We take it as axiomatic that children come to the language learning task with a rich set of linguistic universals. This innate endowment enables children to acquire the language they are exposed to on the basis of very limited data under highly diverse conditions. The past three decades of research have shown the early grammars of two- and three-year-olds to be exceedingly complex. Various syntactic categories (lexical and functional) are evidenced at around two years old or earlier, and by three years old, a wide range of noun phrases and verb phrases, as well as complex clauses, have appeared. The early grammars of children contain abundant evidence of parametric variation, reflecting the characteristics of their target language.

This project examines the grammar of young Cantonese-speaking children, with particular reference to typological characteristics of Cantonese, viz. classifiers and their role in the noun (or determiner) phrase, pro-drop, the interrogative and non-interrogative uses of wh- words, functional categories (including modal auxiliaries, aspect markers, and sentence final particles), focus adverbs, as well as the syntagmatic encoding of the definite/indefinite distinction. The data are based on the longitudinal corpus of 8 children aged between one and a half and three year olds, as well as experimental studies carried out with children up to five and a half years old.

The project has created a tagged computer corpus of the language of 8 children coded in an internationally accepted format, 14 megabytes in size, the largest child language corpus to date in Chinese-speaking communities. This corpus, which we call the Hong Kong Cantonese Child Language Corpus, will be deposited both at the child language archive of Carnegie Mellon University and at the Arts Faculty server of the Chinese University. It will serve as a useful research tool for anyone interested to investigate early language, be they linguists, psychologists, philosophers, sociologists, or educationists.

We have gained a precise understanding of certain aspects of early grammar in Cantonese. The categories of classifiers, aspect markers, modal auxiliaries and sentence final particles are evidenced between 1 year 9 months and 1 year 11 months old, confirming the early onset of functional categories. Classifiers are used productively before age two; a core set of 11 nominal classifiers go3, zek3, zi1, tiu4, zoeng1, lap1, gaan1, bun2, dou6, di1, deoi3, including both sortal and mensural classifiers, are used by all children. Verbal classifiers are rare at this stage. The earliest aspect markers are the perfective marker zo2, followed by the durative marker zyu6, and the progressive marker gan2. Both the postverbal deontic modal auxiliary dak1 and the preverbal modal auxiliaries wui3 and ho2ji3 are established at around two years old. Sentence final particles surface early as a prominent clausal constituent before two years old; at this stage they are found to be primarily non-interrogative in nature, and combinations of final particles are rare.

The articulation of the internal structure of the noun (or determiner) phrase follows a regular pattern. First bare nouns appear, then determiner-noun combinations, followed by numeral-classifier combinations. Classifier-noun combinations have a marked status. With respect to their external distribution, noun phrases of the afore- mentioned types invariably appear first in object position before they occupy subject positions. Pronouns are the exception, preferring subject to object positions.

Our study of pro-drop demonstrates convincingly that as in other languages, the rate of subject drop far exceeds that of object drop in child Cantonese. On the other hand, Cantonese-speaking children drop subjects at a much higher rate than their American counterparts (approx. 80% vs 30% at age two). Similarly, they drop objects much more than American children (approx. 47% vs 7% at age two). Thus, the early grammars of young children already reflect the typological features of their target language. The dropping of subjects and objects cannot be explained by processing limitations, since subjects of complex sentences are not necessarily dropped to a greater extent than those of simple sentences.

In a language in which wh-movement does not occur overtly, we have found asymmetries between subject and object, and between argument and adjunct in the acquisition of wh-words, echoing findings from other languages. Both the naturalistic and experimental data show that what-object (mat1je2) questions are mastered before what-subject questions, and who-subject (bin1go3) questions before who-object questions. Where-argument (bin1dou6) questions are acquired before where-adjunct questions. Verb type is found to be an important factor in interrogative acquisition. With respect to non-interrogative uses of wh-words, a phenomenon present to a much greater extent in Chinese than in languages such as English, children show such uses by three years of age. The earliest non-interrogative uses of wh-words (e.g. gei2 'how-(many/much)'; dim2 'how'; bin1 'where'; me1 'what') are restricted to intensification, rhetorical usage, and labeling of objects without names. The uses of wh-words in polarity contexts are not attested.

A salient feature of Cantonese is its ability to signal the definite/indefinite contrast exclusively in terms of preverbal vs postverbal word order. Our experimental data suggest that children do not develop a reliable grasp of this syntagmatic distinction until very late, around five and a half years old. The findings on existential and ergative verbs show unequivocally that the cue to the definiteness distinction lies in word order and not in the presence of the existential verb jau5 'be/have'.

The syntactic distinction between prepositions and verbs in Cantonese is not as clearcut as in languages such as English. Our study of words which can function both as verbs and prepositions in the language, such as bei2, indicates that the verb use appears earlier than the prepositional use.

In expressing quantificational meanings, Cantonese relies heavily on nonmovable adverbs located between the subject and the predicate. A prominent feature of child Cantonese as opposed to child English is that these quantifying adverbs occur early. We explored how young Cantonese-speaking children understand adverbs that signal the meanings of 'and', 'even', and reiteration, viz the adverbs zoi3, jau6, zung6, dou1. Focus structures are not found before two years old, but emerge between two and three years of age. Children early on are sensitive to quantificational properties of particular adverbs such as the direction of quantification. Pragmatic scales are not yet formed.