In Mandarin, a classifier must be placed between a numeral and a noun, and classifiers encode semantic features. A noun category may take several classifiers, yet speakers typically use one classifier most frequently, which we call the category's conventional classifier. Prototypical members have features that match the conventional classifier's meaning, whereas less typical members may mismatch its meaning. We ask whether Mandarin-speaking children select conventional classifiers for these less typical members, and how input frequency and consistency of each classifier - noun pair interact with typicality to shape classifier selection. Eighty-six 5-7-year-old children produced numeral - classifier - noun phrases for 24 images (8 noun categories & times; 3 typicality conditions: prototypical, less typical, cross-category lookalike items). Conventional classifier use was highest for prototypical items, reduced for less typical items, and near floor for cross-category lookalikes. Frequency increased conventional classifier use for prototypical items, whereas consistency predicted conventional use for both prototypical and less typical items; when modeled together, consistency was a stronger predictor. Children's non-conventional responses to less typical items were overwhelmingly the Mandarin general classifier (g & egrave;). Together, the findings support the conclusion that children integrate both item typicality and pairing statistics while balancing convention-following and semantic-adjustment strategies in classifier selection.
