How do Mandarin-speaking children learn shape classifiers?

Mandarin is a classifier language. A classifier is inserted between a number and a noun for the purpose of quantification (e.g., 一条绳子 one tiáo rope). Each classifier marks semantic characteristics of the noun with which it co-occurs (e.g., 条 tiáo is typically paired with long, narrow and flexible objects). The semantic system of classifiers is complex, and classifier production is a vulnerable area for Mandarin-speaking children (e.g., Hao et al., 2018). However, it is unclear what learning mechanisms drive the acquisition of classifiers in Mandarin-speaking children. In the present study, we explored potential predictors, namely classifier-based semantic categorization and input frequency of classifiers. In addition, we hypothesized that existing vocabulary knowledge would be related to classifier learning. Sixty-four typically-developing monolingual Mandarin-speaking children between 4;1 (year;month) and 6;5 completed two background tasks and two experimental tasks. The background tasks consisted of an object categorization task to index semantic categorization strategy, and a picture selection task and a picture naming task to measure vocabulary knowledge. In experiment 1, we implemented a comprehension and a production task for six real classifiers that emphasize shape. In experiment 2, we administered a learning task for two novel classifiers which encoded different semantic properties (i.e., curly-haired vs. broken). Frequency of classifier input was manipulated using a between-subject design. We analyzed contributions of classifier-based semantic categorization, input frequency of classifiers, and vocabulary knowledge to classifier comprehension and production. Even though children preferred to categorize objects by shared classifier than by other semantic links in the object categorization task, this preference did not significantly predict real classifier comprehension and production in experiment 1. At the same time, vocabulary knowledge was a significant predictor for both. Children may find that the semantic system of real Mandarin classifiers is opaque, and they rely more heavily on an item-by-item learning approach that is used in vocabulary learning (i.e., idiosyncratic learning of individual words). In addition, children showed varied accuracy on different classifier-noun pairs for the same classifier, providing more evidence for item-based learning. For novel classifier learning in experiment 2, classifier-based object categorization was a marginally significant predictor for comprehension. The higher frequency group did not outperform the lower frequency group, and vocabulary knowledge was a significant predictor for neither comprehension nor production. These findings suggest that children mainly took a rule-based approach to learn novel classifiers with transparent semantic categorization. Overall, results from the two experiments showed that the learning approach children primarily use to learn classifiers depend on the transparency of the classifier system.