An empirical study for generating zero pronoun in Korean based on Cost-based centering model

An Empirical Study for Generating Zero Pronoun in Korean based on Cost-based Centering Model Ji-Eun Roh Div. of Electrical and Computer Engineering Pohang University of Science and Tech- nology and Advanced Information Tech- nology Research Center (AITrc) San 31, Hyoja-dong, Nam-gu, Pohang, 790-784, R. of KOREA jeroh@postech.ac.kr fax: +82-54-279-5699 Jong-Hyeok Lee Div. of Electrical and Computer Engineering Pohang University of Science and Tech- nology and Advanced Information Technology Research Center (AITrc) San 31, Hyoja-dong, Nam-gu, Pohang, 790-784, R. of KOREA jhlee@postech.ac.kr fax: +82-54-279-5699 Abstract In Korean, in order to generate a coherent text, a redundantly prominent noun should be replaced by a non-zero pronoun or zero pronoun. Otherwise, the text becomes un- natural. Specifically, a redundant noun in Korean is frequently omitted while a re- dundant noun in English is replaced by a pronoun. This paper proposes a genera- tion algorithm of the zero pronoun, using a Cost-based Centering Model which considers the inference cost. For an objec- tive evaluation of our algorithm, we col- lected 87 texts from three genres, and manually recovered the omitted elements. Using the collected texts, we verify that our algorithm is well defined to explain the phenomenon of the zero pronoun in Korean. We also show that the proposed approach resolves both the over- generation of the zero pronoun in Con- tinue and its under-generation in other transitions in terms of Centering. 1 Introduction Text generation is the process of producing com- prehensible texts in natural languages from non- linguistic representation of information. To gener- ate a coherent text, we must pay attention to each stage of generation, such as content determination, text structuring, aggregation, and generation of anaphoric expression (pronominalization). Among these stages, we are especially interested in the generation of anaphoric expression focusing on zero pronouns, because generating the appropriate zero pronoun is directly connected with text coher- ence. Consider the following short text. (1) Na-nun (I, topic) cinyel-toyn (on display) os (the dresses) cwung (one of) maum-ye tu-nun kes-i iss- ess-ta (attracted me). (One of the dresses on display attracted me.) (2) [Na-nun (I, topic), ø]1 [os-oul (dress, object), ø] ip-e po-ass-ta (putted on). (I (ø) putted it (ø) on.) (3) Haciman (however), [na-nun (I, topic), ø] [os-i (dress, subject), ø] nemu (too) khesu (big) [os-ul (it, object), ø] sal-swu (buy) eps-ess-ta.(can not). (However, the dress (ø) was too big so that I (ø) cannot buy it (ø)) In the above text, ‘na (I)’ appears repeatedly as a topic2 in sentence (1), (2), and (3), and ‘os (dress)’ 1 A bracketed noun, which means the unexpressed argument of the verb, is a zero pronoun. Generally, this kind of omitted element caused by the zero anaphora phenomenon is called zero pronoun, zero element, zero anaphor, or null element. In this paper, we call the omitted element a zero pronoun. appears repeatedly as an object and a subject in sentence (2) and (3). Korean is a highly context- dependent language, and any arguments recover- able from the context are freely dropped. In the …

Content truncated to 500 words