sentences display the canonical grammatical properties of SVCs in that the two successive verbs behave like a complex predicate, sharing one tense and subject value, and even the object value as in (1b). The generation of Korean SVCs is quite flexible as attested by the corpus examples of the verb mek-ta ‘eat’: (2) nanwu-e mek-ta ‘divide and eat’, kkulhi-e mek-ta ‘boil and eat’, mandul-e mek-ta ‘make and eat’, cap-a mek-ta ‘catch and eat’, cip-e mek-ta ‘pick up and eat’, ssip-e mek-ta ‘chew and eat’, kwu-e mek-ta ‘broil and eat’, ppal-a mek-ta ‘suck and eat’, etc. The examples, extracted from the Sejong POS-tagged Corpus, show us that the activity verb mek- ta ‘eat’ can combine with various types of verbs, forming an SVC.1 ⋆ Many thanks go to anonymous reviewers for helpful comments. Main theoretical aspects sketched here follow those in Kim (2010). 1 The Sejong Corpus, released by the National Institute of the Korean Language, is a balanced corpus consisting of about 12 million words with 311,048 sentences. PACLIC 24 Proceedings 439 In this paper, we try to explore the syntactic and semantic properties of the Korean SVCs and build a computationally feasible analysis that can parse the desired set of examples. We then test the analysis with implementing it in the LKB (Linguistic Knowledge Building) System within the preexisting computational grammar, KRG (Korean Resource Grammar). 2 Grammatical Properties of the SVCs 2.1 Syntactic Headedness and Types of the SVC As noted, one main property of the Korean SVCs is that each SVC sentence has only one tense, aspect, and mood value realized on the final verb. The appearance of …

Content truncated to 500 words