Learning Count Classifier Preferences of Malay Nouns

Learning Count Classifier Preferences of Malay Nouns Jeremy Nicholson and Timothy Baldwin NICTA Victoria Research Laboratories University of Melbourne, VIC 3010, Australia {jeremymn,tim}@csse.unimelb.edu.au Abstract We develop a data set of Malay lexemes la- belled with count classifiers, that are attested in raw or lemmatised corpora. A maximum entropy classifier based on simple, language- inspecific features generated from context to- kens achieves about 50% F-score, or about 65% precision when a suite of binary classi- fiers is built to aid multi-class prediction of headword nouns. Surprisingly, numeric fea- tures are not observed to aid classification. This system represents a useful step for semi- supervised lexicography across a range of lan- guages. 1 Introduction This work examines deep lexical acquisition (DLA) of count classifiers in Malay.1 DLA is the process of (semi-)automatically learning linguistic structures for use in linguistically-rich language resources such as precision grammars or wordnets (Baldwin, 2007). Malay is a significant target for DLA in that rela- tively few NLP resources have been developed for it.2 In many languages — notably many South-East and East Asian languages — numbers cannot gener- ally modify nouns. Instead, they modify count clas- sifiers3 which tend to occur as a genitive modifier 1The language is called by a number of names, including bahasa Malayu, bahasa Malaysia, the Malaysian language, and Standard Malay. We simply use Malay, to agree with the Eth- nologue (Gordon, 2005). 2Although some research has been done on bahasa Indone- sia (henceforth Indonesian), another Malayan language with which Malay is (somewhat) mutually intelligible. 3This paper focuses specifically on sortal classifiers. See or in apposition to the noun. In (1) below, biji is an example of a Malay count classifier (CL), premod- ified by the numeral tiga “three”; similarly for the Japanese in (2):4 (1) tiga three biji CL pisang banana “three bananas” (2) saN three boN CL no GEN banana banana “three bananas” The closest English analogue to count classifiers is measure nouns (Flickinger and Bond, 2003), like loaf in 3 loaves of bread (compare 3 breads, which has markedly different semantics). Measure nouns cannot be used to count countable nouns in English, however (e.g. ∗3 loaves of roll). Syntactically, count classifiers are nouns which occur with a (usually numeric) specifier, and are used to quantify uncountable nouns. In languages such as Malay, most nouns are uncountable and numeral–classifier combinations must be used to enumerate instances of a given noun (c.f. ∗tiga pisang; this is not true of the nouns that are clas- sifiers themselves, like orang “person” or buah “fruit”, e.g. tiga orang “three people”). Semanti- cally, count classifiers select for objects of particular semantics, commonly determined by a conceptual class (e.g. HUMAN or BIRD) or the relative dimen- sions of the object (e.g. LONG-AND-THIN or FLAT). Paik and Bond (2001) for discussion of other types of count classifiers, namely event, mensural, group and taxonomic. 4In the Japanese example, the genitive (GEN) linker is re- quired to connect the numeral–classifier combination with the noun in a single noun phrase. In …

Content truncated to 500 words