Constructing Taxonomy of Numerative Classifiers for Asian Languages

Constructing Taxonomy of Numerative Classifiers for Asian Languages Kiyoaki Shirai JAIST kshirai@jaist.ac.jp Takenobu Tokunaga Tokyo Inst. of Tech. take@cl.cs.titech.ac.jp Chu-Ren Huang Academia Sinica churenhuang@gmail.com Shu-Kai Hsieh National Taiwan Normal Univ. shukai@gmail.com Tzu-Yi Kuo Academia Sinica ivykuo@gate.sinica.edu.tw Virach Sornlertlamvanich TCL, NICT virach@tcllab.org Thatsanee Charoenporn TCL, NICT thatsanee@tcllab.org Abstract Numerative classifiers are ubiquitous in many Asian languages. This paper pro- poses a method to construct a taxonomy of numerative classifiers based on a noun- classifier agreement database. The taxon- omy defines superordinate-subordinate rela- tion among numerative classifiers and rep- resents the relations in tree structures. The experiments to construct taxonomies were conducted for evaluation by using data from three different languages: Chinese, Japanese and Thai. We found that our method was promising for Chinese and Japanese, but in- appropriate for Thai. It confirms that there really is no hierarchy among Thai classifiers. 1 Introduction Many Asian languages do not mark grammatical numbers (singular/plural) in noun form, but use nu- merative classifiers together with numerals instead when describing the number of nouns. Numerative classifiers (hereafter “classifiers”) are used with a limited group of nouns, in particular material nouns. In English, for example: “three pieces of paper”. In Asian languages these classifiers are ubiquitous and used with common nouns. Therefore the number of classifiers is much larger than in Western languages. An agreement between nouns and classifiers is also necessary, i.e., a certain noun specifies possible clas- sifiers. The agreement is determined based on var- ious aspects of a noun, such as its meaning, shape, pragmatic aspect and so on. This paper proposes a method to automati- cally construct a taxonomy of numerative classi- fiers for Asian languages. The taxonomy defines superordinate-subordinate relations between classi- fiers. For instance, the Japanese classifier “頭 (tˆo)” is used for counting big animals such as elephants and tigers, while “匹 (hiki)” is used for all animals. Since “匹” can be considered more general than “ 頭”, “匹” is the superordinate classifier of “頭”, rep- resented as “匹” ≻ “頭” in this paper. The taxon- omy represents such superordinate-subordinate rela- tions between classifiers in the form of a tree struc- ture. A taxonomy of classifiers would be fundamen- tal knowledge for natural language processing. In addition, it will be useful for language learners, be- cause learning usage of classifiers is rather difficult, especially for Western language speakers. We evaluate the proposed method by using the data of three Asian languages: Chinese, Japanese and Thai. 2 Noun-classifier agreement database First, let us introduce usages of classifiers in Asian languages. In the following examples, “CL” stands for classifier. • Chinese: yi-ju (CL) dian-hua (telephone) · · · a telephone • Japanese: inu (dog) 2 hiki (CL) · · · 2 dogs • Thai: nakrian (student) 3 khon (CL) · · · 3 students 397 As mentioned earlier, the agreement between nouns and classifiers is observed. For instance, the Japanese classifier “hiki” in the above example agrees with only animals. found in Chinese and Thai.The agreement is also The proposed method to …

Content truncated to 500 words