The Acquisition of the Default Classifier in Taiwanese *

For many years, especially since the publication of Rumelhart and McClelland's (1986) connectionist model of English past tense acquisition, a debate has raged over the reality of an innate rule-learning mechanism in language development (for recent summaries of arguments, see Pinker 1999 for the pro side and Bates and Goodman 1997 for the con). For a debate that is so obviously important for our understanding of the nature of the human language faculty, it is surprising that the linguistic territory that it has been fought over is so small. Following the lead of Rumelhart and McClelland (1986), the vast majority of studies have focused on the difference between irregular and regular verbs in English (e.g. Marcus et al. 1992 vs. Marchman and Bates 1994). Some work has also been done on English noun inflection (e.g. Marcus 1995a vs. Marchman, Plunkett and Goodman 1997), on inflection in languages typologically and/or genetically related to English, such as German (e.g. Marcus et al. 1995 vs. Köpcke 1998), and on inflection in other languages, such as Hebrew (discussed in Pinker 1999). Not all languages inflect, however, most famously Chinese (or more properly, languages in the Sinitic family, including Mandarin, Cantonese, and so forth). Given the narrow way the debate is often framed, it may seem that such noninflecting languages have nothing to tell us about the existence (or lack thereof) of rules in language development. On the contrary, we believe that they have the potential to move the debate in a more fruitful direction. Precisely because Sinitic languages don't have inflection (at least no irregular verbs or nouns), one is forced to rethink the fundamental question: How might a rule-learning mechanism be manifested in a noninflecting language? More specifically, is there any system in Sinitic in which there is a dichotomy between regular (rule-governed) and irregular (idiosyncratic) patterns? We believe that a very good candidate is provided by the nominal classifier system. Like many languages across the world (Allan 1977, Aikhenvald 2000), Sinitic languages require the use of certain morphemes ('classifiers') in certain syntactic contexts (namely after numerals and determiners within an NP). Usually the selection of a classifier depends on the semantics of the noun (shape, animacy, and so on), but there is an exception: a so-called 'general' or 'default' classifier. This acts like a 'miscellaneous' file, being selected for a variety of noun types, often ones that sometimes also go with semantically more specific classifiers. One could argue, then, that while speakers usually choose classifiers through some process involving the semantics of the noun (or perhaps sometimes they simply choose them by rote), there is also a 'default classifier rule' to fall back on in order to fulfill the syntactic obligations when memory (e.g. for which semantic features are encoded by which classifier) fails for some reason or other. If such a default classifier rule actually exists for adults, the natural conclusion is that in order to master the language, children have to learn this rule, and there thus must be a rulelearning mechanism. If the rule-learning mechanism in Sinitic languages then turns out to