Using A Semantic Classification in Parsing Chinese : Some Preliminary Results

Language, Information and Computation (PACLIC12), 18-20 Feb, 1998, 340-347 Using A Semantic Classification in Parsing Chinese: Some Preliminary Results Kok Wee GAN Department of Computer Science Hong Kong University of Science & Technology Clear Water Bay Road, Kowloon, Hong Kong gankw@cs.ust.hk Abstract This paper describes a semantic classification of Chinese and compares its performance with CKIP's word-class classification on the task of identifying the attributive dependency relation and the unmarked coordinative dependency relation of word bigrams made up from nouns, verbs, and adjectives. 1� Introduction In parsing Chinese, the practice of deriving the semantic structure of a sentence from word-class information faces two main challenges: (i) it is difficult to determine word classes due to the lack of inflections in Chinese; (ii) there is a lot of ambiguities in the mapping of word classes to the grammatical roles and to the semantic structure. Table 1 summarizes the types of dependency relation that could exist in word bigrams composed from nouns, verbs, and adjectives. • ActantlRel The actantl relation, denoted as ActantlRel, connects between a lexeme whose meaning is a functor (the head) and its first argument (the dependent). Examples include:1 ActantlRel(gg 'agree', ft 'you')� ActantlRel(t 'big', AR 'tolerance') • ActantlRel(&)4J 'useless', g 'say') Actant2Rel ActantlRel(ki 'is', lb* `deligene) The actant2 relation, denoted as Actant2Rel, connects between a lexeme whose meaning is a functor (the head) and its second argument (the dependent). Examples include: Actant2Rel( 'drink', -,1÷; 'tea')� Actant2Rel(SY1 'feel', aVj 'cordial') Actant2Rel(ge 'like', Mil< 'swim') • MotionStateRel The motion-state relation, denoted as MotionStateRel, exists between two predicates, the dependent (a predicate) expresses the result of the motion conveyed by the other predicate (the head). MotionStateRel(* 'take', A 'walk')� MotionStateRel(IR 'make', t 'big') MotionStateRel(4 'sit', *E 'upright') • AttrRel The attributive relation, denoted as AttrRel, covers all kinds of modifiers: AttrRel(fFT 'cup', 3g34 'glass')� AttrRel(44 'dictionary', Ift 'pocket-size') AttrRel(,' a 'construct', I � 'industry')� AttrRel(EN 'regular',� 'not') AttrRel('M 'method', Via 'train')� AttrRel(gM 'extend', 4V 'stay') • CoordRel The coordinative relation discussed here, denoted as CoordRel, focuses only on unmarked coordinate constructions without any conjunction. The left-hand conjunct is assumed as the head. Examples are: CoordRel(*A 'wood table, 710* 'wood chair') CoordRe1CFW 'rain', ttUR 'wind blow') CoordRel(1)] 'lively', r&rit 'vigorously') In this paper, we propose a semantic classification of Chinese. We compare its performance against the 1 In this paper, a dependency relation between two words is represented as: relation(head, dependent). 340 Language. Information and Computation (PACLIC12), 18-20 Feb, 1998, 340-347 traditional approach on the task of identifying the attributive and the unmarked coordinative relations in word bigrams made up from nouns, verbs, and adjectives. 2� The Semantic Classification of Chinese We propose a semantic classification of Chinese. This classification is derived by adopting the taxonomy of linguistic meanings proposed in the Meaning-Text approach (Polguere, to appear) and developing it further based on (Dong, 1994; 14:13M-Mra, 1993). Figures 1, 2, 3, 4, and 5 shows the details. The object name and predicate classifications are adapted from EMICS (Dong, 1994). The difference is …

Content truncated to 500 words