This paper investigates the sortal classifier system in Zauzou, a Tibeto-Burman language spoken in Southwest
China. Three etymological subclasses – repeater, non-repeater, and quasi-repeaters – are identified in two morphosyntactic
contexts: the bare classifier phrase [n+clf] and the full noun phrase [n+mod+clf], serving as an
“individualizer” that transforms unindividualized nominal concepts into referential individuals. The three types of classifiers
classify nouns on the basis of different semantic parameters that differ in terms of semantic contrastiveness. Animacy, shape,
size, rigidity and quanta are non-contrastive semantic parameters commonly found among non-repeaters. Repeaters and
quasi-repeaters categorize nouns primarily on the basis of semantically contrastive taxonomy, meronomy, or arrangement. With
respect to the function of quantification, repeaters, quasi-repeaters, and non-repeaters are identical in [n+mod+clf] but
diverge in [n+clf]. They do not uniformly render the singular interpretation with all kinds of referents in
[n+clf]. Non-repeaters are more grammaticalized classifiers than (quasi-)repeaters. The Zauzou data provide evidence
for the development from repeaters to non-repeaters, on a par with the commonly recognized grammaticalization path from nouns to
classifiers, which is conditioned by two types of “parameter shift”. This study offers a descriptive model that effectively
captures the relationship between semantic-functional properties of sortal classifiers and their historical development in
“repeater” languages.