A computational theory of child overextension

Overextension—the phenomenon that children extend known words to describe referents outside their vocabulary—is a hallmark of lexical innovation in early childhood. Overextension is subject extensive inquiry linguistics and developmental psychology, but there exists no coherent formal account this phenomenon. We develop general computational framework captures important properties overextension reported separately the previous literature. operationalize as probabilistic inference over conceptual space draws on fusion knowledge from semantics, deep neural networks, psychological experiments support both production comprehension. show how minimally parameterized explains young comprehensive set noun-referent pairs previously child speech, it also predicts behavioral asymmetry children's overextensional comprehension lab settings. Our work offers theory for origins word meaning extension supports single-system view language