The universality of simple distributional methods: Identifying syntactic categories in Mandarin Chinese. Martin Redington, Nick Chater, Chu-Ren Huang, Li-Ping Chang, Steve Finch, & Keh-jiann Chen * Introduction The problem of acquiring language is a difficult and complex one. The traditional as- sumption is that infants acquire language by virtue of innate knowledge, which reduces the problem to one of tuning this innate knowledge of language in general, to the specific characteristics of the language spoken in the infant’s early environment (e.g. Chomsky, 1980). Nevertheless, without empirical assessment of potential learning mechanisms and sources of information, it may be premature to decide which aspects of language, if any, require drawing on innate knowledge. This paper considers a simple distributional learning mechanism, which does not draw on explicit prior knowledge. This method has previously been shown to be informative about the syntactic category membership of individual words in English, French, and German. We ask whether it can provide similar constraints in Chinese. Acquiring Syntactic Categories In acquiring syntactic categories (such as noun, verb, etc.), language learners face two inter-related problems. They must discover the set of syntactic categories in the language, and also identify the syntactic category membership of individual words. It seems both plausible and likely that human infants possess some innate constraints on the identity of the syntactic categories 1 and on the form of the rules governing how these categories can be combined into sentences (the Universal Grammar). However, the extent to which these constraints can aid the identification of the syn- tactic categories of individual words (which cannot be known innately) is unclear. It may well be that semantic or pragmatic information (for instance, knowing that ball refers to a round object [see Pinker, 1984], or that want ball is likely to result in an adult pro- viding a round object [see Snow, 1988]) can provide both clues that ball is a noun, and * Redington & Chater: Department of Experimental Psychology, University of Oxford, South Parks Rd., Oxford, UK. OX1 3UD. E-mail: [fmr|nick]@psy.ox.ac.uk . Huang: Institute of His- tory and Philology, Academia Sinica, Taipei, Taiwan. E-mail: hschuren@ccvax.sinica.edu.tw . Chang & Chen: Institute of Information Science, Academia Sinica, Taipei, Taiwan. E-mail: [lpchang|kchen]@iis.sinica.edu.tw . Finch: Human Communications Research Centre, 1–4 Buc- cleuch Place, Edinburgh, UK. EH8 9LW. E-mail: steve@cogsci.ed.ac.uk . 1 For example, it has been proposed that infants might possess innate ‘action’ and ‘object’ categories, corresponding to nouns and verbs, (Pinker, 1984).