Biased statistical learning of closed-class items
Heidi Getz, Elissa Newport
November 2021

In natural languages, closed-class items predict open-class items but not the other way around. For example, in English, if there is a determiner there will be a noun, but nouns can occur with or without determiners. Here we asked whether statistical learning of closed-class items is also asymmetrical. In three experiments we exposed adults to a miniature language with the one-way dependency “if X then Y”: if X was present, Y was also present, but Y could occur without X. We created different versions of the language in order to ask whether learning depended on which category (X or Y) was an open or closed class. In one condition, X had the main properties of a closed class and Y had the main properties of an open class; in a contrasting condition, X had properties of an open class and Y had properties of a closed class. Learners’ exposure in these two conditions was otherwise identical. Learning was significantly better with closed-class X. Additional experiments demonstrated that it is the perceptual distinctiveness of closed-class items that drives learners to analyze them differently, and that the mathematical relationship between closed- and open-class items influences learning more strongly than their linear order. These results suggest that statistical learning is biased: learners privilege computations in which closed-class items are predictive of, rather than predicted by, open-class items. We suggest that the distributional asymmetries of closed-class items in natural languages—and perhaps the asymmetrical structure of linguistic representations—may arise in part from this learning bias.
Format: [ pdf ]
Reference: lingbuzz/006325
(please use that when you cite this article)
Published in:
keywords: language acquisition, statistical learning, functional heads, learning biases, universal grammar, morphology, syntax
previous versions: v1 [November 2021]
Downloaded:89 times


[ edit this article | back to article list ]