On the Idiosyncrasies of the Mandarin Chinese Classifier System

Proceedings of NAACL-HLT 2019, pages 4100–4106 Minneapolis, Minnesota, June 2 - June 7, 2019. c⃝2019 Association for Computational Linguistics 4100 On the Idiosyncrasies of the Mandarin Chinese Classifier System Shijia Liu@ and Hongyuan Mei@ and Adina WilliamsP and Ryan Cotterell@, H @Department of Computer Science, Johns Hopkins University PFacebook Artificial Intelligence Research HDepartment of Computer Science and Technology, University of Cambridge {sliu126,hmei}@jhu.edu, adinawilliams@fb.com, rdc42@cam.ac.uk Abstract While idiosyncrasies of the Chinese classi- fier system have been a richly studied topic among linguists (Adams and Conklin, 1973; Erbaugh, 1986; Lakoff, 1986), not much work has been done to quantify them with statisti- cal methods. In this paper, we introduce an information-theoretic approach to measuring idiosyncrasy; we examine how much the un- certainty in Mandarin Chinese classifiers can be reduced by knowing semantic information about the nouns that the classifiers modify. Using the empirical distribution of classifiers from the parsed Chinese Gigaword corpus (Graff et al., 2005), we compute the mutual information (in bits) between the distribution over classifiers and distributions over other lin- guistic quantities. We investigate whether se- mantic classes of nouns and adjectives differ in how much they reduce uncertainty in classi- fier choice, and find that it is not fully idiosyn- cratic; while there are no obvious trends for the majority of semantic classes, shape nouns re- duce uncertainty in classifier choice the most. 1 Introduction Many of the world’s languages make use of nu- meral classifiers (Aikhenvald, 2000). While theo- retical debate still rages on the function of numeral classifiers (Krifka, 1995; Ahrens and Huang, 1996; Cheng et al., 1998; Chierchia, 1998; Li, 2000; Nis- bett, 2004; Bale and Coon, 2014), it is generally ac- cepted that they need to be present for nouns to be modified by numerals, quantifiers, demonstratives, or other qualifiers (Li and Thompson, 1981, 104). In Mandarin Chinese, for instance, the phrase one person translates as 一个人 (y¯ı g`e r´en); the classi- fier 个 (g`e) has no clear translation in English, yet, nevertheless, it is necessary to place it between the numeral 一 (y¯ı) and the word for person 人 (r´en). There are hundreds of numeral classifiers in the Mandarin lexicon (Po-Ching and Rimmington, Classifier P¯ıny¯ın Semantic Class 个 g`e objects, general-purpose 件 ji`an matters 头 t´ou domesticated animals 只 zh¯ı general animals 张 zh¯ang flat objects 条 ti´ao long, narrow objects 项 xi`ang items, projects 道 d`ao orders, roads, projections 匹 pˇı horses, cloth 顿 d`un meals Table 1: Examples of Mandarin classifiers. Classi- fiers’ simplified Mandarin Chinese characters (1st col- umn), p¯ıny¯ın pronunciations (2nd column), and com- monly modified noun types (3rd column) are listed. 2015, Table 1 gives some canonical examples), and classifier choices are often argued to be based on inherent, possibly universal, semantic properties associated with the noun, such as shape (Kuo and Sera, 2009; Zhang and Jiang, 2016). Indeed, in a summary article, Tai (1994) writes: “Chinese classifier systems are cognitively based, rather than arbitrary systems of classification.” If classifier choice were solely based on conceptual features of …

Content truncated to 500 words