On the general classifiers ge and zàg in Hakka

This paper investigates the distribution and properties of the Hakka general classifiers ge and zàg. We focus on the [determiner/numeral + classifier + noun] construction where we observe the relations between the general classifiers and their following nouns, chosen based on their frequency in this construction. We adopt a corpus-based collostructional analysis which calculates the collocational strength values of ge and zàg with following nouns. A Hakka corpus was compiled for the study. The three-way distinction in the collostructional analysis (attractive, neutral, and repulsive) is directly mapped to acceptability of various degrees. The results show that ge is highly correlated with human-denoting nouns, whereas zàg is highly correlated with animal-denoting nouns. Nouns denoting abstract entities or concrete objects without physical properties like size or shape usually lack specific classifiers, and both ge and zàg can collocate with them, albeit with varying degrees of preference. We argue that both ge and zàg are general classifiers because both are more frequently used than specific classifiers and both exhibit disjointed semantic distribution and allow abstract nouns. While they show preferences for different nouns, requirements to qualify as general classifiers are equally met.